IOCCG Report (Draft)

Earth Observations in Support of Global Water Quality Monitoring Report of a GEO Water Quality Community of Practice, chaired by Steven

Greb, and based on contributions from (in alphabetical order): Stewart Bernard

Caren Binding

Carsten Brockmann

Paul Digiacomo

Derek Griffiths

Steve Groom

Tiit Kutser

Veronica Lance

Mark Matthews

Chris Mannaerts

Daniel Odermatt

Lisl Robertson

Blake A. Schaeffer

Rick Stumpf

Andrew Tyler

Erin Urquhart

1

Water Quality Monitoring Report PrefaceChapter 1: Water Quality Monitoring and Assessment: Needs, benefits and frameworks

1.1 Introduction1.2 The need for protecting water’s designated uses1.3 Benefits of maintaining water designated uses

1.3.1 Environmental Stewardship1.3.2 Human Health and Well-being1.3.3 Economic Growth

1.4 Frameworks for monitoring and protecting designated uses1.4.1 Global challenges related to water designated uses1.4.1 Africa1.4.2 Asia1.4.3 Australia1.4.4 Europe1.4.5 North America1.4.6 South America

1.5 How satellite remote sensing can contribute1.6 Challenges and Why Now?References

Chapter 2: Introduction to deriving water quality measures from satellites2.1 Introduction 2.2 Theoretical Basis of Water Quality Remote Sensing 2.3 What Can be Measured Using Remote Sensing? 2.4 Water Quality Retrieval Algorithms 2.5 Platform, Resolution, and Data Processing Considerations 2.6 Advantages and Limitations of Remote Sensing for Water Quality Monitoring 2.7 Case Studies

2.7.1 Operational Satellite Water Quality Monitoring 2.7.2 Eutrophication and harmful algal blooms

ReferencesChapter 3: Complementarity of in-situ and satellite measurements

3.1 Water quality monitoring approaches3.2 Discrete in situ measurements and water sample analyses

3.2.1 Prevalent techniques and suitability3.2.2 Examples

3.3 Continuous in situ measurements by automated moorings, drifting buoys and AUVs

3.3.1 Prevalent techniques and suitability3.3.2 Examples

3.4 Remotely sensed water quality products3.4.1 Prevalent techniques and suitability3.4.2 Examples

2

3.5 Discussion of measurement statistics and related attributes for all three approaches

3.5.1 Accuracy 3.5.2 Representativeness considerations3.5.3 Complementary role of remote sensing with in situ measurements3.5.4 Complementary role of remote sensing with modelling3.5.5 Costs

3.6 Resources3.6.1 In situ data bases3.6.2 Integrated data systems

ReferencesChapter 4: Linkages Between Data Providers and End Users

4.1 Context4.2 Data delivery

4.2.1 Data formats4.2.2 Data extraction and analysis tools.

4.3 Training4.4 Quality Control

4.5 Validation Data4.6 Timeliness

4.6.1 Real-time Monitoring and Response.4.6.2 Current Assessment4.6.3 Retrospective Assessment

4.7 Future Collaborative EffortsReferences

Chapter 5: Measures from satellites5.1 Understanding the satellite signal measured from inland and coastal waters

5.1.1 The water-leaving signal5.1.2 The atmospheric signal

5.2 Deriving water quality products from satellite: algorithms and issues5.2.1 Algorithm types: partial atmospheric corrections5.2.2 Algorithm types: full atmospheric correction

5.3 OutlookReferences

Chapter 6: Sensors6.1 Introduction

Spaceborne sensor requirements and technical constraints across multispectral and hyperspectral configurations and spatial and temporal resolution

6.1 Spectral band requirements for estimating water variables6.2 Spectral bands for atmospheric correction6.3 Spatial Resolution Requirements6.4 Radiometric accuracy requirements

6.4.1 Signal-to-Noise Ratio or radiometric sensitivity requirements

3

Trade-offs and solutions (blending) between spectral, spatial, radiometric and temporal requirements

6.5 The application potential of relevant existing and planned spaceborne sensors (where the specifications are already set)6.6 Recommendations on desired specifications on future sensors.References

Chapter 7: Coastal and Inland Water Quality Observing System Assets and Supporting Infrastructure

7.1 Introduction 7.2 Observing System Assets: Current Status and Future Directions

7.2.1 Satellite assets7.2.2 In situ assets 7.2.3 Modeling and data assimilation capabilities7.2.4 Needs, issues and gaps 7.2.5 Future directions, solutions and priorities for ocean observing assets

7.3 Constellations and Integration of Observing Assets 7.3.1 Current integration approach across observing system assets7.3.2 Future constellation and integration approach

7.4 Supporting observing system infrastructure 7.4.1 Calibration of satellite observations7.4.2 Validation of satellite data products7.4.3 Quality monitoring and assessment of data and product utility7.4.4 Capacity building for developing and developed nations

7.5 Recommendations to agencies and programmes 7.5.1 CEOS7.5.2 Global Observing Systems: GOOS, GCOS and GTOS7.5.3 GODAE OceanView7.5.4 GEOSS: Blue Planet and Water Cycle Initiatives

7.6 Summary and conclusions Chapter 8: Ground and User Segment

8.1 Introduction8.2 Mission or satellite ground segment

8.2.1 Overview8.2.2 Mission ground segments

8.2.2.1 Nasa/gsfc (oceancolor)8.2.2.2 ESA/Sentinel (2,3) collaborative GS ;8.2.2.3 NOAA STAR and CoastWatch

8.2.3 Data distribution mechanisms8.2.3.1 FTP, tape and web downloads8.2.3.2 GEONETCast near real time dissemination system (GEO)8.2.3.3 Web-based analysis systems

8.3 User Ground Segments8.4 SummaryReferences

4

Chapter 9: New Concepts (Virtual Laboratory)9.1 Introduction9.2 “Big data” from Space

9.2.1 Historic archives and future data volumes 9.1.2 Challenges for data access and data processing

9.3 Moving software to the data (Carsten)9.3.1 Data harmonization (“Data Cube”)9.3.2 Exploratory tools (e.g. Product Feature Extraction[1],[2]) 9.3.3 Service models

9.3.3.1 Infrastructure as a Service (IaaS)9.3.3.2 Software as a Service (SaaS)

9.3.4 Community tools 9.3.5 Mobile Technology

9.4 Examples9.4.1 The Australian Geoscience Data Cube 9.4.2. The ESA Thematic Exploitation Platform 9.4.3 Climate, Environment and Monitoring from Space (CEMS) 9.4.4 Something from US Paul

9.5 Summary and Conclusions: Benefits and drawback from inland water points of viewReferences

Chapter 10: Data Exploitation Programmes and User Support10.1 Introduction10.2 User support programs10.3 Thematic workshops10.4 Dedicated R&D studies10.5 Validation10.6 Summary of recommendations from inland water point of view

5

Water Quality Monitoring Report PrefaceGoals of the IOCCG

The International Ocean-Color Coordinating Group (IOCCG) was established nearly 20 years ago to global scale develop consensus and synthesis of satellite ocean color radiometry (OCR). Composed of experts from data providers (Space Agencies) and scientific and management user communities, the IOCCG’s primary objectives are to:

● Foster expertise and broaden the user community of ocean-color data ● Develop consensus among the OCR user community and provide a common voice to

international bodies and space agencies ● Promote the importance of OCR data to the global community ● Encourage and facilitate standard international calibration and validation ● Advocate collection of essential ocean and atmosphere data ● Encourage and facilitate common access and exchange of ocean color data

The IOCCG Earth Observations in Support of Global Water Quality Monitoring Working Group

Declining inland and estuarine water quality has become a global issue of significant concern as anthropogenic activities expand and climate change threatens to cause major alterations to the hydrological cycle (UNEP, 2002; ICPP, 2008, GOOS, 2012). The measurement of water quality variables via radiometric measurements of the water’s optical properties has rapidly grown over recent years. Improvements in algorithms and product development, sensor technology and maturity, and data accessibility and provision have led to demonstrated confidence in remotely sensed data with potential applications to water resources management. However, to date, management agencies have been slow to embrace satellite derived measurements even though important parameters such as chlorophyll, cyanophycocyanin, suspended solids, colored dissolved organic matter, and light attenuation, Secchi Disk transparency and turbidity have been quantified with required accuracies.

The IOCCG formed a working group on Earth Observations in Support of Global Water Quality Monitoring in 2014 to support the implementation of a global water quality monitoring service that contributes to the broader implementation of the Global Earth Observation System of Systems (GEOSS) under the auspices of the Group on Earth Observations (GEO). The goal of the working group is to provide a strategic plan that incorporates current and future earth observation (EO) information into national and international near-coastal and inland water quality monitoring efforts. The working group aims to accomplish this by promoting best practices, coordination of efforts and partnerships, and proposinge specific new linkages

6

between data providers and data end users. The members of the working group bring a diverse range of backgrounds and perspectives to this task. The membership includes individuals from space agencies, local and national management agencies, and the scientific community <?appendix one-table of members/contributors?>. Purpose of this report The purpose of this report is to provide the overview, background and detailed information needed to support the development of an earth observation based global water quality monitoring service. The objectives of this report are to assess current knowledge and gaps regarding coastal and inland water quality and associated use of remote sensing data and assess existing space-based and in situ observing capabilities. This report also identifies user needs and requirements, new observing capabilities and new and improved data streams and products, mission requirements, supporting research and development activities, and best practices. Finally, this report explicitly discusses user engagement and outreach with the objective to strengthen linkages between data providers and end users. Who should read this report? This report is intended to reach a broad audience, but particularly targets three primary audience constituencies: 1) water resources management end users, 2) science community, and 3) OCR data providers . The goal of targeting these audiences is to build a community of practice and strengthen linkages between the groups for broader “downstream” end-user utilization of ocean color data. How to read this report The report has three sections, and each section is intended for one of the three audiences identified above.

● Section A Earth observations in support of water quality monitoring and assessment for stakeholders This section is written for the end user and will provide background on water quality monitoring frameworks and tools available to the end user. It includes discussion of the tools accuracies, uncertainties and representativeness. This section will also describe the fundamentals of remote sensing, what can and can’t be measured with remote sensing, the complementary role of remote sensing and in situ measurements and where to find information and decision support tools.

● Section B Remote sensing science in support of water quality monitoring and

assessment for stakeholders

7

The second section is directed towards the science community and will focus on our current understanding of the water and atmospheric satellite signal over inland and coastal waters, current methodologies for deriving optical properties and derived water quality products and current and future sensor requirements, constraints and applications.

● Section C Perspectives for agencies in support of water quality monitoring and

assessment for stakeholders The third section is directed towards agencies with responsibilities associated with space or ground segments. This section also explores new concepts in data processing and applications in Chapters 9 and 10, and will be of interest to all of the audience groups.

The three sections are also organized from a timeframe perspective. Section A examines the immediate needs (<1 year) of the end-user; section B addresses the research and development questions with a 1-5 year timeframe; and section C has a long-term perspective with recommendations for future programs. <What we hope readers will learn from the report: Major takeaways and conclusions> I like this but would like to defer until after chapter rough drafts are composed.

Linkages with Previous IOCCG Reports

<need to briefly discuss how this links with each of these>● Report 3 (2000): Remote Sensing of Ocean Colour in Coastal, and Other Optically-

Complex, Waters ● Report 5 (2006): Remote Sensing of Inherent Optical Properties: Fundamentals, Tests of

Algorithms, and Applications ● Report 7 (2008): Why Ocean Colour? The Societal Benefits of Ocean- Colour Tech● Report 8 (2009): Remote Sensing in Fisheries and Aquaculture ● Report 10 (2010): Atmospheric Correction for Remotely-Sensed Ocean-Colour Products● Report 13 (2012) Mission Requirements for Future Ocean-Colour Sensors.

Acknowledgements

<need to complete>

8

Chapter 1: Water Quality Monitoring and Assessment: Needs, benefits and frameworksBlake A. Schaeffer, Steve Greb, Caren Binding, Erin Urquhart

1.1 IntroductionThe information throughout this report is provided for water quality managers and various stakeholders interested in monitoring and assessing coastal and inland waters with satellite remote sensing. Water quality managers are identified as anyone who is responsible for protecting the designated, or beneficial, uses of water for any purpose. Water quality is defined as the biological, physical, and chemical characteristics required to maintain these uses. Broad categories of designated uses may include drinking, recreation, irrigation, and food supply. Stakeholders are more broadly defined as any person or entity that may have interest in, interact with, or benefit from a waterbody’s uses. An end-user of the satellite remote sensing information may include both water quality managers and stakeholders.

It is important to have a robust scientific understanding of environmental processes to inform stakeholders when addressing biological, physical and chemical water quality dynamics that impact designated uses. Management of water quality generally focuses within watersheds and the connected inland and coastal waters. Here, we focus on surface waters in wide rivers, lakes, reservoirs, estuaries and coastal waters that can be resolved by current satellite remote sensing technology (Figure 1).

9

Figure 1. Global map of population density projected for 2015 (CIESIN 2005) with world-wide lakes, reservoirs, rivers and permanent open water bodies with a surface area ≥ 0.1 km2 (black points) from the World Wildlife Foundation.

As there are numerous definitions of coastal waters that depend on boundaries delineated scientifically, politically, or through other mechanisms, a single definition of coastal waters is not practical (IGOS 2006). This report considers coastal waters to cover the interface between land and sea toward the furthest ocean extent necessary, dependent on the stakeholder requirements for specific local, regional, or national interests. Historically, ocean color satellite sensors focused on the global oceans (McClain 2009) and these same sensors are repurposed for coastal and inland water quality monitoring (Mouw et al. 2015). Therefore, there is no limit as to how far these capabilities can be applied into the ocean beyond any single definition of coastal waters.

1.2 The need for protecting water’s designated usesManaging the Ddemand for finite water resources to support societal uses requires information on the quality of the water, so stakeholders can make sustainable decisions. Population growth, climate change and variability (Vorosmarty et al. 2000), and changing land use practices (Meyer and Turner 1992) all contribute to the stress of national water quality. Water quality can be measured in terms of biological, physical, and chemical indicators such as turbidity, chlorophyll-a, harmful algae, pollution-sediment, submerged habitat, temperature, metals, dissolved oxygen, nutrients (primarily phosphorus and nitrogen) and many other contaminants. Here, we focus on core water quality indicators that can be derived directly from satellite remote sensing including turbidity, chlorophyll-a, harmful algae, pollution-sediment, submerged habitat and temperature (Muller-Karger 1992). These six core indicators enable water quality managers and stakeholders to link anthropogenic stressors to water environmental responses that may impact designated uses.

1.3 Benefits of maintaining water designated usesThe world population was estimated at over seven billion (7 × 109) people in 2016 (U.S. Census Bureau 2016). An estimated 30% to 70% of this population lives within 100 km of a marine coastline, and 90% lives within 10 km of a freshwater body (Kummu et al. 2011; UNEP 2007; Wilson and Fischetti 2010). Sustainable water management practices are critical for ensuring overall environmental stewardship, human health and well-being, and continued economic growth.

1.3.1 Environmental StewardshipSince the last century, water use has exceeded long-term supply and the status of these water resources are inadequately monitored to inform stakeholders of change in quality (MEA 2005b; UNESCO 2006). The decline in the quality of water resources is causing extinction of freshwater species and putting many ecosystems at risk, including a loss of biodiversity (MEA 2005b). Coastal zones, the most productive ecosystems on Earth, are particularly vulnerable,

10

and linked to human and animal life, as well as entire ecosystem health (Barbier et al. 2011; Day et al. 2012; UNEP 2012). In recent decades, increasing pollution from inland, along with loss of coastal habitats that filter pollution, has led to extensive “dead zones” – or areas with low amounts of oxygen, where aquatic animals are unable to survive, such as in the Black Sea (Sorokin 2002), Baltic Sea (Conley et al. 2009), and the Gulf of Mexico (Diaz and Rosenberg 2008) (Figure 2). Ecosystem function is known to be of major importance to the well-being of humans and the conservation of biological diversity is one key element in maintaining ecosystems in good condition (MEA 2005a).

Figure 2. Global estimates for anthropogenic biomes (Ellis et al. 2013.) with distribution of 476 hypoxic zones (black points), as of 2011, from the World Resources Institute (Diaz et al. 2011).

1.3.2 Human Health and Well-beingCoastal area recreation benefits human well-being and quality of life due to increased contact with the natural environment (Cox et al. 2006). Furthermore, all socioeconomic levels associated with human health increase with living proximity to the coast (Wheeler et al. 2012). And yet bBy far the largest causes of water quality degradation and subsequent decline of aquatic systems originate fromis a primary result of anthropogenic activities, which threatensimpacting both human and ecosystem health. More than 80% of the global health burden is water related, and at any given time, people suffering from water illnesses occupy more than half of the world’s hospital beds (REF?). The lack of access to clean water and adequate sanitation remains the world’s largest health problem, resulting in the estimated death of an estimated 3 million people per year, the majority of which are children under the age of five (DFID et al. 2002; WHO 2006).

11

1.3.3 Economic GrowthWater quality impacts economies through, for example, , such as changes in property valuation and visitor decisions, which in turn impact tax revenues (Dodds et al. 2009). For example, eEutrophication in lakes has historically been identified as a cause for appraisal declines in residential property, especially in proximity to the degraded waterbody (Gibbs et al. 2002; Michael et al. 1996). Additionally, visibility of coastal waters has provided a premium for homes with waterfront views (Major and Lusht 2004). In freshwater systems, total eutrophication-related losses in the U.S. are estimated at $4.6 billion annually (Dodds et al. 2009), and in the UK $105-160 million annually (MEA 2005c), in South Africa $250 million (Frost and Sullivan International 2010)and Australia A$200 million (Atech 2000). Cyanobacterial blooms in Lake Tai during 1998 resulted in economic losses of $6.5 billion (Le et al. 2010).; annual costs of freshwater algal blooms in Australia were A$200 million in 2000 (Atech 2000), with similar annual eutrophication costs in the United Kingdom at $150 million (Pretty et al. 2003), and in South Africa at $250 million (Frost and Sullivan International 2010). Global economic losses due to pollution, eutrophication, and declining water quality are estimated at $6-16 billion (MEA 2005c; OECD 2012). Combined global ecosystem benefit estimates were $18x1012/yr for coastal waters and $2.3x1012/yr for lakes and rivers (Costanza et al. 2014). Therefore, the value of satellite data to inform decisions for water quality management could be significant based on a value of information calculation (Macauley 2006).

1.4 Frameworks for monitoring and protecting designated usesFirst, a global overview from the perspective of the United Nations (UN), World Health Organization (WHO), and World Bank are provided to describe the context of water designated uses and their related challenges and water designated uses that cross all continents. Then, select examples of water quality frameworks are presented to demonstrate the common goals for monitoring and protecting water uses within each continent. These frameworks define needs and requirements, some of which could be addressed with satellite remote sensing of water quality to assist in monitoring and maintaining the designated uses of water.

1.4.1 Global challenges related to water designated usesThe WHO’sWorld Health Organization (WHO) primary water focus is the prevention of waterborne and water-related diseases under the World Health Assembly resolution 64.24 of May 2011. There are five objectives through 2020: (1) collect information on water quality and health, (2) provide water quality management guidelines, (3) strengthen capacity to manage water quality, (4) facilitate implementation, and (5) monitor the impact on policies and practice. The Water Quality and Health Strategy 2013-2020 will support public health, socioeconomic development and well-being. This WHOWorld Health Organization strategy also supports the United Nations Millennium Development Goals which supportsed human rights to sustainable water access (WHO 2013), and the new Sustainable Development Goals.

The UNUnited Nations hasd eight Millennium Development Goals that are all connected to water quality management. Millennium Development Goals include eradicatinge extreme

12

hunger and poverty, universal primary education, gender equality, reducinge child mortality, improvinge maternal health, combating the human immunodeficiency virus and other diseases, ensuringe environmental sustainability, and developing a global partnership for development. The Goal of ensuring environmental sustainability was directly dependent on safe and clean water for growth, social and economic development, and reduction in poverty (Carr and Neary 2008). Global assessments on the status and trends of freshwater resources is supported by the UN’sUnited Nations Global Environment Monitoring System Water Programme. At the time this report was being developed, New Sustainable Development Goals (SDGs) were being formulated that include Clean Water and Sanitation and SustainableSustainabile Use of Oceans at the time this report was being developed.

The World Bank views water management as essential for continued human development, food and energy security, and job creation (Radstake and Tuinhof 2003). Water management is an important aspect for human lives, and continued economic and social growth. Maintaining beneficial water uses for growth and development support agriculture, energy production, and overall reduction in poverty. Water demand has increased and is expected to continue increasing in the future, which may threaten human health, agriculture, industry, and biodiversity. The World Bank has historically focused on sanitation, treatment, and nutrient pollution to protect water resources and mitigate the economic consequences.

1.4.1 AfricaThe National Water Act 36 of 1998 established the national framework for South Africa water management, protection, and use to benefit the public interest. The National Water Act required the development of the National Water Resource Strategy, with the objective to manage water resources for sustainable social and economic development. This strategy must provide information on water requirements and propose management actions. The National Environmental Management Act 107 of 1998 states everyone has the right to an environment that is not harmful, while supporting sustainable development to serve current and future generations. The Integrated Coastal Management Act 24 of 2008 ensures that the coastal zone is socially, economically, and ecologically sustainable; and to protect against adverse effects on the coastal environment. The National Eutrophication Monitoring Programme and the National Aquatic Ecosystem Health Monitoring Programme are aimed at maintaining the quality of surface water, such as reporting trophic status and related problems such as harmful algal blooms.

1.4.2 AsiaThe Water Environment Partnership in Asia includes 13 countries and looks to achieve sustainable socio-economic development and manage water resources. In Japan, the Water Pollution Control Law preserves public water areas to protect human health and the living environment, and is managed by the Ministry of Environment. The Mekong River Commission is comprised of China, Burma, Laos, Thailand, Cambodia, and Vietnam to support sustainable management and development of water and related resources to balance economic development, environmental protection and social sustainability. The Environment Programme

13

conducts monitoring activities to report water quality results to the involved nations, and the objective is to provide up to date environmental and social knowledge and efficient environmental management cooperation mechanisms for basin management and development (MRC 2010).

1.4.3 AustraliaThe National Water Quality Management Strategy is the primary water quality management policy in Australia, overseen by the Natural Resource Management Ministerial Council and the Environment Protection and Heritage Council, andthen implemented by state and territory governments. The objective is to protect and enhance water quality while maintaining economic and social growth. The Bureau of Meteorology provides regular reports on the status of water resources and use, which may be used for reporting on inland water quality and condition under the Environment Protection and Biodiversity Conservation Act of 1999. This Act requires a report to Parliament every five years on Australia’s biophysical, ecological, social, and culturally related environmental issues.

1.4.4 EuropeThe European Union (EU) Water Framework Directive provides protection of aquatic ecology, drinking water resources, and bathing water across EU member states. This directive is also complimented by a number of other directives such as the Marine Strategy Framework Directive 2008. The objective of the Water Framework Directive is to protect human health, water supply, natural ecosystems, and biodiversity by maintaining good ecological and chemical status of all ground and surface waters. Ecological status indicators are measured by the departure from the biological community that would be expected in conditions under no anthropogenic influence. The ecological status includes measures of the aquatic flora and fauna, nutrients, salinity, temperature and chemical pollutants.

The Oslo Paris Commission (OSPAR Commission 2010) consists of 15 nations; including Belgium, Denmark, Finland, France, Germany, Iceland, Ireland, Luxembourg, The Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and United Kingdom. The Commission focuses on protecting land-based sources, ecosystems, biodiversity, and economic development in the North-East Atlantic. Relevant strategies to this topic include a Biodiversity and Ecosystem Strategy and an Eutrophication Strategy as part of the North-East Atlantic Environment Strategy. The overall goal is to manage anthropogenic activities that impact the maritime area, maintain healthy ecosystems, safeguard human health, and restore marine areas.

The Helsinki Commission is responsible for protecting the Baltic Sea in a multi-national effort between Denmark, Estonia, Finland, Germany, Latvia, Lithuania, Poland, Russia, and Sweden. There are four main segments to the Baltic Sea Action Plan: eutrophication, hazardous waste, biodiversity, and maritime activities (HELCOM 2007).

14

1.4.5 North AmericaMexico’s National Water Law is administered by the National Water Commission (CONAGUA) to preserve water and associated services to achieve sustainable use. The 2030 Agenda includes protecting water quality of rivers and lakes to meet needs of the population. One of the principles defined in the 2030 Agenda is to meet society’s needs with an adequate water quality and quantity to maintain public health. In Canada, the responsibility for water quality is shared between federal, provincial/territorial, and municipal levels of government, developing various governance mechanisms to protect and enhance the quality of Canada’s water resources, promote the wise and efficient management and use of water, and develop guidelines for water quality standards. Several Federal departments and agencies (including Environment Canada, Health Canada, and the Department of Fisheries and Oceans) work closely to address nationally significant freshwater concerns, ensuring national policies and guidelines are in place on environmental and health-related water issues, and are responsible for administering federal legislation such as The Canada Water Act, Canadian Environmental Protection Act, and Fisheries Act. The Canada Water Act sets the framework established the Environmental Quality Guidelines (CCME 2006) set by the Canadian Council Ministries of the Environment. Water guidelines include coverage of recreational waters, aquatic life, and drinking water supplies. Recreational guidelines protect against chemical and physical impairments such as aesthetics and nuisance conditions (CCME 1999). Other recreational use measures include cyanobacteria, temperature, turbidity, water clarity, and colour (Health Canada 2012).

Canada and the United States share many waterways, including the Great Lakes which are among the world's largest bodies of freshwater, and as such the two countries maintain several treaties and agreements (e.g. The International Boundary Waters Treaty Act) which provide a mechanism for cooperation and coordination in managing these transboundary waters. The Great Lakes Water Quality Agreement (GLWQA) between Canada and the United States identifies shared priorities and coordinates actions to restore and protect the chemical, physical and biological integrity of the waters of the Great Lakes.

In the United States, the Clean Water Act, Safe Drinking Water Act, and Harmful Algal Bloom and Hypoxia Research and Control Act provide the basic structure for monitoring and maintaining water quality standards. Specifically, the Clean Water Act established the structure for water quality standards in surface waters, and the Safe Drinking Water Act allows for standards to ensure the quality of drinking water. The Harmful Algal Bloom and Hypoxia Research and Control Act is primarily focused on detecting, predicting, controlling, mitigating, and responding to marine and freshwater harmful algal bloom and hypoxia events. Case study # - Numeric water quality criteriaIn the United States, the Clean Water Act requires states to identify designated uses of their waters and when necessary to develop science-based water quality criteria to ensure protection of the designated uses. Numeric water quality criteria are concentrations or levels of a pollutant that, if achieved, provide an expectation that designated uses will be supported. The U.S. Environmental Protection Agency established a national strategy for development of numeric criteria identifying chlorophyll-a as a nutrient-related response variable. Schaeffer et al.

15

developed an approach to numeric criteria using satellite remote sensing to derive chlorophyll-a (Schaeffer et al. 2013a; Schaeffer et al. 2012). This approach was illustrated using data from the State of Florida coastal waters (Figure 1). The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) satellite was used to determine a quantitative reference baseline to protect coastal waters from eutrophication impacts. Briefly, the 90th percentile of annual geometric means between 1998 and 2009 from the SeaWiFS mission was used to determine the criteria value (Figure 2). In addition, approaches to enable transition of assessments from the SeaWiFS to newer platforms such as the Moderate Resolution Imaging Spectroradiometer Aqua (MODIS) and the Medium Resolution Imaging Spectroradiometer (MERIS) were established.

Figure 3. Fixed coastal segments used in the numeric criteria approach for the Florida Panhandle. Numbers are coastal segment numbers ranging from 1 through 17.

Figure 4. Candidate criteria for each coastal segment calculated on the basis of the 90 th percentile of the annual chlorophyll-a geometric means for the Florida panhandle. Coastal segment numbers correspond to segments identified in Figure 3 above.

16

1.4.6 South America Brazil Law No. 9,433 of January 1977 established the National Water Resource Policy and National Water Resource Management System, which is implemented by the Brazil National Water Agency to ensure sustainable use of rivers and lakes. The Argentina National Constitution states people should have access to a healthy environment for human uses. The provinces of Argentina have developed specific environmental laws that regulate water quality. The Peru Water Resources Law established the National Water Authority (ANA) to manage water resources for sustainable use. In 1981, Chile enacted its new Water Code (later reformed in 2005) that guarantees the right of citizens to live in an environment free from contamination. Since then Chile’s National Commission on the Environment have made improvements in wastewater and industrial water treatment.

1.5 How satellite remote sensing can contributeData from satellite remote sensing can address and inform communities on water quality changes that impact societal uses (Vörösmarty et al. 2015). Mechanistic models are necessary for management, but current models cannot resolve events because of limitations on knowing when and where these events occur. Satellite data provides information on dynamic and ephemeral events over extensive spatial and temporal scales. Feasibility of incorporating satellite information into water quality monitoring has been demonstrated and operational applications are expanding. Many state, federal, and non-government organizations conduct field-based activities with limited resources to provide information. Satellites have the potential to help address the limitations of geographic extent and temporal coverage. When coupled with field-based observations, satellite data provide a more comprehensive ability to monitor, assess and forecast changes in the environment. Satellite information can be used to assess baseline conditions and to understand trends for management. Furthermore, inland and estuarine waters present a challenge due to complex hydrologic connections and spatial separation, variable ecological drivers, and anthropogenic stressors. Satellites provide an integrated and system level approach that will be beneficial as extreme events increase (IPCC 2012).

1.6 Challenges and Why Now?Tools required for satellite remote sensing of water quality such as in-water algorithms, atmospheric corrections, and land adjacency effects will continue to mature over the coming decade and will be addressed throughout this report. However, other challenges to integrating satellite remote sensing into water quality management still exist today and can only be solved with open and effective discussion and forum between scientists, stakeholders, and water quality managers. Schaeffer et al (2013b) found four main challenges inhibited the consideration of satellite remote sensing data in water quality management including perceived cost of purchasing data, lack of understanding accuracy of satellite products, questions about data continuity between missions, and programmatic support and understanding. This report is intended to begin addressing some of these challenges and is one of many forms of communication (such as the International Ocean Colour Science Meeting Breakout Session “Tools to Harness the Potential of Earth Observations for Water Quality Reporting and

17

Management” (IOCCG 2015) to continue an effective dialogue between the scientific community and stakeholders.

There is increased demand from state, federal, and non-governmental organizations for the ability to monitor water quality by re-purposing ocean color and land imaging satellites. Significant progress has been demonstrated in deriving data from inland and estuarine waters using satellite ocean color sensors such as the Moderate-resolution Imaging Spectroradiometer (MODIS) and MEdium Resolution Imaging Spectrometer (MERIS). Due to the coarse spatial resolution of these two sensors (1km and 300m respectively), water quality science and applications also depend on terrestrial sensors such as Landsat (Mouw et al. 2015; Palmer et al. 2015). We expect that with launch of Sentinel-2 and Sentinel-3 satellites, temporal resolutions will improve, approaching those associated with ocean color missions (Hestir et al. 2015). In addition, recent technology advances in cloud-based infrastructure allows for coordinated data sharing with centralized, open access, publically available data (Mouw et al. 2015).

References

Atech (2000). Cost of algal blooms. In (p. 42). Canberra, ACT, Australia: LWRRDC Occasional Paper 26/99. Land and Water Resources Research and Development Corporation

Barbier, E.B., Hacker, S.D., Kennedy, C., Koch, E.W., Stier, A.C., & Silliman, B.R. (2011). The value of estuarine and coastal ecosystem services. Ecological Monographs, 81, 169-193

Carr, G.M., & Neary, J.P. (2008). Water Quality for Ecosystem and Human Health, 2nd Edition. Ontario, Canada: UN GEMS/Water Program Office

CCME (1999). Recreational water quality guidelines and aesthetics. In, Canadian environmental quality guidelines. (p. 3). Winnipeg, CA: Canadian Council of Ministers of the Environment

CCME (2006). A Canada-wide framework for water quality monitoring. In W.Q.M. Sub-Group (Ed.) (p. 25). Victoria, BC: Canadian Council of Ministers of the Environment

CIESIN (2005). Gridded Population of the World, Version 3 (GPWv3): Population Density Grid, Future Estimates. NASA Socioeconomic Data and Applications Center (SEDAC), http://dx.doi.org/10.7927/H4ST7MRB. Accessed April 2016

Conley, D.J., Björck, S., Bonsdorff, E., Carstensen, J., Destouni, G., Gustafsson, B.G., Hietanen, S., Kortekaas, M., Kuosa, H., Meier, H.E.M., Müller-Karulis, B., Nordberg, K., Norkko, A., Nürnberg, G., Pitkänen, H., Rabalais, N.N., Rosenberg, R., Savchuk, O.P., Slomp, C.P., Voss, M., Wulff, F., & Zillén, L. (2009). Hypoxia-related processes in the Baltic Sea. Environmental Science and Technology, 43, 3412-3420

18

Costanza, R., Groot, R.d., Sutton, P., Ploeg, S.v.d., Anderson, S.J., Kubiszewski, I., Farber, S., & Turner, R.K. (2014). Changes in the global value of ecosystem services. Global Environmental Change, 26, 152-158

Cox, M.E., Johnstone, R., & Robinson, J. (2006). Relationships between perceived coastal waterway condition and social aspects of quality of life. Ecology and Society, 11, 35-59

Day, J.W., Hall, C., Kemp, W.M., & Yanez-Arancibia, A. (2012). Estuarine Ecology 2nd. Edition. Hoboken, New Jersey, USA: John Wiley and Sons, Inc.

DFID, EC, UNDP, & World Bank (2002). Linking Poverty Reduction and Environmental Management: Policy Challenges and Opportunities. In. Washington, DC: The World Bank

Diaz, R., Selman, M., & Chique, C. (2011). Global Eutrophic and Hypoxic Coastal Systems. World Resources Institute. Eutrophication and Hypoxia: Nutrient Pollution in Coastal Waters, docs.wri.org/wri_eutrophic_hypoxic_dataset_2011-03.xls

Diaz, R.J., & Rosenberg, R. (2008). Spreading dead zones and consequences for marine ecosystems. Science, 321, 926-929

Dodds, W.K., Bouska, W.W., Eitzmann, J.L., Pilger, T.J., Pitts, K.L., Riley, A.J., Schloesser, J.T., & Thornbrugh, D.J. (2009). Eutrophication of U.S. freshwaters: analysis of potential economic damages. Environmental Science and Technology, 43, 12-19

Ellis, E.C., Goldewijk, K.K., Siebert, S., Lightman, D., & Ramankutty, N. (2013.). Anthropogenic Biomes of the World, Version 2: 2000. NASA Socioeconomic Data and Applications Center (SEDAC), http://dx.doi.org/10.7927/H4D798B9. Accessed April 2016

Frost and Sullivan International (2010). Eutrophication Research Impact Assessment. . In. Pretoria, South Africa: Water Research Commission

Gibbs, J.P., Halstead, J.M., Boyle, K.J., & Huang, J.-C. (2002). An hedonic analysis of the effects of lake water clarity on New Hampshire lakefront properties. Agricultural and Resource Economics Review, 31, 39-46

Health Canada (2012). Guidelines for Canadian Recreational Water Quality. In A.a.C.C.B. Water,

Healthy Environments and Consumer Safety Branch (Ed.) (p. 155). Ottawa, Ontario: Health Canada

HELCOM (2007). HELCOM Baltic Sea action plan. In (p. 101). Krakow, Poland

19

Hestir, E.L., Brando, V.E., Bresciani, M., Giardino, C., Matta, E., Villa, P., & Dekker, A.G. (2015). Measuring freshwater aquatic ecosystems: The need for a hyperspectral global mapping satellite mission. Remote Sensing of Environment, 167, 181-195

IGOS (2006). A Coastal Theme for the IGOS Partnership — For the Monitoring of our Environment from Space and from Earth. In (p. 60). Paris: UNESCO

IOCCG (2015). Proceedings of the 2015 International Ocean Colour Science Meeting. In, International Ocean Colour Science Meeting (pp. 29-30). San Francisco, CA, USA

IPCC (2012). Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change. New York, NY, USA: Cambridge University Press

Kummu, M., Moel, H.d., Ward, P.J., & Varis, O. (2011). How close do we live to water? A global analysis of population distance to freshwater bodies. PLoS ONE, 6, e20578

Le, C., Zha, Y., Li, Y., Sun, D., Lu, H., & Yin, B. (2010). Eutrophication of lake waters in China: cost, causes, and control. Environmental Management, 45, 662-668

Macauley, M.K. (2006). The value of information: Measuring the contribution of space-derived earth science data to resource management. Space Policy, 22, 274-282

Major, C., & Lusht, K.M. (2004). Beach proximity and the distribution of property values in shore communities. The Appraisal Journal, Fall, 333-338

McClain, C. (2009). A decade of satellite ocean color observations. Annual Review of Marine Science, 1, 13-42

MEA (2005a). Ecosystems and human well-being : health synthesis. In J. Sarukhán, A. Whyte, P. Weinstein, & M.B.o. Review (Eds.). France (World Health Organisation, WHO Publications, Millennium Ecosystem Assessment by Corvalan, C., Hales, S. and McMicahel, A.– review editors J. Sarukhán, A. Whyte, P. Weinstein and other members of the MA Board review)

MEA (2005b). Ecosystems and Human Well-being: Current State and Trends, Volume 1. In R. Hassan, R. Scholes, & N. Ash (Eds.). Washington, DC

MEA (2005c). Ecosystems and Human Well-being: Synthesis. In J. Sarukhán, A. Whyte, & M.B.o.R.E. (Eds.). Washington, DC

Meyer, W.B., & Turner, B.L. (1992). Human population growth and global land-use/cover change. Annual Review of Ecology and Systematics, 23, 39-61

20

Michael, H.J., Boyle, K.J., & Bouchard, R. (1996). Water quality affects property prices:A case study of selected Maine lakes. In (p. 18). Maine Agricultural and Forest Experiment Station: University of Maine

Mouw, C.B., Greb, S., Aurinc, D., DiGiacomo, P.M., Lee, Z., Twardowski, M., Binding, C., Hu, C., Ma, R., Moore, T., Moses, W., & Craig, S.E. (2015). Aquatic color radiometry remote sensing of coastal and inland waters: Challenges and recommendations for future satellite missions. Remote Sensing of Environment, 160, 15-30MRC (2010). Environment Programme 2011-2015. In (p. 75): Mekong River Commission for Sustainable Development

Muller-Karger, F.E. (1992). Remote sensing of marin pollution: A challenge for the 1990s. Marine Pollution Bulletin, 25, 56-60

OECD (2012). OECD environmental outlook to 2050. OECD Publishing, 351

OSPAR Commission (2010). The North-East Atlantic Environmental Strategy: Strategy of the

OSPAR Commission for the Protection of the Marine Environment of the North East Atlantic 2010-2020. In (p. 27)

Palmer, S.C.J., Kutser, T., & Hunter, P.D. (2015). Remote sensing of inland waters: Challenges, progress and future directions. Remote Sensing of Environment, 157, 1-8

Pretty, J., Mason, C., Nedwell, D.B., Hine, R.E., Leaf, S., & Dils, R. (2003). Environmental costs of freshwater eutrophication in England and Wales. Environmental Science and Technology, 37, 201-208

Radstake, F., & Tuinhof, A. (2003). Water quality: assessment and protection. In R. Davis, & R. Hirji (Eds.), Water resources and environment technical note D.1 (p. 34). Washington, DC: The World Bank

Schaeffer, B., Hagy, J.D., & Stumpf, R.P. (2013a). An approach to developing numeric water quality criteria for coastal waters: a transitiion from SeaWiFS to MODIS and MERIS satellites. Journal of applied remote sensing, 7, 073544

Schaeffer, B.A., Hagy, J.D., Lehrter, J.C., Conmy, R.N., & Stumpf, R. (2012). An approach to developing numeric water quality criteria for coastal waters using the SeaWiFS satellite data record. Environmental Science and Technology, 46, 916-922

Schaeffer, B.A., Schaeffer, K.G., Keith, D., Lunetta, R.S., Conmy, R., & Gould, R.W. (2013b). Barriers to adopting satellite remote sensing for water quality management. International Journal of Remote Sensing, 34, 7534-7544

21

Sorokin, Y.I. (2002). The Black Sea: Ecology and Oceanography. Leiden: Backhuys Publishers

U.S. Census Bureau (2016). U.S. and world population. In

UNEP (2007). United Nations Environment Programme Annual Report. Nairobi, Kenya: UNEP

UNEP (2012). Water. Nairobi, Kenya: UNEP

UNESCO (2006). Water, a shared responsibility. In, The United Nations World Water Development Report 2. New York, USA

Vorosmarty, C.J., Green, P., Salisbury, J., & Lammers, R.B. (2000). Global water resources: vulnerability from climate change and population growth. Science, 289, 284-288

Vörösmarty, C.J., Hoekstra, A.Y., Bunn, S.E., Conway, D., & Gupta, J. (2015). What scale for water governance? Science, 349, 478-479

Wheeler, B.W., White, M., Stahl-Timmins, W., & Depledge, M.H. (2012). Does living by the coast improve health and wellbeing? Health and Place, 18, 1198-1201

WHO (2006). Meeting the MDG drinking water and sanitation target : the urban and rural challenge of the decade. In (p. 42). Switzerland: World Health Organization and UNICEF

WHO (2013). Water quality and health strategy 2013-2020. In W.H. Organization (Ed.) (p. 15). http://www.who.int/water_sanitation_health/publications

Wilson, S.G., & Fischetti, T.R. (2010). Coastline population trends in the United States: 1960 to 2008. In. Washington, D.C.: U.S. Census Bureau

22

Chapter 2: Introduction to deriving water quality measures from satellitesCaren Binding, Rick Stumpf, Blake Schaeffer, Andrew Tyler

2.1 IntroductionRemote sensing of water quality is based on the concept that measureable changes in the colour of the water caused by variations in the concentrations of key water quality constituents may be detected from a remote platform such as a satellite or aircraft. Chapter 1 defined water quality as those biological, physical, and chemical characteristics of natural water required to maintain beneficial use of a water body. There are several recurring basic needs of water quality monitoring programs which may include routine observations of water clarity (transmission, turbidity, Secchi Depth), algal biomass (chlorophyll-a), harmful or nuisance algal blooms (HNABs), trophic status, suspended sediment loads, temperature, nutrients, dissolved oxygen, organic/inorganic pollutants, microbial contamination. Some of these parameters have a definite and well-defined influence on the colour of water, some do not, and it is this distinction that determines whether or not it may be directly measurable with remote sensing. In this chapter we present a broad summary of the theoretical basis for remote sensing of water quality, including a discussion of what key water quality parameters can and cannot be measured, an introduction to platforms and algorithms, and advantages and limitations of the remote sensing approach.

Historically, in situ measurements and collection of water samples for subsequent laboratory analyses have been the conventional methods of water quality monitoring programs. While providing accurate information for particular points in space and time, acquiring such information in the required frequency and geographic coverage to adequately characterize spatial heterogeneity and temporal variability in water quality is often prohibitively expensive and logistically difficult. Remote sensing offers one of the most cost-effective, spatially and temporally comprehensive, tools for observing often highly dynamic water quality phenomena in coastal and inland waters. New and improved sensor technologies, novel algorithm development, and considerable improvements in data availability and image processing capabilities have resulted in major advancements in recent years in remote sensing of coastal and inland waters and a demonstrable confidence in derived water quality products. Many examples now exist where remote sensing is delivering operational water quality information, providing prompt, reliable, synoptic maps of water quality in support of in situ monitoring activities. A subset of these will be included here and in other chapters throughout this volume as case studies to demonstrate the state of development and implementation of water quality remote sensing applications.

23

2.2 Theoretical Basis of Water Quality Remote Sensing When sunlight reaches a water body, some of it is reflected directly off the surface, but most penetrates into the water column and interacts not only with water molecules but with organic and inorganic materials dissolved and suspended within the water (Figure 1). Materials that interact with light, termed optically active constituents (OACs), do so through the processes of absorption and scattering. These OACs have measureable, often unique, absorption and scattering signatures, also referred to as Inherent Optical Properties (IOPs). For practical purposes, OACs are often grouped into four main components: pure water, phytoplankton, mineral particles, and coloured dissolved organic material. By preferentially absorbing or scattering light at different wavelengths, OACs determine the intensity and colour of light scattered upwards back through the surface (the water-leaving radiance). It is this water-leaving radiance signal, detected by sensors mounted on a remote platform such as a satellite, which can be interpreted in terms of key water quality parameters.

The depth to which sunlight penetrates (and therefore the satellite signal represents) depends on both the wavelength of light and the clarity of the water through which it travels. In highly turbid waters the satellite signal may be representative of the upper centimeters of the water column while in clear waters up to tens of meters. In shallow, clear waters, sunlight may penetrate to the bottom substrate and be reflected back to the surface making bottom-reflectance an additional contributor to the water-leaving radiance. In such waters, bottom substrate type, submerged vegetation and water depth may be estimated.

As a passive remote sensing device, an aquatic colour satellite sensor measures the response of the water surface to solar illumination, and so provides meaningful data only during daytime, cloud free observations. Scanning devices, and the orbit of the satellite, allow observations to be made on a per pixel basis, providing instantaneous, synoptic images over large swaths of the globe. Imagery is geolocated such that for each pixel, one can obtain a measure of the water-leaving radiance (or the derived water quality parameter of interest) along with its location. Repeat imagery, therefore, affords a mechanism for tracking seasonal cycles, long term trends or episodic events over time for any specified area of interest.For a more detailed technical introduction to remote sensing and aquatic optics relevant to observations of coastal and inland waters the reader is directed to chapters 5 and 6 within this report, as well as IOCCG Report #3 (IOCCG, 2000).

24

Figure 1: Schematic describing key water quality concerns for coastal and inland waters in relation to satellite remote sensing of water colour.

2.3 What Can be Measured Using Remote Sensing?The primary products of any aquatic colour sensor are measures of either water-leaving or top-of-atmosphere radiance or reflectance at a range of visible and near-infra-red wavelengths. These form the basis of any subsequent water quality determinations. Often a suite of secondary optical properties (absorption, backscattering, diffuse attenuation, for example) can also be derived.

Water quality parameters that can be measured directly from these primary optical parameters are those that have a measureable effect on water colour through their ability to absorb or scatter visible light (B). There are many water properties of interest that cannot be measured remotely, and othersthat can only be derived by combining remotely-sensed data with in situ or ancillary data (C). Integrated approaches based on assimilation of remote sensing data into diagnostic and prognostic models have expanded the potential of remote sensing beyond retrieval of directly measurable quantities, for example, in studies of primary productivity, and carbon cycling. Other, non-optically active WQ parameters of interest (such as nutrients, organic/inorganic pollutants, microbial contamination, dissolved oxygen) although impossible to make direct measures, have been estimated through inference using proxy relationships and empirical models. Such relationships, however, cannot be relied upon as physics-based solutions; they are typically stochastic, may not be causal, and may have a limited validity

25

range, both spatially and temporally. The use of proxies depends strongly on the spatial/temporal robustness of the proxy relationship with a directly measureable WQ parameter. As such, proxy-based observations may be appropriate for regional, often qualitative applications more so than quantitative, global applications, unless linked to more mechanistic modelling approaches and ancillary data. (A) Primary optical parameters

● Water-leaving or top-of-atmosphere radiance or reflectance, required with adequate accuracy to minimize uncertainty in derived WQ products

● Derived IOPs (absorption , backscatter) – essential first step in many analytical methods to derive WQ (IOCCG, 2006)

(B) WQ parameters directly measureable from space

● Water Clarity - diffuse attenuation, Secchi disk depth, euphotic depth(Olmanson et al., Binding et al., Lee et al, Doron et al)

● Algal blooms - chlorophyll-a (total algal biomass), marker pigments such as phycocyanin (for HNAB discrimination), quantitative bloom indices such as aerial extent, timing, duration (Stumpf et al, Binding et al, Odermatt, Matthews, Tyler)

Substrate - submerged aquatic vegetation, bathymetry, benthic ecosystems (Vahtmäe & Kutser, 2007….)

● Suspended minerals – River plumes, shore erosion, bottom resuspenion, whiting events (CaCO3 precipitation) (Refs)

● Coloured Dissolved Organic Matter (Refs)● Temperature (using thermal infrared or microwave technology)(Refs)● Surface oil slicks (typically detected using microwave SAR but optical imagery offers

some additional benefits, e.g. discriminating surface oil slicks from algae) (refs)

(C) Additional Desirable WQ parameters not directly measureable from space● Trophic status - can be inferred if reasonably defined by (B) from algal biomass and/or

water clarity (Matthews et al, 2012; Binding et al, 2011)● Primary productivity, carbon cycling – can be estimated from satellite-derived

chlorophyll and algal absorption, using various models/assumptions for contributing factors such as algal depth distributions, light availability, temperature, and algal physiology variables (e.g. photosynthetic quantum yield – the efficiency of energy conversion to organic carbon). See Lee et al. (2015) for recent review.

● Algal Toxins – toxins are colourless and so are not directly measureable from aquatic colour. Toxicity is often correlated regionally with phycocyanin (e.g. Hunter et al, 2010) but not all cyanobacteria are toxic, and the relationship between toxic strain cell density and toxin concentration has been shown to be widely variable (McGregor and Fabbro, 2000,).Nutrients – dissolved nitrogen and phosphorus are colourless so they do not add to the optical properties of a waterbody directly. However nutrients can affect colour indirectly by influencing the growth of algae. While nutrient concentrations may be related to algal biomass on a broad scale (particulate P at least, not always so for TP), a proxy-based approach to mapping nutrients may vary widely in space and time due to

26

complex nutrient-algae dynamics including variations in bioavailability and other growth-limiting factors.

● Dissolved Oxygen – not measureable directly….● Pollutants/metals/microbial contamination - not measureable directly, some regional

links to turbidity (e.g. E. Coli released from beach sediment after resuspension and/or rain events)

2.4 Water Quality Retrieval AlgorithmsThe signal retrieved by a satellite sensor is confounded by large contributions from the intervening atmosphere and sun glint, for which corrections must first be applied. Rigorous atmospheric correction algorithms therefore form a critical primary step in satellite image processing for most water quality applications. After removing the effects of the atmosphere and reaching a measure of the water-leaving radiance, extracting quantitative information on water quality parameters requires the application of appropriate algorithms. Retrieval algorithms take many forms, from simple empirical approaches, to complex multivariate and neural network techniques, to semi-analytical radiative transfer models. (Figure **Note from Andrew - we could add a figure here illustrating the types of algorithm).. Tyler et al., (accepted) categorised three groups of existing inwater algorithms for the estimation of biogeochemical properties:

(1) The first group comprise algorithms whose performance is driven by the shape and magnitude of the water leaving radiance signal, where a single band of the water leaving reflectance or single ratio or combinations of multiple spectral bands in the band ratio algorithms, utilizing specific spectral attributes that appear to be unique to specific parameters such as Chl a. Such ratios are used to derive empirical relationships with in water constituents including Chl a. This approach is limited to the need for well define features, which under conditions of low concentration or poor atmospheric correction cannot be guaranteed.

(2) The second group of algorithms are more physics based bio-optical models, driven by the relationship between the Inherent optical properties (IOPs) and the water leaving radiance signal. These approaches implement spectral inversion techniques such as spectral optimization, linear and non-linear matrix inversion to retrieve IOPs [IOCCG 2006] and the biogeochemical properties. The inversion approach has often outperformed those in group (1), but depend strongly in the initial parameterization (IOCCG 2000).

(3) The final group of algorithms are based on machine learning techniques. Based largely on neural networks, support vector machines and hybrid active learning models, these approaches appear to be robust to noise and allow the application of complex bi-directional radiative transfer models. They also have the advantage of reducing the impact of residual atmospheric correction errors. However, these machine learning require large training data sets. Extending these training data sets with either empirical or simulated data can result in poorer accuracies and increased computational times.

27

The choice of approach or combination of approaches depends largely on the level of optical complexity of the waters under consideration, source and format of imagery, the amount of information available regarding in situ optical properties, and computation time/resources. For a more in-depth discussion of specific algorithm approaches for water quality retrievals the reader is directed to Chapter ** of this volume.

2.5 Platform, Resolution, and Data Processing ConsiderationsInstrument and orbit specifications define the pixel size (spatial resolution), revisit (how frequently a site is viewed, or temporal resolution), and number of wavebands which are measured (spectral resolution). Much as the human eye sees visible light in three bands (red, green, and blue), aquatic colour sensors view a scene in several spectral bands. Typically the wavelength range for water quality applications extends across the visible and near-infra-red domains of the electromagnetic spectrum, with sensors ranging from multi-spectral (measuring at multiple discrete wavelengths) to hyperspectral (measuring near continuously across the full spectrum). Historically, sensor and orbit capabilities have resulted in trade-offs between the spatial, temporal, and spectral resolutions available on satellite missions. For example, the Landsat series of satellites provide high spatial resolution (~30m) but coarse spectral resolution and revisit times of 16 days, while a sensor such as MODIS operates with a coarser ground spatial resolution of 1km, enhanced spectral resolution, and with daily revisit. Such limitations are decreasing with advances in modern sensor technology but still play a part in remote sensing application decision making. Table 2 summarises the existing and forthcoming satellite capabilities for value for water quality applications. Alternate platforms such as aircraft or more recent experimental devices such as drones provide additional solutions for targeted nearshore high resolution observations (although frequently used for research or demonstration purposes, airborne remote sensing is often prohibitively expensive for operational use in water quality monitoring). For brevity, this chapter deals only with satellite platforms and their applications (note from Andrew - we could add a table from our review paper).

28

Table 2. Summary of Satellite Platform Capabilities (from Tyler et al., accepted)

2.6 Advantages and Limitations of Remote Sensing for Water Quality Monitoring

Advantages:● Synoptic observations over large areas● Frequent revisit times● Increasing commitment from space agencies to data and product continuity allowing for

continuous time series observations● Advancements in data delivery and processing allowing for NRT (near-real-time)

observations – providing value in early warning● Variety of data availability – large suite of products, resolutions● Providing data for lakes that have not been monitored, or with limited monitoring data,

especially remote sites● Archive of data allowing for retrospective analysis● Non-commercial imagery allows for low cost monitoring● Increasingly robust algorithms for optically complex coastal and inland waters

29

● Increasingly available validation datasets● Increasing sophistication of satellite sensors offering scope for advanced suite of

products● Increasing capability as operational monitoring tool

Figure 1 Example output from ESA’s Sentinel 2 of the Baltic Sea (note from Andrew - do we want to add something like this to illustrate what is possible from Sentinel 2)

Limitations/challenges● Effective revisit frequency is reduced by cloud cover over area of interest● Potential spatial/temporal bias – cloud cover or atmospheric phenomena affecting data

validity may result in nearshore and seasonal bias in data availability● Products directly measurable from remote sensing are limited to those with a discernible

effect on colour ● Surface only observations – as the remote sensing signal originates from the sunlit

surface waters, WQ observations are representative of that layer only● Nearshore signal contamination such as adjacency effects (whereby the bright

reflectance from land contaminates nearshore pixels), bottom reflectance (in clear shallow waters the bottom substrate can contribute to the remote sensing signal) and

30

complex atmospheric conditions, contribute to the challenges of developing robust retrieval algorithms in coastal and inland waters

● Satellite data, derived products, and processing software, are not always easily manipulated/interpreted by the non-specialist

2.7 Case Studies

2.7.1 Operational Satellite Water Quality MonitoringSeveral examples exist where satellite derivations of water quality parameters have been adopted for operational water quality monitoring.

Operational services from Geneva? Reference operational services… Australia, - decker. The Great Barrier Reef Marine Park Authority uses methods developed by CSIRO for systematic and cost-effective assessment of water quality variables (chlorophyll and suspended sediments) in the Great Barrier Reef lagoon using satellite data. Baltic Sea –Hansson, M., & B. Hakansson, 2007, "The Baltic Algae Watch System - a remote sensing application for monitoring cyanobacterial blooms in the Baltic Sea", Journal of Applied Remote Sensing 2007, 1(1):011507.

2.7.2 Eutrophication and harmful algal bloomsEutrophication and harmful algal blooms are perhaps the most pervasive problems affecting inland and coastal water quality globally. Several key studies have demonstrated the real potential for remote sensing in monitoring of algal blooms and HNAB detection. Matthews (SA), Stumpf, Binding (N. Am), Taihu…Europe…phenology from Lake Balaton (Tyler)

2.7.3 Water Clarity

Water clarity is a key water quality measure that reflects not only ecosystem health but also Great Lakes – Binding et al L&O 2015Minnesota lakes - Olmanson

31

References

IOCCG, 2006. Remote sensing of inherent optical properties: fundamentals, tests of algorithms, and applications. In: Lee, Z.-P. (Ed.), Reports of the International Ocean-ColourCoordinating Group, No. 5. IOCCG, Dartmouth, Canada, p. 126. Lee, Z., Marra, J., Perry, M.J., Kahru, M. 2015. Estimating oceanic primary productivity from ocean color remote sensing: A strategic assessment. Journal of Marine Systems, Volume 149, September 01, 2015, Pages 50-59 Matthews, Bernard, and Robertson. 2012. An algorithm for detecting trophic status (chlorophyll-a), cyanobacterial-dominance, surface scums and floating vegetation in inland and coastal waters. Remote Sensing of Environment, Volume 124, September 2012, Pages 637–652 Binding, C. E., Greenberg, T. A., Jerome, J. H., Bukata, R. P. 2011. Time series analysis of algal blooms in Lake of the Woods using the MERIS maximum chlorophyll index. Journal of Plankton Research, 33(12):1847-1852. Binding, C. E., T. A. Greenberg, S. B. Watson, S. Rastin, and J. Gould. 2015. Long term water clarity changes in North America’s Great Lakes from multi-sensor satellite observations. Limnology and Oceanography, in press.Hunter, P.D., Tyler, A.N., Carvalho, L., Codd, G.A., Maberly, S.C. 2010. Hyperspectral remote sensing of cyanobacterial pigments as indicators for cell populations and toxins in eutrophic lakes. Remote Sensing of Environment, Volume 114, Issue 11, November 2010, Pages 2705-2718 McGregor, G. B., and Fabbro, L. D, 2000. Dominance of Cylindrospermopsis raciborskii (Nostocales, Cyanoprokaryota) in Queensland tropical and subtropical reservoirs: Implications for monitoring and management. Odermatt, D., Gitelson, A., Brando, V.E., Schaepman, M. 2012. Review of constituent retrieval in optically deep and complex waters from satellite imagery. Remote Sensing of Environment, Volume 118, 15 March 2012, Pages 116-126 Stumpf, R.P., Wynne, T.T., Baker, D.B., Fahnenstiel, G.L. 2012. Interannual variability of cyanobacterial blooms in Lake Erie. PLoS ONE. 7 (8), e42444

Olmanson, L.G. , Bauer, M.E., Brezonik, P.L. 2008. A 20-year Landsat water clarity census of Minnesota's 10,000 lakes. Remote Sensing of Environment, Volume 112, Issue 11, 15 November 2008, Pages 4086-4097 Vahtmäe, E., Kutser, T. 2007.Mapping bottom type and water depth in shallow coastal waters with satellite remote sensing. Journal of Coastal Research, 50, Pages 185-189

32

33

Chapter 3: Complementarity of in-situ and satellite measurementsSteve Greb*, Blake A. Schaeffer, Paul DiGiacomo, Menghua Wang, Daniel Odermatt, Evangelos Spyrakos

3.1 Water quality monitoring approachesThis chapter provides an overview of water quality monitoring approaches. Various tools or options are available depending on the needs and objectives of the user. In addition, the technical advantages and disadvantages of the approaches including accuracies, limits of detection and representativeness are discussed in the context of stakeholder frameworks and institutional requirements. The objectives of monitoring may include water characterization for trends, problem identification, compliance, design of pollution and remediation programs, and emergency response (Mahadevan et al., 2003). New and future methodologies including complimentary monitoring will be highlighted along with providing references to resources to those who wants to further pursue these strategies.

Water quality monitoring efforts need to recognize the underlying temporal and spatial variability of relevant processes as summarized in Figure 1, and monitoring strategies need to be designed appropriately.

34

Figure 1: Spatial and temporal scales of processes related to water quality in inland and near-shore systems and their coverage by satellite sensors (red box) airborne data (blue box) and field-based efforts (green box).

3.2 Discrete in situ measurements and water sample analyses

3.2.1 Prevalent techniques and suitabilityHistorically, in situ discrete sampling has been the only way in which water quality management authorities could assess the condition of inland waters. The frequency at which point-based sampling programs are carried out may vary from daily for drinking water reservoirs to weekly, monthly or seasonal. Sampling schedules may often be dependent on perceived threats and/or the intended uses for the water. Currently, most water quality monitoring programs rely on discrete in situ point samples.

The advantages include flexibility to measure a wide range of chemical, biological and physical parameters, including trace metals, organic and inorganic micro pollutants, nutrients, cyanobacteria, and optical properties. The measurements are usually highly replicable within a sample; and space and time considerations such as depth profiles and time of day are investigator dependent. Disadvantages include logistics, potentially missing important events due to foul weather or long distances, collecting representative samples over large areas, consistency in methodologies and open access to data. Table 1 lists several methods and instrumentations that are being typically used for discrete (D) and continuous (C) field measurements of the selected water quality parameters.

35

Water quality concerns

Relevantmeasures

Typical field-based methods and instrumentations

Turbidity Nephelometric Turbidity Units (NTUs) and

attenuation coefficients

· Nephelometric ratio (D)· Nephelometric, backscatter sensors and transmissometers (D, C)

Visual water clarity · Secchi disk (D)

Suspended solids

Sediment concentration/total suspended matter

· Gravimetric method (D)· Nephelometric; backscatter sensors and transmissometers (D, C)

Chlorophyll Chlorophyll a · Spectrophotometric; HPLC-based and fluorometric methods (D)· Fluorescence sensors (D, C)

Harmful algae Phycocyanin (

Phycocyanin/chlorophyll a as index for

cyanobacteria presence)

· Spectrophotometric and fluorometric procedures (D)· Fluorescence sensors (D, C)

Identification · Microscopy (D)· molecular methods (D)· chemotaxonomic methods (D)· flow cytometry (D, C)

Cell abundance/biovolume

· Microscopy (D)· Chlorophyll a/phycocyanin (D, C)· Imaging particle analysers (D, C)

Toxins · Biological assays (D)· Chromatographic methods (D)

Dissolved organic carbon

Chromophoric dissolved organic matter

· Spectrophotometric methods (D)· Fluorescence sensors (D, C)

Submerged habitat

Density · Boat or in-water observations (D)

Health

Habitat

Temperature Temperature · Thermistor temperature sensors (D, C)

36

Dissolved Oxygen

Dissolved Oxygen · Iodometric (Winkler) method (D)· Optical and amperometric sensors (D, C)

Dissolved inorganic micronutrients

Dissolved inorganic micronutrients

· Colorimetric methods (D)· Spectral photometer sensors for selected micronutrients (D, C)

Table 1: Overview of the typical field-based methods and instrumentations used for discrete (D) and continuous (C) field water-quality measurements.

3.2.2 ExamplesThe parameters, methods and protocols in use vary strongly. A reference book by the UNEP GEMS/Water programme provides an overview of several hundred different techniques applied by GEMS partner institutions (GEMS/Water, 2014). In the case of turbidity, references to visual, photometric and two nephelometric techniques are given, along with protocols for probe and turbidimeter measurements, providing for turbidity in units of JTU, NTU or FNU.

3.3 Continuous in situ measurements by automated moorings, drifting buoys and AUVs

3.3.1 Prevalent techniques and suitabilityThere have been significant technological developments in such instruments in recent years, opening up the window into a greater understanding of water quality variability, particularly at short time intervals. In situ systems can run continuously and capture daily and diurnal as well as extreme events. Examples of in situ measurements from permanently installed instrumentation include algal pigments, Colored or Chromophoric Dissolved Organic Matter (CDOM), nitrates and turbidity as well as physicochemical measurements including conductivity, salinity, dissolved oxygen and temperature. An excellent review of commercially available in-situ sensors can be found in Zielinski et al. (2009). The obvious advantages are the ability to measure several physicochemical and bio-optical variables simultaneously and continually at one location with high sensitivity (note from Steve - and high frequency?). In situ sensors can store data, or transmit data in (near) real time using radio telemetry, mobile phone or wireless networks. The limitations of autonomous systems include limited number of parameters one can potentially measure, surrogate measurements not reflecting true quantity, and equipment maintenance, vulnerability, power usage, fouling and high capital expense. Much like in situ collection, autonomous instrument do not address spatial representation.

3.3.2 ExamplesSummarize specific practical cases (note from steve - Trying to think of something similar to the GEMS document to match your example for discrete. Something like an IOOS for

37

in-land/coastal waters? There is an aquatic sensor workgroup, but they only have protocols for rivers and streams. http://www.watersensors.org/qamatrix.html)

3.4 Remotely sensed water quality products

3.4.1 Prevalent techniques and suitabilityOptical remote sensing of water quality makes use of various ground-based, air- and spaceborne platforms. Polar orbiting satellites are currently the most suitable platforms for monitoring surface water quality dynamics, enhanced by UAVs and the upcoming geostationary satellites. Remote sensing implicates a resolution trade-off. As far as polar orbiters are concerned, Landsat-8 or Sentinel-2 provide for a 10-30 m spatial resolution at the expense of a relatively high revisit time of 10-16 days and limited spectro-radiometric sampling. On the other hand, moderate resolution instruments (MODIS, MERIS, VIIRS and Sentinel-3) have lower spatial resolution but a significantly better potential to sample spectral absorption features and a revisit time of 2-3 days. Geostationary satellites such as GOCI and the GEO-CAPE (note from Steve - do we spell this out here?), enable almost equivalent spatial and spectro-radiometric resolutions at revisit times as low as 1 hour, but for limited geographic areas. Various retrieval algorithms are available whose suitability depends on the selected sensor and the parameter of interest, as well as on the optical water type (see Chapter 2 for more details).

The strength of remote sensing techniques lies in their ability to provide both spatial and temporal views of surface water quality parameters that is typically not possible from in situ measurements. Additional advantages include spatially synoptic objective measurements at a point in time, resulting in increasing data availability. Some satellite programs such as Landsat have a legacy of historical imagery, which may be processed for retrospective analysis (e.g. long-term trends). Many satellite imageries are now offered free and easily assessable. A strong science program supporting satellite remote sensing continues to improve robust algorithms (see Matthews, 2011; Odermatt et al., 2012b for recent reviews) and develops information and operational products for societal use. There are disadvantages to the remote sensing approach to monitoring. One of the main drawbacks is weather (cloud cover) and atmospheric conditions (smoke, haze and dust), which interfere with the optical signal. Inclement weather periods may also be times of critical monitoring such as runoff events and satellite monitoring will be ineffective. Other disadvantages include only those parameters that are optically active can be detected, other need to be inferred. Spatially, remote sensing can generally only measure surface water quality conditions (no profiles) and smaller lakes, rivers and streams may be excluded because of pixel size. Some satellites have long revisit periods, thereby decreasing the probability to monitor episodic events.(note from Steve - Probably need to touch on land adjacency effects, that satellites can’t see near land-water boundary?)

3.4.2 ExamplesNote from Daniel: - Diversity II inland water products

38

- CSIRO reef rescue monitoring programme- Great Lakes Observing System (http://data.glos.us/portal/)

3.5 Discussion of measurement statistics and related attributes for all three approaches(note from Daniel - I would change this title to make it more catchy; “recommendations on combined monitoring concepts“, and then maybe separately address spatial, temporal, vertical dimension and accuracy in subchapters)

The three approaches discussed in the previous sections all have their merits and drawbacks. The monitoring objective will dictate which approach is most appropriate. Often a balance may be required with the general reciprocal relationship between spatial coverage and amount of information. Zielinski et al. (2009) illustrated this relationship in the context of marine hazardous substances. Thus, in situ measurements evaluate processes at different time/space scales than satellite-based measurements.

Note from Steve: We probably need to think about this more. Airborne campaigns are not cheap, at least right now. Also, think we need to change “ocean color” to water quality or something else. Spatial resolution looks off too. Satellite could be 10s-1000s of meters, in-situ can be 0-10s kms.

Irrespective of the measurement approach it is important to understand the measurement uncertainties in the context of its use. Because of the rapid evolution of sensor technologies, the challenge has been to provide systematic evaluations of these emerging technologies and often researchers and managers rely too much on manufacturer’s specifications who do not fully understand the instruments performance in the context of its fields application (Waldmann et al., 2010). It is critical to have agreed upon calibration, quality control/ quality assessment

39

procedures because data is increasingly being made available to the larger end user community, not just project specific use.

3.5.1 Accuracy The accuracy of water quality constituent concentrations derived from remote sensing will vary depending on derived product region of interest, and time of the year. Accuracy will also vary due to concentration of the substance, the matrix of other optically active substances in the water, availability of appropriate wavebands, atmospheric corrections, and the accuracy and representativeness (see below) of the in situ data set. Typical accuracy targets for most ocean colour products are better than ±30% (IOCCG 2000) and inland and coastal waters are generally expected to have the same magnitude of accuracies. Examples??? The additional value of satellite remote sensing is not always absolute accuracy of individual retrievals, but synoptic and frequent coverage of numerous waterbodies. Trends?? Threshold? Even with larger uncertainties in accuracy estimates, parameter concentration change could be detected if the product was derived with a consistent methodology (Stumpf et al. 2003; Hu et al. 2005). Algorithms applied to operational efforts need pass the rigors of peer-reviewed publications and be robust in covering the range of constituent concentrations with quantifiable errors.

Parameter Accuracy Limit of Detection

Discreet Contin. RS Discreet Contin. RS

Water Clarity

Chl-a/Biomass

HAB

Sediment/Turbidity

Temperature

Phosphorus

Nitrate

Dissolved Oxygen

Table 2: Comparitive summary table

40

Water quality parameters

Typical accuracy/limit of detection

Visual water clarity · 0.1-70 m (D) with typical accuracy ± 0.01m or higher depending on the object size and weather/water conditions.

Turbidity · 0.1-50 NTU by nephelometry (D) with possible extension of the range with dilution of the sample and accuracy around ± 0.01 NTU.· Nephelometric sensors can operate between 0.1 to 3000 NTU with the accuracy to depend on the turbidity range.· Backscatter sensors can measure higher values than the above but they normally show lower sensitivity at low ranges due to the emitter-detector arrangement.· The range of attenuation coefficients measured with transmissometers is generally low and depends on the emitter-detector distance.

Chlorophyll a · Spectrophotometric methods (D) show generally higher detection limits than HPLC and fluorometric methods while they overestimate the chla concentration when phaeophytin or/and certain carotenoid pigments are present.· Fluorometry (D, C) provide better detection limits than spectrophotometric methods but overestimate or underestimate depending on algae community and pigment composition (D) and can be affected by biotic and abiotic parameters (C).· HPLC methods may result lower concentrations for chlorophyll a than the above due to the ability to separate the pigments.

Phycocyanin ·

Etc. ·

3.5.2 Representativeness considerations Space and time considerations of system representativeness are important when considering the three monitoring approaches. For example, a discreet surface sample may represent a depth of 0 to 0.5 meters at a single point in a lake. Its representativeness will depend on environmental conditions such as light, wind, currents, degree of both vertical and horizontal mixing and algal bloom patchiness. Patalas and Salki (1993) suggested an increase in the number of sampling locations is required with increased lake size to properly represent lake conditions. Conversely, remote sensing generally represents a greater unit of coverage (e.g. one pixel= 300m2 – 1km2 for medium resolution sensors) and presents a better picture of the

41

aggregated spatial distribution over a much greater areas. Using MODIS satellite imagery, Ostendorf (2011) illustrated how taking a highly accurate transect in a Australian coastal area would be essentially sampling conditions of a chaotic system and is potentially a waste of time and resources/ effort, resulting in little increased understanding of the system variability.

Figure 2. Measurements resolution

Comparisons (match-ups) are often made between remotely sensed parameters and in-situ measurements. These exercises are sometimes incorrectly termed “ground-truthing” for each of these approaches measure different spatial scales. For example, in a mesotrophic lake, a surface sample collected with a Kemmerer bottle might estimate/represent chlorophyll concentrations in the top 0.1 m2 surface area to a depth of 0.5 meters (volume=0.05 m3), whereas satellite-derived chlorophyll concentrations are based on pixel size and photic depth. The depth to which the satellite “sees” will depend on the concentration of water constituents, and is wavelength dependent, meaning that red wavelengths are attenuated more than blue wavelengths. A Landsat pixel represents an area of 900 m2 and, when averaged over a 3x3 pixel matrix, represents a surface area of 8100m2 (a volume of 8100m3 for a one meter photic depth). The extent to which discrete and satellite samples correlate (match-up) depends on the underlying spatial variability (uniformity) and how well the point sample represents the greater average satellite coverage. In some cases the correlation between discrete transects and satellite observations is poor (see Galat & Verdin, 1989). Kutser (2004) concludes a single-point in situ measurement is inadequate for validation of satellite chlorophyll estimates, particularly during cyanobacterial blooms with surface scums.Other factors may influence the differences seen in match-up values, including sub-surface algal maxima, timing of sample collections, analysis methodologies and weather conditions. The use of continuous automated instrumentation and well-timed discrete samples (within 30 minutes of satellite overpass) will increase the correlation between satellite and in situ observations. The question is not which monitoring approach is better but which monitoring approach is best suited for the intended objective. In situ sampling might be most appropriate for fine-scale, short-term temporal processes whereas remote sensing might be better suited for larger scale regional processes or trends over time. The continued improvement in remote sensing capabilities has provided new opportunities to the end-user community.

42

3.5.3 Complementary role of remote sensing with in situ measurementsThe power of utilizing remote sensing is not as a replacement of traditional sampling approaches but in its complementary role. Use of either in situ or remote sensed data has limited value as compared to these monitoring tools used together (Delvin et al. 2013). Remote sensing has the ability to build or extrapolate limited in situ measurements by expanding the spatial coverage and providing better characterization of the water body. Combining these two data sources in their study of chlorophyll variability in the Great Barrier Reef, Delvin et al used remote sensing to extrapolate point data into a spatial framework and better determine the conditions in which chlorophyll concentrations became elevated. Kallio et al. (2003) used airborne remote sensing to present a complete spatial chlorophyll coverage for two Finnish lakes. Comparing mean chlorophyll concentrations from the three routine lake stations with mean chlorophyll obtained from the remote sensing product yielded a 47 percent difference. Remote sensing chorophyll products can be used to help target “hot spots” for more intense sampling and identify future sampling stations that better represent lake conditions.

3.5.4 Complementary role of remote sensing with modellingRobust models are necessary for management, but current models cannot resolve events because of limitations on knowing when and where these events occur. The complexity and dynamic nature of inland and coastal waters can be difficult to characterize. New modelling methods are being advanced which merge satellite observations as input parameters in an effort to better understand underlying biogeochemical processes and use their predictive powers to forecast future changes. Modelling coastal and inland waters is complex because connections between environmental variation and ecological systems occur across multiple and interactive spatial, temporal and organizational scales ((Delvin et al. 2013). Modelling tools can simulate and predict transport of nutrients, pollutants and contaminants. Modelling efforts can expand the temporal coverage of the remote sensing products. For example, persistent cloud cover can hamper satellite remote sensed products and these gaps can be filled in with modelling efforts. Remote sensing data can be used to setting initial conditions, boundary conditions, model parameterization, forcing and validation. Remotely sensed data assimilation into biogeochemical models is commonly used to make midcourse simulation corrections and model parameterization. A key issue in utilizing remote sensed data in modelling efforts is whether the data products represent the same quantities modelled. This can be due to depth of light penetration or imagery spatial scales vs. modelling grid.

3.5.5 CostsIt is difficult to do a comparative cost analysis of the three major monitoring approaches describes above in section XX. Each approach has its strengths and weaknesses and though cost is one factor, one must start will the monitoring objectives and determine whether each of the approaches will provide the necessary information. Clearly, costs of traditional network approaches continue to rise. The implementation of the European water directive framework

43

costs are estimated at 6-11 million Euros in UK and Netherlands (Nocker et al., 2007) and 6-9 million Euros for Norway, Sweden and Finland (Halleraker et al., 2012).Autonomous data sondes are becoming increasing popular for monitoring constituents such as dissolved oxygen, temperature, conductivity, pH, turbidity and NO3. Decreasing costs and increased memory allows enormous amounts of data to be collected in key locations. Though initial costs may be 5-10K for some of these devices, the cost per measurement (10’s thousands over a season) will be quite low. A good resource guide to current in situ monitoring equipment costs and new emerging technologies was compiled by the Chesapeake Conservancy’s Conservation Innovation Center.http://www.chesapeakeconservancy.org/images/Low_Cost_Water_Quality_Monitoring_Needs_Assessment_small1.pdf Dekker and Hestir (2012) provide a cost-benefit analysis for setting up a prototype earth observation monitoring system for Australia utilizing either Landsat and MODIS/MERIS satellites. Their analysis provides both the initial parameterization of service and the operational aspects and builds upon current research activities and infrastructure. For other countries or regional programs, the situation may be different. Campbell (2013) points out it is probably not possible to provide a generic cost assessment of remote sensing given each situation has variable factors e.g. expediency of collecting a point sample vs. need for spatial characterization. Bukata (2005) suggests a perspective of “cost per monitored area”. For example, if only one chlorophyll sample/year was required on a 5000 hectare (ha) lake, then the costs for remote sensing (infrastructure and FTE time) is probably unreasonable. On the other hand, if water quality measurements were required for every 10 ha of the 5000 ha lake, then remote sensing would be more attractive. In Wisconsin, where water clarity for 8000 lakes is determined annually from Landsat (LINK), costs to generate those values are approximately US$1.50. It would cost far more to use a traditional approach involving staff time, boats, travel costs, etc.

In conclusion, cost considerations, though important, must be taken in the context of the monitoring objective where the informational needs define the monitoring requirements (note from Steve: I think we are missing a huge element here, based on the IJRS Barriers paper.” We need to highlight the cost to acquire the imagery $0. And the approximate cost to implement processing (e.g. decent computer, seadas software free of charge, perhaps a training course $3,000K., etc).

3.6 Resources

3.6.1 In situ data basesThe US Water Quality Portal (WQP) is a cooperative service sponsored by the United States Geological Survey (USGS), the Environmental Protection Agency (EPA), and the National Water Quality Monitoring Council (NWQMC). It serves data collected by over 400 state, federal, tribal, and local agencies. http://www.waterqualitydata.us/

44

The United Nations Global Environment Monitoring System (GEMS) Water Programme is dedicated to providing environmental water quality data and information of the highest integrity, accessibility and interoperability. These data are used in water assessments and capacity building initiatives around the world. http://www.gemstat.org/The Water Data Centre, hosted at the European Environment Agency (EEA), provides a central access point to several web-services: interactive maps, data viewers, European datasets and indicators. These services are mostly based on reporting from countries as part of implementation of EU directives or via the Eionet framework. www.eea.europa.eu/themes/water/dc

3.6.2 Integrated data systemsThe Copernicus Marine Service has been designed to respond to issues emerging in the environmental, business and scientific sectors. Using information from both satellite and in situ observations, it provides state-of-the-art analyses and forecasts daily, which offer an unprecedented capability to observe, understand and anticipate marine environment events http://marine.copernicus.eu/ Temporary contents possibly on the interface with other chaptersIV. Utilizing remote sensing in current and future monitoring efforts (other chapter)a. Institutional hurdles (reg. requirements, standard methods, history)b. Developing a new WQ monitoring paradigm (establishing continuity, data uptake, trust).c. Ecological systems are dynamic and interact across a wide range of interacting spatial, temporal, and organizational scales (Ostendorf, 2011) multipled. Data quality should be documented as a function as a combination of accuracy metrics and spatio-temporal representativeness.e. Capturing temporal and spatial at correct environmental frequencies and pattern variabilities critical for the decision making process.f. Using multiple monitoring tools, taking advantage of each’s strength and complimentary role. A combination of scales may provide additional insight and decision-making information (Zielinski et al, 2009). Though in situ continuous monitoring may provide fine-scale information on diurnal variability, combined with “match-up” satellite imagery can provide information for modeling studies, surface structure at the meso-scale and applications for fisheries information V. Water quality data repositories (not sure whether this should go in this chapter?)a. Current sources (in situ measurement) i. GEMS The United Nations Environment Programme Global Environment Monitoring System (GEMS) Water Programme is the most complete system worldwide on fresh water quality ii. EPA Storet iii. EEAb. Discussion of future remote sensing data archive needs, Water MLVI. Guidance for managers (reference section)a. How to acquire EO information

45

b. Tools availablec. Costsd. Infrastructuree. Training

References

Bukata, R.P. (2005). Satellite Monitoring of Inland and Coastal Water Quality: Retrospection, Introspection, Future Directions (Boca Raton (United States): Taylor & Francis (CRC Press)).

Campbell, G. (2013). Remote sensing of algal blooms in inland waters using the matrix inversion method and semiempirical algorithms. In Advances in Mapping from Remote Sensor Imagery, (Boca Raton (United States): Taylor & Francis (CRC Press)), pp. 279–308.

Dekker, A.G., and Hestir, E.L. (2012). Evaluating the Feasibility of Systematic Inland Water Quality Monitoring with Satellite Remote Sensing (CSIRO: Water for a Healthy Country National Research Flagship).

Doxaran, D., Ehn, J., Bélanger, S., Matsuoka, A., Hooker, S., and Babin, M. (2012). Optical characterisation of suspended particles in the Mackenzie River plume (Canadian Arctic Ocean) and implications for ocean colour remote sensing. Biogeosciences 9, 3213–3229.

GEMS/Water (2014). Analytical Methods for Environmental Water Quality (Burlington (Canada): Environment Canada). Kutser, T. 2004. Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnol. Oceanogr., 49, 2179–2189.

Mahadevan, T.N., Saradhi, I.V., Gaware, J.J., and Puranik, V.D. (2003). Assessment of Environmental Quality: Systems Approach and Management. In National Symposium on Biochemical Sciences: Health and Environmental Aspects, S. Prakash, ed. (New Delhi, India: Allied Publishers Pvt. Ltd.), pp. 3–16.

Matthews, M.W. (2011). A current review of empirical procedures of remote sensing in inland and near-coastal transitional waters. Int. J. Remote Sens. 32, 6855–6899.

Mellard, J.P., Yoshiyama, K., Litchman, E., and Klausmeier, C.A. (2011). The vertical distribution of phytoplankton in stratified water columns. J. Theor. Biol. 269, 16–30.

Odermatt, D., Pomati, F., Pitarch, J., Carpenter, J., Kawka, M., Schaepman, M., and Wüest, A. (2012a). MERIS observations of phytoplankton blooms in a stratified eutrophic lake. Remote Sens. Environ. 126, 232–239.

46

Odermatt, D., Gitelson, A., Brando, V.E., and Schaepman, M.E. (2012b). Review of constituent retrieval in optically deep and complex waters from satellite imagery. Remote Sens. Environ. 118, 116–126.

Ostendorf, B. (2011). Overview: Spatial information and indicators for sustainable management of natural resources. Spat. Inf. Indic. Sustain. Manag. Nat. Resour. 11, 97–102.

Patalas, K., and Salki, A. (1993). Spatial variation of crustacean plankton in lakes of different size. Can. J. Fish. Aquat. Sci. 50, 2626–2640.

Piskozub, J., Neumann, T., and Wozniak, L. (2008). Ocean color remote sensing: Choosing the correct depth weighting function. Opt. Express 16, 14683–14688.

Pitarch, J., Odermatt, D., Kawka, M., and Wüest, A. (2014). Retrieval of vertical particle concentration profiles by optical remote sensing: a model study. Opt. Express 22, A947.

Stramska, M., and Stramski, D. (2005). Effects of a nonuniform vertical profile of chlorophyll concentration on remote-sensing reflectance of the ocean. Appl Opt 44, 1735–1747.

Waldmann, C., Tamburri, M., Prien, R., and Fietzek, P. (2010). Assessment of sensor performance. Ocean Sci. 6, 235–245.

Zaneveld, J.R.V., Barnard, A.H., and Boss, E. (2005). Theoretical derivation of the depth average of remotely sensed optical parameters. Opt. Express 13, 9052–9061.

Zielinski, O., Busch, J.A., Cembella, A.D., Daly, K.L., Engelbrektsson, J., Hannides, A.K., and Schmidt, H. (2009). Detecting marine hazardous substances and organisms: sensors for pollutants, toxins, and pathogens. Ocean Sci. 5, 329–349.

47

Chapter 4: Linkages Between Data Providers and End UsersRichard P. Stumpf, Carsten Brockmann, Paul DiGiacomo, Blake Schaeffer, Arnold Dekker.

4.1 Context Multiple stakeholders and user requirements for products derived from remote sensing (Chapter 1) lead to a variety of complexities involved in providing those products. These complexities cover several broad categories. At the highest level is the nature of the data delivery; this covers identifying appropriate products, tools, and training to support the manager’s applications. It is frequently tempting for remote sensing scientists to automatically process products for delivery to a Web page, with the assumption that the availability of the product will solve the manager’s concern. However, if data delivery issues such as format, training, and access are not properly addressed, the data will be unusable or not used. The next key concern involves data quality and consistency. The managers need to have an evaluation of data quality, and information on data consistency and potential failures. Can the managers be confident in all the data; how much work must they do to identify errors or failures? However, remote sensing scientists require a return flow where user communities provide data sets that support validation. The next two concerns involve timeliness characteristics; does the time perspective require retrospective data or current data, and whether a “current” issue requires data in real-time data or if a delay of a season or more is acceptable. These four areas of concern—delivery, quality, temporal context, and timeliness—shape the linkages between data provides and end users.

4.2 Data deliveryData delivery captures a range of complexities, including access methods, data format, metadata requirements, analysis and analysis tools, and training. The fundamental challenge with data delivery is that none of them can be correctly identified without direct interaction with the end users. This interaction must identify the user capabilities, knowledge, and resource commitment. In short, the end users need to extract the maximum information for their requirements with the minimum effort in processing image data.

4.2.1 Data formatsTypically products that require advanced remote sensing analysis tools, such as NASA’s SeaDAS (Sea Data Analysis System) will be beyond the scope of a resource manager. Compatibility with geographic information system (GIS) packages increases the value for some applications, especially habitat mapping. For water quality products, GIS compatibility will

48

depend on the user requirement. Recognition of these distinctions by the data provider will lead to the greatest value of the product. For map products, GIS compatibility at some level should be maintained. Graphics files, such as GIF, PNG, and JPG, are common but they are the least useful to a user, as they do not internally store geographic information (geography must be stored in a separate “world” file) and do not typically preserve data values. For small lakes this is not necessarily a problem; if users can readily identify the lake, they can use color to qualitatively assess the pattern. For larger or unfamiliar water bodies, or for habitats, the user has to guess locations from these products. This is helpful as a way of distributing satellite data. [SB1] There are better options appearing today that allow at least the extraction of latitude and longitude (Table 1). Websites can be constructed to allow extraction of latitude and longitude from pictures with no knowledge of a program (NASA WorldView; NASA 2015). More advanced options exist within programs. For the general user, the standard PDF Acrobat reader supports extraction of spatial information, so geoPDFs can provide a simple usable information set (Figure 1, Lake Erie geoPDF). Vector data values can be imbedded in the PDF, so more information can be extracted. Online mapping tools, such as GoogleMaps or GoogleEarth, allow similar extraction of geo graphic and vector (or graphical) information. At a next level, GeoTIF files provide pictorial information as well as being geographic data files. They can maintain both data values and the color information in “8-bit” (256 value) or 24-bit (true color) files. This provides a dual use—an immediate view of the information, while allowing data extraction in a GIS program.

Table 1.

Product Pictorial Geographic data

GIF, PNG, JPG Yes Only with separate world file

No

PDF Yes Yes as geoPDF Yes, vector

geoTIFF Yes (as 8-bit or 24-bit) Yes Yes

Data files (HDF, netCDF, etc.)

Not at present Yes Yes

49

Figure 1. GeoPDF file of Lake Erie displayed in Acrobat Reader. Location of cursor is shown.

Habitat mapping requires GIS formats, as most questions revolve around the relationship between the habitat and other land use patterns, or spatial changes in habitat type. Water quality applications have a more variable requirement. Monitoring of algal blooms or sediment plumes in large lakes or bays may need the image information to capture the spatial patterns. In small water bodies, data extraction may become more useful. A manager addressing a specific water body may want to obtain temporal patterns, is the bloom increasing over the month, or has the turbidity spiked. In this case data extraction tools would be potentially more useful. Metadata for files will need to capture the key points: what the values mean (whether scaled integers or real numbers), units, algorithms, applied masks or flags, as well as information on quality control.

4.2.2 Data extraction and analysis tools.Data extraction tools will develop rapidly over the next few years. Among the options for most managers are ArcGIS based, smartphone app-based, and Web-based. ArcGIS based extraction tools GIS data, of course, would allow a GIS lab in the management office to extract the time series of interest, and adjust the methods and results. The incorporation of Python as a scripting tool, enhances this option when the tools are written in python. Web-based options exist, where data can be extracted using a Web site. One of the best examples at present is

50

NASA’s Giovanni site (NASA, 2015b). It requires nominal knowledge and can deliver data from a rectangle to a form compatible with a spreadsheet. Some sites automatically create time series from fixed locations. With the rise of cloud storage and computing, web-based options of user selection will become more frequent and flexible. The development of smartphone apps may push the capability much further along, and rapidly enhance get us to this type of approach (Figure C, Sidebar; US EPA app), particularly as app capabilities are being integrated into computer operating systems.

Figure C. Example of the ability to drop pin locations in water bodies of ineterst. The pin extracts satellite derived data from the specific lat/long coordinates. The application can also display geotiff files of satellite derived products (Mattas-Curry, Schaeffer, Conmy, & Keith. 2015. A satellite perspective to monitor water quality using your mobile phone. Lake Line. 35(2): 28-31.)

4.3 Training Training of managers will involve understanding the options for data acquisition and data extraction, and also understanding the data quality and data limitations. Training is not a substitute for robust algorithms and proper quality control. This may seem obvious, but it is not. Providing a standard ocean chlorophyll algorithm to estuary managers, then “training” the managers to ignore the “high chlorophyll” when the water is optically shallow, or turbid, or high in CDOM, will not lead to use of the data product, and is a waste of time for both the producer and the user. Training strategies are rarely presented in the literature (any examples of training in the literature XXX?), so producers often start from scratch with little knowledge of what has been learned by other data producers.

51

The training needs to cover the quality control, accuracy, and uncertainties. Also, as no remote sensing model is 100% reliable, the managers need to know what to look for in order to exclude the small percentage of suspect data. Training may lead to interesting feedback. A producer may automatically disregard image portions that are suspect, while a user will assume that they are valid. Noting this discrepancy in quality can lead to review and improvement of the algorithms. (note from Steve:There are also likely different levels of comfort for the data between scientists and managers. Scientists may want most accurate and robust model, managers may not need this level of detail. See the IOCS meeting report http://iocs.ioccg.org/wp-content/uploads/2012/10/report-iocs-2015-meeting.pdf)

4.4 Quality ControlQuality control is inconsistent across the remote sensing areas. Habitat mapping has a well-established procedure for skill assessment going back decades (Congalton 1999; Foody, 2002). This includes user and producer errors across habitat classes, and methods for change evaluation. Quality control strategy for water quality is much thinner. The emphasis tends to be on algorithm accuracy, following the methods for the global chlorophyll algorithm, while reproducibility, robustness and image accuracy are not frequently considered. Algorithm accuracy is Table 2 has examples of factors that should be addressed. Simple regression metrics do not address algorithm robustness. Spatial and temporal patchiness (Kutser, 2009) introduce artifacts that confound data misfit with data error (Lynch and McGillicuddy, 2009) For some topics, such as bloom detection, type 1 and type 2 errors are of particular concern to managers (Tomlinson et al., 2004).

Image quality control Parameter quality control

Consistency within the image (no false contours)

Accuracy over the data range

Robustness (consistency between images having different atmosphere, look angles etc.)

Reproducibility under varying water quality conditions

Invalid data failure rate Algorithm failure rate

4.5 Validation Data

52

Validation data may originate in many sources. For habitat mapping projects, the producers may collect much of the validation data. In most cases, however, the user communities may have programs that collect the necessary data, or may enhance those programs if they perceive the remote sensing product to be of value. The major challenges with obtaining validation are access, inconsistent formatting, and variable methods. These problems can be solved for programs that are coupled with remote sensing (Table 3). However, for other data sets, the remote sensing data producer faces some significant challenges. The data collected in a standardized program may not be optimal for the remote sensing validation. However, that is the data set that is available. One program may systematically measure chlorophyll by laboratory fluorometry, uncorrected for phaeopigments, another may use HLPC. One program may use cell counts for HABs, another will use biovolume. These variations mean that a single validation strategy will not work. Classification metrics may be necessary, and relative concentrations may become part of the validation. The best way to obtain validation data is through a monitoring program that fully couples both field and remote sensing data (Table 3), and exists for several HAB issues. A coupled program assures that the field data is available, usable, and directly compatible with the remote sensing data. The next best option is drawing on a systematic measurement program. The data set is self-consistent and the management community established format and access protocols. This is most effective in water quality programs, however, HAB monitoring in the Gulf of Mexico also fits this description (Stumpf et al., 2003). Intensive programs, such as the US National Lake Assessment, have the consistency and access of other systematic programs, however, they are focused in time, and not repeated each year.

Ad hoc programs—those that are conducted in response to an event or as part of an independent research project—may provide key validation data, but they have significant challenges. The data may be poorly or inconsistently formatted, including simple issues like the use of station names, without longitude and latitude. For the remote sensing producer, the key challenge is to convince the data provider to address these issues. One effective way to solve the problem is to fund the field data provider to address the data deficiencies. This is often not possible. Another quite effective method is to demonstrate that the remote sensing product has real value to the field data provider. A data provider will invest time in providing data in exchange for a product they can use. This demonstration requires real effort on the part of the remote sensing data producer. The data producer needs to understand the management need and have already conducted quality control and some evaluation. The result may be that the effort is not worth the return. (note from Steve - perhaps this can be shortened. The topic seems to drift off a bit)

Validation program examples Typical parameters issues

Integrated satellite Finland, Norway, HABs Establishing the

53

and field Lake Erie program

Systematic national and regional field collection

EU Water Directive, GBRMPA, US EPA Estuary programs (Chesapeake Bay, San Francisco Bay, Tampa Bay)

Chlorophyll, turbidity, HABs

limited areas,

National intensive programs

US National Lake Assessment

Chlorophyll, nutrients, oxygen, taxonomy, etc

Limited time, parameter

Routine local programs

States, municipalities, parks

HABs, chlorophyll, turbidity

Format, parameters,

Ad hoc: research, event response

Universities, various government

various Format, parameters, access

Table 3. Types of Validation data

4.6 Timeliness

4.6.1 Real-time Monitoring and Response. For users that potentially need to respond to events, products need to be delivered in a timely and useful way. Health agencies would want to know whether potentially toxic blooms are present. Water suppliers need to deal with toxicity and biomass. Environmental agencies need to anticipate aquatic life and wildlife impacts, clean-ups of fish kills, and public reports of unusual water conditions. Harmful algal blooms are an obvious application of real-time monitoring, and several examples exist (Florida, Stumpf et al., 2003; Lake Erie, Wynne et al., 2013; the Finnish Baltic, Hansson and Hakansson, 2007; Norway, XXXX; Ireland, Cusack XXXX). Monitoring and response place a burden on the data producer. The weather forecast community refers to models as “model guidance”, not as forecasts. The forecasts are generated from the combination of models, data, and analysis. If the models are distributed as the product, it leaves to the end user the job of creating a forecast. A similar view is needed for satellite data. When supporting a response, the satellite data set is a model, leaving the question of how the forecast (hazard) is determined. Automatic distribution of satellite data is

54

the equivalent to distributing a a weather model. The user has been provided the satellite guidance, but not the forecast. This can be appropriate, but requires the user to become expert at interpreting the satellite data, understanding product idiosyncrasies, and generating the appropriate forecast. There are alternatives. Using the weather analogy, the U.S. National Weather Service produces a National Digital Forecast Databased (NDFD), where the actual forecasts (made by weather forecasts) are distributed in a compatible format that is readable as a model or dataset. In the realm of satellite data, the Florida HAB bulletin (Stumpf et al., 2003) uses analysts to incorporate previous knowledge to identify features in the satellite image that are known to be the HAB. The value of such a product means that false positives or false negatives are not repeated for days on end when the algorithm systematically fails, or random failures do not translate into the product when anomalous conditions occur (such as resuspension events). Similar strategies can be developed by coupling satellite data with hydrodynamic or ecological models (Wynne et al., 2013; XXXX Baltic or GBR?). The value of the satellite data increases as a result of the additional analysis.

4.6.2 Current AssessmentCurrent assessment involves identifying the relative health of the water bodies or habitats at present. Assessment and monitoring (section 4.5) can cover a continuum, and repeating assessment over multiple years lead to retrospective analysis (4.4.2). Assessment provides the status of the system, while monitoring would typically support an immediate management response. The quality control for assessment will be more stringent than real-time applications. Assessments will require algorithms that are valid across water bodies for the variable of interest. Assessments have been applied with a variety of satellite sensors. Lakes have been examined every few years with Landsat (Minnesota, known for having 10,000 lakes, Olmanson et al., 2008), and the Great Barrier Reef (Roelfsema et al., 2006). Habitat studies, such as extent of seagrass or coral health are a typical example of current assessment.

4.6.3 Retrospective AssessmentRetrospective assessment involves evaluation of trends in water quality and whether water bodies have changed. These have particular value in developing scenarios to that might support forecasts or allow development of management strategies. Some examples of this sort of work are studies of cyanobacteria blooms in the Baltic (Kahru and Elmgren, 2014) and in Lake Erie (Stumpf et al., 2012). In these cases, evaluation of the inter-annual variability allowed determination of the role of nutrient loading. The study by Stumpf et al. (2012) helped inform development of nutrient target loads by the U.S. and Canada through the Great Lakes Water Quality Agreement. Retrospective studies also capture the phenology of blooms (Hungary, Palmer et al., 2015), and patterns between lakes (South Africa, Matthews 2014) and the frequency of events that pose a risk for various uses, such as the potential risk of toxic cyanobacteria to drinking water suppliers (Lake Erie, Wynne and Stumpf, 2015). They can also identify when conditions changed prior to development of monitoring programs. For example, researchers in the late 1980s discovered dying seagrass in Florida Bay that appeared to lead to an increase in turbidity. A retrospective analysis determined the scale of seagrass loss and

55

timing of development of turbidity before a field monitoring program was implemented (Stumpf et al., 1999). Retrospective assessments have a particular challenge in the difficulty of evaluating the data products. The monitoring programs that support algorithm development typically post-date the available satellite data (especially with Landsat). As a result, the quality control requirements (section 4.2) should become both rigorous and flexible for these studies. This appears to be an oxymoron. The rigor is in assessing the robustness of the algorithm under the potential variation in conditions that may have occurred. The flexibility is in the types of observations used to provide confidence in the retrospective conditions. For example, if turbidity changed, a chlorophyll algorithm that may be valid in clear water may perform poorly during periods of turbid water. This means that the algorithm must be evaluated over the range of turbidity and chlorophyll that was observed over the time of interest, that increases the rigor of the quality control. For flexibility, qualitative validation methods become necessary. In the Florida Bay case mentioned above, seagrass researchers reported seeing rapid die-offs of seagrass starting in a narrow time window (late 1989); Stumpf et al (1999) evaluated the satellite data to confirm that it identified a change in the appropriate area during this narrow period.

4.7 Future Collaborative EffortsSeveral routine management applications around the world use satellite data products. Each of these has developed skills and knowledge that can help advance the application of satellite data in other areas. The insights and understandings developed by the collaboration between data providers and managers need to be shared. Frequently, the remote sensing community works with managers (or research users) that are extremely knowledgeable about remote sensing. However, the key user communities should not require a depth of knowledge in remote sensing. They need to understand the satellite product as well as they understand a chlorophyll measurement, but not to the same level as being able to manage the nuances of HPLC, or atmospheric correction. This means that taking what is learned from supporting different non-expert user needs and exchanging that information between countries. Note from Steve: Might consider that Schaeffer et al 2013 highlighted “To best realize the full application potential of these emerging technologies an open and effective dialogue is needed between scientists, policy makers, environmentalmanagers, and stakeholders at the federal, state, and local levels.” We also found that after 22+ years the same issues identified by Specter and Gayle (1990) regarding training and understanding still stand today. Schaeffer, Schaeffer, Keith, Lunetta, Conmy, & Gould. 2013. Barriers to adopting satellite remote sensing for water quality management. International Journal of Remote Sensing. 34(21): 7534-7544.

56

Specter, C., and D. Gayle. 1990. “Managing Technology Transfer for Coastal Zone Development:Caribbean Experts Identify Major Issues.” International Journal of Remote Sensing 11:1729–1740.

Based on these issues, we need international strategies to:

1 use assessments within existing monitoring assessment systems to identify algorithm strengths and deficiencies, and gaps in remote sensing. International programs, such as GEO, Blue Planet, the European Union Directives, need to make it easier to share this information;2. have international programs and journal editors promote transfer and comparison of algorithms between regions in order to evaluate the robustness of methods; the results will identify strengths, improve quality control, transfer knowledge and identify gaps that can inform future research;3. have international programs organize workshops to help share successes and “lessons learned” between different communities in sharing data with users;4. develop a common resource on strategies for supporting managers; journal editors need to be encouraged to help promote this exchange.

References Congalton R, Green K (1999) Assessing the accuracy of remotely sensed data: Principles and practices. Lewis Publishers, New York, New York Dekker GBR. Foody GM (2002) Status of land cover classification accuracyassessment. Remote Sens Environ 80:185‒201 Hansson, M., & B. Hakansson, 2007, "The Baltic Algae Watch System - a remote sensing application for monitoring cyanobacterial blooms in the Baltic Sea", Journal of Applied Remote Sensing 2007, 1(1):011507. Kahru, M, and R Elmgren. 2014. Multidecadal time series of satellite-detected accumulations of cyanobacteria in the Baltic Sea. Biogeosciences 11.13 (2014): 3619-3633. Kutser, T. 2009. Passive optical remote sensing of cyanobacteria and other intense phytoplankton blooms in coastal and inland waters. International Journal of Remote Sensing. 30:4401-4425. Lynch, D. R., McGillicuddy, D. J., & Werner, F. E. (2009). Skill assessment for coupled biological/physical models of marine systems. Journal of Marine Systems, 76(1), 1-3

57

Lunetta, R.S., B.A. Schaeffer, R.P. Stumpf, et al., 2015. Evaluation of cyanobacteria call count detection derived from MERIS imagery across the eastern USA. Remote Sensing of Environment. V.157, pp.24-34. DOI: 10.1016/j.rse.2014.06.008 Matthews, M. W. (2014). Eutrophication and cyanobacterial blooms in South African inland waters: 10years of MERIS observations. Remote Sensing of Environment, 155, 161-177.NASA, 2015a. EOSDIS Worldview. https://earthdata.nasa.gov/labs/worldview/. Accessed 01 Sept 2015. NASA, 2015b. Giovanni. http://giovanni.gsfc.nasa.gov/giovanni/. Accessed 01 Sept 2015 NERC, 2015. Globolakes. http://www.globolakes.ac.uk/. National Environment Research Council. Accessed August 2015. Olmanson, L. G., Bauer, M. E., & Brezonik, P. L. (2008). A 20-year Landsat water clarity census of Minnesota's 10,000 lakes. Remote Sensing of Environment, 112(11), 4086-409 Palmer, S. C. J., Odermatt, D., Hunter, P. D., Brockmann, C., Presing, M., Balzter, H., & Tóth, V. R. (2015). Satellite remote sensing of phytoplankton phenology in Lake Balaton using 10years of MERIS observations. Remote Sensing of Environment, 158, 441-452. Roelfsema CM, Phinn SR, Dennison WC, Dekker AG, Brando VE (2006) Monitoring Toxic Cyano‒bacteria Lyngbya majuscula in Moreton Bay, Australia by Integrating Satellite Image Data and Field Mapping. Harmful Algae 5:45‒56 Schaeffer, B.A., J.D. Hagy, R.N. Conmy, J.C. Lehrter, R.P. Stumpf, 2011. An approach to developing numerical water quality criteria for coastal waters using the SeaWiFS satellite data record. Environmental Science & Technology, dx.doi.org/10.1021/es2014105 Stumpf, R.P. , M.E. Culver, P.A. Tester, G.J. Kirkpatrick, B.A. Pederson, M. Tomlinson, E. Truby, V. Ransibrahmanakul, K. Hughes, M. Soracco. 2003. Monitoring Karenia brevis blooms in the Gulf of Mexico Using Satellite Ocean Color Imagery and Other Data. Harmful Algae, v. 2(2), pp. 147-160. Stumpf RP, Frayer ML, Durako MJ, Brock JC (1999) Variations in water clarity and bottom albedo in Florida Bay from 1985 to 1997. Estuaries 22:431‒444 Stumpf, R.P., T.T. Wynne, D.B. Baker, and G.L. Fahnenstiel. 2012. Interannual variability of cyanobacterial blooms in Lake Erie. PLoS ONE. . 7(8): e42444. doi:10.1371/journal.pone.0042444.

58

Tomlinson, M.C., R.P. Stumpf, T.T. Wynne. 2009, An evaluation of remote sensing techniques for enhanced detection of the toxic dinoflagellate, Karenia brevis, in the Gulf of Mexico, Remote Sensing of Environment, v. 113, pp. 598-609. Wynne, T.T. and R.P. Stumpf. 2015. Spatial and temporal patterns in the seasonal distribution of toxic cyanobacteria in western Lake Erie from 2002-2014. Toxins.7:1649-1663. Wynne, T.T., R.P. Stumpf, M.C. Tomlinson, G.L. Fahnenstiel, J. Dyble, D.J. Schwab, S. Joseph-Joshi, 2013. Evolution of a cyanobacterial bloom forecast system in western Lake Erie: development and initial evaluation. Journal of Great Lakes Research. http://dx.doi.org/10.1016/j.jglr.2012.10.003

59

Chapter 5: Measures from satellites Mark Matthews, Stewart Bernard, Lisl Robertson, Daniel Odermatt, Derek Griffiths , Tiit Kutser

List of terms & symbols

6S Second Simulation of a Satellite Signal in the Solar Spectrum

aCDOM Absorption coefficient of CDOM

ALI Advanced land imager

AOT Aerosol optical thickness

ap(443) Absoriton coefficient of particulate matter at 442 nm

ARPH Adaptive reflectance peak height

BRR Bottom of Rayleigh reflectance

CDOM Coloured dissolved organic matter

Chl Chlorophyll-a

FAI Floating algae index

FAV floating aquatic vegetation

FLH Fluorescence line height

HICO Hyperspectral imager for the coastal ocean

IOPs Inherent optical properties

MCI Maximum chlorophyll index

MERIS Medium resolution imaging spectrometer

MODIS Moderate resolution imaging spectrometer

MPH Maximum peak height

MSI Multispectral imager

OLCI Ocean and land colour instrument

60

Rrs Remote sensing reflectance

SAV submerged aquatic vegetation

SPM Suspended particulate matter

TOA Top of Atmosphere

Tripton Non-living suspended matter

TSM Total suspended matter

5.1 Understanding the satellite signal measured from inland and coastal waters The third report in the IOCCG series (IOCCG 2000) addressed remote sensing in coastal and other optically-complex waters. In this report, an introduction was presented on the colour of case 2 waters, defined as those waters where the water-leaving signal is significantly affected by constituents other than phytoplankton, namely tripton composed of detritus and minerals, and coloured dissolved organic matter (CDOM) consisting of humic and fulvic acids (the reader is referred to Chapter 2 in IOCCG (2000) for further details). General challenges for optically complex waters are extensively addressed in IOCCG (2000). More recently however, a significant body of literature has developed related to remote sensing of spatially constrained inland waters, which sets apart particular challenges that do not equally apply for coastal waters. In addition, the development of new sensors and concomitant algorithms and approaches means that significant advances have been made since 2000. Understanding the signal measured at the top of the atmosphere (TOA) by a satellite in space begins with radiative transfer in water and the atmosphere. As described in Chapter 2, the signal leaving the water arises from several sources including the water itself, the water constituents (namely phytoplankton, tripton and CDOM), benthic or submerged vegetation (if present), and the lake bottom (if optically shallow). The signal from the atmosphere is affected by atmospheric gases (e.g O2, H2O) and aerosols (e.g. dust and smoke particles). In addition, the atmospheric signal may contain large components resulting from stray light scattered into the instrument field of view from adjacent land (called the adjacency effect which depends on the target size and the optical properties of the surrounding vegetation or soil), cloud (if present) and light reflected from the water surface (sun glint or specular reflectance, if present). The unique challenge faced in small inland water bodies results from processes occurring both in the water and in the atmosphere. This includes sometimes extreme ranges in concentration of independently varying water constituents; the presence of blooms of cyanobacteria driven by

61

nutrient concentrations that are extremely high when compared with coastal and open ocean waters; bottom signals in shallow water and benthic, submerged or floating vegetation; unpredictable and extreme concentrations of atmospheric aerosols resulting from, e.g., urban pollution and fires; high levels of stray light from adjacent land and vegetation; highly variable optical properties of inorganic matter dependent on the surrounding geological formations and landscape, e.g., sediment features. In combination, these present a substantial challenge for obtaining quantitative estimates of biogeophysical variables in inland waters, and demand methods which account for some of these challenges simultaneously. The following section discusses the variability and ranges of the water-leaving and atmospheric signals typically encountered over inland waters, as measured by satellites in space at top-of-atmosphere.

5.1.1 The water-leaving signalThe water constituents in optically complex water are generally summarized as belonging to phytoplankton, tripton and CDOM. However, characterising the water-leaving signal requires an understanding of the considerable optical complexity within these groups, and having a quantitative ability to assess their cumulative effects on the underwater light field. Figure 1 provides a quantitative illustration of the complexity and variability of the remote sensing reflectance routinely observed in inland waters, arising from the optical complexity and range of concentrations of in-water components, which are discussed below. The phytoplankton component consists of groups with highly variable optical properties from eukaryotes (algae) and prokaryotes (cyanobacteria). This

62

Figure 1. Examples of measured remote sensing reflectance illustrating variability in reflectance over five orders of magnitude often encountered in inland and near-coastal waters. Note log scale on y-axis. See legend for water constituent concentrations. Figure: Mark Matthews. Data courtesy of: David Doxaran; Claudia Giardino; Erin Hestir; Tiit Kutser; Mark Matthews; Stefan Simis.

variability results from differences in pigmentation, spectral fluorescence emission, population particle size distribution, and intra-cellular structural features. For example, cyanobacteria have significantly different optical properties than algae caused by significant scattering from internal gas vacuoles (Matthews & Bernard, 2015), a lack of characteristic chlorophyll-a fluorescence emission (Seppälä et al., 2007), and phycobilipigments which alter their spectral features. Added to this complexity arising from optical features is the very large range of productive biomass that occurs in inland waters. Chlorophyll-a concentration routinely varies over five orders of magnitude from 0.1 to 1 000 mg m-3, from alpine lakes to subtropical reservoirs. The

63

large concentration range, and optical variability for which chlorophyll–a algorithms are expected to perform, presents a substantially different set of challenges for inland remote sensing in comparison to the oceanic environment. The tripton component composed of detritus and minerals also varies significantly in optical characteristics and concentration ranges. Detritus material composed of dead phytoplankton cells and decaying organic matter is often stirred into the water column in very high concentrations in shallow eutrophic lakes (e.g. Lake Ijsselmeer in the Netherlands, or Zeekoevlei in South Africa). This significantly reduces the availability of light in the blue spectral region. Mineral particles composed of silt and clay depend on the surrounding geology, and the variability in the particle shapes and sizes, and absorption characteristics, results in very wide ranging optical properties (e.g Stramski et al., 2007). Suspended clay particles (usually measured as total suspended matter) exhibit sometimes extremely high concentrations ranges greater than 2000 g m-3 (Doxaran et al., 2006). The result is bright water-leaving signals, including for wavelengths greater than 700 nm. Such high concentrations cause significant interference with conventional retrieval algorithms for chlorophyll-a based on reflectance ratios or fluorescence (e.g. Kobayashi et al. 2011; McKee et al., 2007). High concentrations of humic and fulvic acids (CDOM) leached from surrounding vegetation and soils significantly reduces the light availability in the blue spectral region in boreal and brackish waters. This is in addition to detrital CDOM produced by the phytoplankton themselves. Dark waters have absorption values that may exceed 10 m-1 at 440 nm (Kutser et al., 2016), with typical values for inland water between 1 and 2 m-1 at 440nm (Kirk, 1994). In such cases water-leaving signal may be so small as to make detection at TOA very challenging or even infeasible, taking into account the signal-to-noise ratios of current satellite instruments. The frequent presence of submerged or floating aquatic vegetation (SAV or FAV) also further complicates the signal in inland waters, particularly in shallow marginal systems. FAVs negate the influence of absorption and scattering by water resulting in spectra resembling land or vegetation. Various types of SAV cause significant alterations in the shape and magnitude of the water-leaving spectra, such that these might be discriminated from other vegetation types (Hestir et al., 2008). Such composition, concentration ranges and variability in water constituents described above are foreign to ocean colour applications further than a few kilometers from the coastline. This demands new and improved approaches using instruments that are capable of accurately measuring the variability of the water-leaving signal across five orders of radiance magnitude. Using a radiative transfer model, the expected range of water-leaving signals resulting from variable phytoplankton types and concentrations, CDOM absorption and non-algal particle concentrations were modeled. This provides a preliminary assessment of the range of variability in remote sensing reflectance expected to be caused by the three primary determinants of the water-leaving signal typically retrieved by algorithms. The model used three phytoplankton types commonly occurring in inland waters: dinoflagellates, cryptophytes and vaculoate cyanobacteria

64

(Matthews & Bernard, 2015; Evers-King et al., 2014; Robertson Lain et al., 2014) at chlorophyll-a concentrations between 0.1 to 2.5 mg m-3, and 20 to 60 mg m-3. For each chlorophyll-a concentration, the reflectance was modeled for low/high CDOM absorption, and low/high non-algal particle (NAP) concentrations scenarios. Figure 2 shows the Rrs and Figure 3 shows the corresponding frequency plots for the 1st and 2nd derivatives in order to highlight the significant peaks, troughs and inflection points in the Rrs. It is apparent from Figure 2 and 3 that accounting for variability only due to three phytoplankton types and two CDOM and NAP ranges, large complexity is introduced in the water-leaving signal, evidenced by significant changes in shape and magnitude. For the high biomass range the red peak near 700 nm becomes an important peak feature, highlighting the use of red wavelengths for high biomass scenarios.

Figure 2. Modelled remote sensing reflectance for three phytoplankton groups (dinoflagellate, cryptophyte, cyanobacteria) for two biomass ranges and combinations of high/low CDOM and NAP scenarios. Left panel: low concentration ranges 0.1 to 2.5 mg m-3. Right panel: high concentration ranges 20 to 60 mg m-3. The spectra were modelled at 2 nm resolution using Ecolight (Sequoia Scientific). Credit: Lisl Robertson Lain.

65

Figure 3. Corresponding frequency plots for Figure 2 for the 1st (peaks/troughs) and 2nd derivatives (inflection points) overlaid on the OLCI wavebands (grey shading). Top: low concentration range 0.1 to 2.5 mg m-3. Bottom: high concentration range 20 to 60 mg m-3. Credit: List Robertson Lain.

5.1.2 The atmospheric signal

The atmosphere typically contributes around 90% of the signal at TOA over a clear ocean. Thus the signal from the water is challengingly small in comparison to the atmosphere. In inland waters, the water-leaving signal may be substantially larger (in the case of high concentrations of phytoplankton or suspended matter) or smaller (in the case of high CDOM) than that observed in the open ocean. This makes correcting for the atmosphere in a consistent manner in inland waters substantially more challenging, potentially requiring the use of different atmospheric correction schemes dependent on conditions. The main challenges for atmospheric correction over small to intermediate inland targets arise from contamination by atmospheric aerosols and stray light, as discussed below and illustrated in Figure 4.

66

Figure 4. Effect of the atmosphere on the signal measured at TOA by a satellite sensor for low and high biomass scenarios. TOA radiances were modeled using MODTRAN for clear and turbid atmospheres, with and without the adjacency effect. Water leaving reflectance was generated with Hydrolight at 1 nm resolution. A, B show TOA radiance, C, D show contribution of water leaving radiance (Lw) to TOA radiance (Ltot). Simulations: Mark Matthews (Ecolight) and Derek Griffiths (Modtran).

Concentrations of aerosols over land are often more extreme than over the ocean, resulting from smoke and dust particles from industrial, agricultural and urban activities. This may lead to a substantially higher contribution by aerosol scattering to the total signal at TOA and significant dampening of the available signal (Figure 4). This has the effect of reducing the relative signal from water, especially over oligotrophic waters (Figure 4C). Typical methods used to retrieve aerosol optical thickness estimates from NIR satellite reflectance measurements are often invalid due to non-negligible water-leaving reflectance from high concentrations of in-water scattering constituents. Correcting for the atmosphere in such cases usually requires detailed

67

atmospheric profiles and/or simultaneous AOT measurements from AERONET stations, not conventionally available for most targets. Therefore, alternative methods are required to obtain aerosol estimates over inland waters. The adjacency effect causes enlarged reflectance at NIR wavelengths, and is particularly severe for small dark water bodies surrounded by dense vegetation (Figure 4). The adjacency effect may be significant up to 30 km from the shoreline, depending on the surrounding vegetation and shape of the target (Santer and Schmechtig, 2000). The effect complicates the estimation of aerosol properties from NIR reflectance. The adjacency effect over a clear alpine lake may contribute more than 30% of the signal at TOA (Odermatt et al., 2008). This makes oligotrophic water bodies the most challenging targets from a quantitative perspective, due to the small water-leaving signal with a relatively low dynamic range but high stray light contamination. Various approaches to correct for the adjacency effect have been proposed (e.g., Santer and Gazolski, 2008; Sterckx et al. 2014; Kiselev et al. 2014). In most cases, the implementation of a correction for the adjacency effect is required in order to derive accurate water-leaving reflectance estimates over inland waters; however, the performance of correction methods needs careful assessment, especially when used operationally. Due to the commonly large concentration of phytoplankton or mineral particles in inland waters, the contribution to the signal at TOA from water may be substantially higher than 10%. For these ‘bright’ waters the relative signal from the water may be comparable to that of the atmosphere (Figure 4). This provides an opportunity to take advantage of the resulting increase in SNR at TOA. It also means that various approaches using TOA data are feasible, reducing the need for an atmospheric correction for the full aerosol component (see Figure 5 below).

5.2 Deriving water quality products from satellite: algorithms and issuesGiven the complexity present in the water and atmosphere briefly discussed above, pragmatic approaches are required with respect to quantitative retrievals of water quality parameters suitable for general application in inland waters. In the absence of a satisfactorily complete knowledge of water and atmospheric constituent optical properties, the requirements remain unmet in order to satisfy a fully reductionist bio-optical modeling approach leading to quantitative water quality estimates. Here we discuss applications of recent algorithms suitable for use with both intermediate TOA and atmospherically corrected bottom-of-atmosphere reflectance products over inland waters. Candidate atmospheric corrections applicable in various scenarios are discussed. Algorithms and approaches are illustrated for the typical parameters of interest in inland waters namely chlorophyll-a, cyanobacteria, FAV, CDOM and total suspended matter.

5.2.1 Algorithm types: partial atmospheric corrections

68

A growing number of publications present approaches which utilise partial atmospheric corrections and resulting intermediate reflectance products to obtain various quantitative and diagnostic products for inland waters. These approaches often take advantage of the relatively higher signal at TOA caused by high constituent concentrations or bright surface features. By correcting for the more predictable and stable Rayleigh (non-aerosol) component of gaseous absorption and scattering, the reflectance at bottom-of-Rayleigh (BRR) can serve as a useful intermediate product. Partial atmospheric corrections are most often used in combination with empirical band ratios and peak height (derivative) subtraction methods to effectively normalize for aerosol effects. As illustrated in Figure 5A, TOA and BRR data types preserve both the magnitude and shape of Rrs in the case of high biomass blooms, especially for red wavelengths. Thus the need to correct for aerosols in the case of bright waters is negligible when implementing empirical-type algorithms. Indeed, in these cases application of unsupervised atmospheric corrections to retrieve Rrs often leads to warping of spectral shapes and errors due to out-of-range concentrations associated with underlying bio-optical models. This is especially true for coupled water-atmosphere models, such as the Coast Colour Level 2 Reflectance (CCL2R) Neural Network. If desired, improved Rrs estimates are provided through spatial interpolation methods such as SCAPE-M (Guanter et al., 2009) for bright waters. In the case of oligotrophic and dark waters, partial atmospheric corrections, as expected, do not often adequately reproduce the shape or magnitude of Rrs (Figure 5B,C). The high values for BRR are the result of correcting for absorption by atmospheric gases. Nevertheless, some studies demonstrate the value of partial atmospheric corrections for simple trophic status estimation even in oligotrophic waters (e.g. Matthews & Odermatt, 2015), when the requirements for inland waters are focused on water quality rather than biogeochemistry. Various products types which may be derived from TOA data types are shown in Figure 6. Chlorophyll-a fluorescence features are clearly discernable enabling quantitative chl-a estimates using algorithm like the FLH (Gower et al., 1999) and MPH (Matthews et al., 2012) from MODIS or MERIS. The spectral features associated with cyanobacteria are also preserved at TOA, enabling discrimination of cyanobacteria using feature detection algorithms such as the CI (Lunetta et al., 2014) and MPH. Red-edge features in bands near 700 nm associated with algal blooms enable quantitative chlorophyll-a estimates to be obtained using indices such as the MCI (Palmer et al., 2014), MPH, and ARPH (Ryan et al., 2014) algorithms from MERIS or hyperspectral type sensors, e.g., HICO, with suitable red-edge bands. Detection of FAV can also be performed from TOA using indices such as the FAI (MODIS) (Hu et al., 2010) or the MPH (MERIS). The TOA or partial atmospheric correction approach can therefore be very valuable and relatively simply implemented for monitoring phenomena such as eutrophication (chlorophyll-a), cyanobacteria blooms and aquatic vegetation in inland waters. However, as discussed in the next section, it is less well-suited to applications requiring blue wavelengths or the full visible water-leaving reflectance spectra required for CDOM determination or analytical reflectance inversion algorithms.

69

Figure 5. Remote sensing reflectance measurements compared to TOA reflectance, partially corrected bottom-of-Rayleigh reflectance (BRR) and Rrs obtained using spatial interpolation of aerosols (Scape-M) and a coupled water-atmosphere Neural Network (CoastColour) from MERIS for a) high biomass – Albufera b) oligotrophic – Lake Constance and c) dark waters – Päijänne. The left panels show dimensionless reflectance, whereas the right panel shows data normalized to the integral to indicate the retrieval of spectral shape. Credit: Daniel Odermatt. In situ Rrs courtesy: Antonio Ruiz-Verdù (Centre for Hydrographic Studies CEDEX), Thomas Heege (EOMAP GmbH & Co.KG), Sampsa Koponen, Kari Kallio, Timo Pyhälahti (SYKE).

70

Figure 6. Various approaches making use of TOA data for the detection of chl-a fluorescence FLH, MPH, ARPH algorithms (A,B,G); cyanobacteria spectral features CI, MPH algorithms (C,D); red-edge bloom features (C, E); and floating algae and vegetation (C,E,F). NB- must be inland examples!

A. FLH dragon bloom (MODIS)? B. MCI

C. MPH cyano detection D. CI cyano detection - Michigan

E. ARPH (HICO) F. FAI veg detection (MODIS!) - Taihu

5.2.2 Algorithm types: full atmospheric correction Given the challenges with aerosol retrieval over inland waters, further correcting the TOA signal for the aerosol component may lead to large uncertainties in the water-leaving reflectance product. From a signal perspective, it is likely only necessary to correct for the aerosol component in scenarios where the water-leaving reflectance constitutes less than approx. 10% of the signal at TOA, typical of ocean reflectance. However, estimates of Rrs can be obtained using radiative transfer codes such as 6S (Giardino et al., 2014) or by interpolating aerosol estimates made over land (e.g. SCAPE-M, Guanter et al., 1999) even for bright targets (Figure 5A).

Methods to derive Rrs include coupled water-atmosphere models that allow for non-negligible NIR reflectance. However, these models are usually unable to account for the extreme range of constituent concentrations and variability of the IOPs of inland waters. In oligotrophic waters coupled neural network approaches more adequately reproduce spectral shapes and magnitudes (Figure 5C). In strongly absorbing waters with very small water-leaving signal, these approaches may also be more useful due to the flexibility of IOP variability afforded the atmospheric correction (Figure 5C).

In the case of Rrs derived over bright waters, and simultaneously accurately correcting for the adjacency effect from surrounding land, large uncertainties are likely to be present in the result. In reality only a three dimensional radiative transfer correction accounting for a full atmospheric profile of aerosols and surrounding land terrain will likely yield systematically accurate Rrs estimates for small inland water bodies. As this is currently not routinely implemented for any satellite data sources, this remains a challenge for future missions.

Notwithstanding the above challenges, and assuming an atmospheric correction provides satisfactory results, using Rrs can yield significant benefits. Numerous methods exist using Rrs to provide the full suite of physical and biogeochemical products, from algorithmic derivation based on the IOPs, to neural networks and advanced bio-optical radiative transfer inversions. One such example of an available routine algorithm is the MERIS Lakes Neural Network

71

processor which provides a variety of physical and biological estimates (Figure 7). Nevertheless, the challenges from optical variability remain, and in order to obtain accurate quantitative estimates from inversion methods, the optical characteristics of the water in question should be accurately known. However, these complex models are for the moment incapable of adequately characterizing all scenarios encountered in inland waters. It may be necessary to use water-type classification (e.g. Moore et al., 2014) in combination with algorithms targeted at specific water or IOP types.

Fig 7. MERIS Lakes Neural Network Processor.

Chl, tss and cdom products side by side

5.3 OutlookSignificant progress has been made to address the challenges of optical complexity encountered in inland waters, and more knowledge of the IOPs and range and variability of water types are being obtained. This should lead to greater quantitative accuracy from algorithms, especially when used in combination with water type classifications. Whilst methods which use spectral shape and magnitudes applied to partially atmospherically corrected data provide useful information, adequately correcting the signal for atmospheric contamination remains the greatest challenge in inland waters. It remains to be seen whether recently developed methods can be used to provide an accurate and consistent reflectance product for inland waters, for general applications of inversion algorithms.

Whilst low signal-to-noise ratios, and impractical band combinations have limited the use of high-spatial resolution sensors, the next generation of high resolution sensors such as Sentinel-2 MSI and Landsat ALI are likely to yield significantly improved products (e.g., Tomin et al., 2016). Hyperspectral sensors being launched have the challenge of obtaining adequate radiometric sensitivity and calibration to meet the highly quantitative demands of dark water targets. The gains from hyperspectral imagery will see the addition of various improved phytoplankton type detection methods, provided the aforementioned challenges are adequately addressed. The Sentinel-3 OLCI series of satellites will continue to be the go-to product for inland waters given its frequent global coverage, highly quantitative radiometry, sufficient spatial resolution and data availability, and growing suite of algorithm products.

ReferencesDoxaran, D., Cherukuru, N., & Lavender, S. J. (2006). Apparent and inherent optical properties of turbid estuarine waters: measurements, empirical quantification relationships, and modeling. Applied Optics, 45(10), 2310–2324.

72

Evers-King, H., Bernard, S., Robertson Lain, L., & Probyn, T. a. (2014). Sensitivity in reflectance attributed to phytoplankton cell size: forward and inverse modelling approaches. Optics Express, 22(10), 11536–11551. doi:10.1364/OE.22.011536

Giardino, C., Bresciani, M., Cazzaniga, I., Schenk, K., Rieger, P., Braga, F., … Brando, V. (2014). Evaluation of Multi-Resolution Satellite Sensors for Assessing Water Quality and Bottom Depth of Lake Garda. Sensors, 14(12), 24116–24131. doi:10.3390/s141224116

Gower, J. F. R., Doerffer, R., & Borstad, G. a. (1999). Interpretation of the 685nm peak in water-leaving radiance spectra in terms of fluorescence, absorption and scattering, and its observation by MERIS. International Journal of Remote Sensing, 20(9), 1771–1786. doi:10.1080/014311699212470

Guanter, L., Ruiz-Verdú, A., Odermatt, D., Giardino, C., Simis, S., Estellés, V., … Moreno, J. (2009). Atmospheric correction of ENVISAT/MERIS data over inland waters: Validation for European lakes. Remote Sensing of Environment, 114(3), 467–480.

Hestir, E. L., Khanna, S., Andrew, M. E., Santos, M. J., Viers, J. H., Greenberg, J. A., Rajapakse, S.S., Ustin, S. L. (2008). Identification of invasive vegetation using hyperspectral remote sensing in the California Delta ecosystem. Remote Sensing of Environment, 112(11), 4034–4047. doi:10.1016/j.rse.2008.01.022

Hu, C., Lee, Z., Ma, R., Yu, K., Li, D., & Shang, S. (2010). Moderate Resolution Imaging Spectroradiometer (MODIS) observations of cyanobacteria blooms in Taihu Lake, China. Journal of Geophysical Research, 115(C4), 1–20. doi:10.1029/2009JC005511

Kirk, J. T. O. (1994). Light and photosynthesis in aquatic ecosystems. Bristol: Cambridge University Press.

Kiselev, V., Bulgarelli, B., & Heege, T. (2014). Sensor independent adjacency correction algorithm for coastal and inland water systems. Remote Sensing of Environment. doi:10.1016/j.rse.2014.07.025

Kobayashi, H., Toratani, M., Matsumura, S., Siripong, A., Lirdwitayaprasit, T., & Jintasaeranee, P. (2011). Optical properties of inorganic suspended solids and their influence on ocean colour remote sensing in highly turbid coastal waters. International Journal of Remote Sensing, 32(23), 8393–8420.

Kutser, T., Paavel, B., Verpoorter, C., Ligi, M., Soomets, T., Toming, K., & Casal, G. (2016). Remote Sensing of Black Lakes and Using 810 nm Reflectance Peak for Retrieving Water Quality Parameters of Optically Complex Waters. Remote Sensing, 8(497), 1–15. doi:10.3390/rs8060497

73

Lunetta, R. S., Schaeffer, B. a., Stumpf, R. P., Keith, D., Jacobs, S. a., & Murphy, M. S. (2014). Evaluation of cyanobacteria cell count detection derived from MERIS imagery across the eastern USA. Remote Sensing of Environment. doi:10.1016/j.rse.2014.06.008

Matthews, M. W., & Bernard, S. (2013). Using a two-layered sphere model to investigate the impact of gas vacuoles on the inherent optical properties of Microcystis aeruginosa. Biogeosciences, 10, 8139–8157. doi:10.5194/bgd-10-10531-2013

Matthews, M. W., & Odermatt, D. (2015). Improved algorithm for routine monitoring of cyanobacteria and eutrophication in inland and near-coastal waters. Remote Sensing of Environment, 156, 374–382. doi:10.1016/j.rse.2014.10.010

Matthews, M. W., Bernard, S., & Robertson, L. (2012). An algorithm for detecting trophic status (chlorophyll-a), cyanobacterial-dominance, surface scums and floating vegetation in inland and coastal waters. Remote Sensing of Environment, 124, 637–652. doi:10.1016/j.rse.2012.05.032

Mckee, D., Cunningham, A., Wright, D., & Hay, L. (2007). Potential impacts of nonalgal materials on water-leaving Sun induced chlorophyll fluorescence signals in coastal waters. Applied Optics, 46(31), 7720–7729.

Moore, T. S., Dowell, M. D., Bradt, S., & Ruiz Verdu, A. (2014). An optical water type framework for selecting and blending retrievals from bio-optical algorithms in lakes and coastal waters. Remote Sensing of Environment, 143, 97–111. doi:10.1016/j.rse.2013.11.021

Odermatt, D., Kiselev, S., Heege, T., Kneubühler, M., & Itten, K. I. (2008). Adjacency effect considerations and air/water constituent retrieval for Lake Constance. In ESA/ESRIN (Ed.), Proceedings of the 2nd MERIS/(A)ATSR user workshop. Frascati, Italy.

Palmer, S. C. J., Hunter, P. D., Lankester, T., Hubbard, S., Spyrakos, E., N. Tyler, A., … Tóth, V. R. (2014). Validation of Envisat MERIS algorithms for chlorophyll retrieval in a large, turbid and optically-complex shallow lake. Remote Sensing of Environment. doi:10.1016/j.rse.2014.07.024

Robertson Lain, L., Bernard, S., & Evers-King, H. (2014). Biophysical modelling of phytoplankton communities from first principles using two-layered spheres: Equivalent Algal Populations (EAP) model. Optics Express, 22(14), 16745–16758. doi:10.1364/OE.22.016745

Ryan, J., Davis, C., Tufillaro, N., Kudela, R., & Gao, B.-C. (2014). Application of the Hyperspectral Imager for the Coastal Ocean to Phytoplankton Ecology Studies in Monterey Bay, CA, USA. Remote Sensing, 6(2), 1007–1025. doi:10.3390/rs6021007

Santer, R., & Schmechtig, C. (2000). Adjacency effects on water surfaces: primary scattering approximation and sensitivity study. Applied Optics, 39(3), 361–375.

74

Santer, R., & Zagolski, F. (2008). Improve Contrast between Ocean and Land (ICOL): Algorithm Theoretical Basis Document (ATBD): The MERIS Level-1C.

Seppälä, J., Ylöstalo, P., Kaitala, S., Hällfors, S., Raateoja, M., & Maunula, P. (2007). Ship-of-opportunity based phycocyanin fluorescence monitoring of the filamentous cyanobacteria bloom dynamics in the Baltic Sea. Estuarine, Coastal and Shelf Science, 73(3-4), 489–500. doi:10.1016/j.ecss.2007.02.015

Sterckx, S., Knaeps, S., Kratzer, S., & Ruddick, K. (2014). SIMilarity Environment Correction (SIMEC) applied to MERIS data over inland and coastal waters. Remote Sensing of Environment. doi:10.1016/j.rse.2014.06.017

Stramski, D., Babin, M., & Wozniak, S. B. (2007). Variations in the optical properties of terrigenous mineral-rich particulate matter suspended in seawater. Limnology and Oceanography, 52(6), 2418–2433.

Toming, K., Kutser, T., Laas, A., Sepp, M., Paavel, B., & Nõges, T. (2016). First Experiences in Mapping Lake Water Quality Parameters with Sentinel-2 MSI Imagery. Remote Sensing, 8(640), 1–14. doi:10.3390/rs8080640

Wynne, T. T., Stumpf, R. P., Tomlinson, M. C., Warner, R. A., Tester, P. A., Dyble, J., & Fahnenstiel, G. L. (2008). Relating spectral shape to cyanobacterial blooms in the Laurentian Great Lakes. International Journal of Remote Sensing, 29(12), 3665–3672.

75

Chapter 6: SensorsErin Hester, Arnold Dekker, Menghua Wang, Mark Matthews

6.1 IntroductionThe GEOSS Water Strategy (2014) highlights the importance of Earth observation satellites in providing information for monitoring the water cycle. In many regions around the world, there is little or no information available for water management, and satellite data may meet some of the information gaps. Additionally, their systematic, global coverage also helps to address problems of data continuity in trans-boundary basins where complete, consolidated, and consistent information may be difficult to obtain. A specific observation need for the future that is not covered in the current strategy is ”enhanced water quality monitoring over inland water bodies and in coastal zones.” The GEOSS Water Strategy recommends a global-scale coordinated effort that advances the future use of satellite remote sensing for water quality applications, and emphasizes continuity of existing satellite capabilities and development of new and improved sensor and platform technologies.

In support of GEO observation objectives, the Committee on Earth Observation Satellites (CEOS) coordinates internationally with member agencies to ensure that space and ground segment observations are coordinated around a particular set of parameters. These coordinated space-based observations are called virtual constellations. Unfortunately, as of the publication of this report, there is no virtual constellation for inland and near-coastal water quality. The GEOSS Water Strategy (2014) has recommended that CEOS assess the benefits and technological challenges of a hyperspectral satellite mission focused specifically on water quality.

This chapter addresses these recommendations by first discussing the optically active water quality variables that we have mentioned throughout this report that need to be measured from space. We then show the specific inherent optical properties that determine the effect of varying concentrations of OAC’s on the spectrum. We show through some simulations and through analyses of measured spectra how the water remote sensing reflectances spectra vary. Next we look at the spectral band requirements for atmospheric correction. We then discuss the spatial requirements by looking at global distribution of size classes of lakes and of rivers. Finally we discuss the radiometric accuracy and sensitivity requirements. For an actual sensor specification a trade-off is required between these requirements as in the end there is only a limited amount of photons (limited by spectral resolution, radiometric resolution and spatial resolution) that can be measured from space at a velocity approaching 30,000 km per hour for a polar orbiting satellite at altitudes between 450-800 km. For geostationary satellites this equation different due to a longer sensor –earth distance (~30,000 kms) but higher integration times as the sensors remains fixed at the same location over the Earth (it can dwell).

76

The second half of this chapter assesses the application potential of relevant existing and planned spaceborne sensors (where the specifications are already set) as well as makes recommendations on desired specifications on future sensors.

It is relevant to briefly discuss the issue of accuracy and validity of retrieval of water quality variables:

The accuracy of water quality constituent concentrations derived from remote sensing will vary depending on derived product region of interest, and time of the year. Accuracy will also vary due to concentration of the substance, the matrix of other optically active substances in the water, availability of appropriate wavebands, atmospheric corrections, and the accuracy and representativeness (see below) of the in situ data set. Typical accuracy targets for most ocean colour products are better than ±30% (IOCCG 2000) and inland and coastal waters are generally expected to have the same magnitude of accuracies. Examples??? The additional value of satellite remote sensing is not always absolute accuracy of individual retrievals, but synoptic and frequent coverage of numerous waterbodies. Trends?? Threshold? Even with larger uncertainties in accuracy estimates, parameter concentration change could be detected if the product was derived with a consistent methodology (Stumpf et al. 2003; Hu et al. 2005). Algorithms applied to operational efforts need pass the rigors of peer-reviewed publications and be robust in covering the range of constituent concentrations with quantifiable errors.

Before we discuss spectral bands it is relevant to first look at Fig 1 that gives a typical inland water concentration specific or pure water optical properties: and the next 4 figures that show the effect of varying concentrations on the reflectance spectrum of inland waters when we vary either the tripton, the CDOM or the chlorophyll levels or all variables together including cyanophycocyanin.(comment AD- this needs fine-tuning still).

The validity is more important: we need to know that an algorithm that makes use of spectral bands is actually (through a causal relationship) indeed measuring that variable and not some co-varying variable.

77

78

79

80

81

82

Spaceborne sensor requirements and technical constraints across multispectral and hyperspectral configurations and spatial and temporal resolution

83

Several of the mission requirements for future inland and near-coastal water quality sensors are similar to those of ocean colour sensors. These requirements are thoroughly documented in the IOCCG Report 13 “Mission requirements for future ocean colour sensors.” The 13th IOCCG report focuses on ensuring a minimum set of requirements for any coastal-ocean colour sensor to ensure high quality data. In this chapter, we only discuss where mission requirements for

84

inland and near-coastal water quality observations deviate from open ocean and coastal waters, namely:

● Differences in spectral band requirements due to differences in the range and composition of concentrations of optically active substances and their relative contributions to the measured signal as well as needing to be able to cope with optically shallow waters.

● Differences in spatial scales such as smaller water bodies and high spatial complexity of land-water boundaries.

● Atmospheric correction, including land adjacency effect, effects of highly turbid waters, strongly absorbing aerosols, etc.

We refer the reader to IOCCG report 13 for a full discussion of mission requirements for future ocean colour sensors.

6.1 Spectral band requirements for estimating water variables Derivative analysis of hyperspectral data has been used widely (e.g. Lee et al., 2007) to determine the bands that encompass the largest signal variability and therefore to suggest the optimal location of the spectral bands for remote sensing of environment. Here we present some preliminary results of spectral derivatives using a representative subset (n=1500) of in-situ remote sensing reflectance (Rrs(λ) in sr-1) sourced from the in-situ bio-optical data repository LIMNADES [note from Menghua: will these data be open to the public? I don’t feel this is a good idea to use the data] (Lake Bio-optical Measurements and Matchup Data for Remote Sensing) (Figure 6.1).

Figure 6.1. Mean and standard deviations of water reflectance calculated from radiometric measurements in a wide variety of inland and nearshore environments.

85

These data were collected from more than 150 aquatic systems, mainly lakes but also nearshore environments and rivers, around the world. Chlorophyll a (Chla) concentration in this dataset ranged between 0.01 and 10000 mg m-3. All data were interpolated to 1 nm and smoothed prior to the analysis. The peak wavelengths in Figure 6.2 correspond to 0 values in the first (black line) and second-order derivatives (grey line). Figure 2 also shows the location of MERIS, MODIS, OLI and OLCI spectral bands for comparison. The peaks between 700 and 800 nm may reflect variations in the Rrs(λ) due to very high load of sediments, presence of cyanobacteria scum or instrument noise and sun-glint in those wavelengths. It should be also noted the peak between 600 and 650 nm for both derivatives as a result of the phycocyanin absorption at the wavelength around 625 nm. Table 6.1 summarises the bands that embrace the inflections of remote sensing reflectance calculated from measurements in inland, coastal and oceanic waters. These bands could be used to guide future missions for remote sensing of aquatic environments.

86

While this is a powerful analytical tool for identifying parts of the spectrum where most of the variability occurs, it does not account for the fact that the position of relevant peaks can undergo significant shifts. For example, several studies have documented a change in the position and amplitude of the reflectance peak around 700nm associated with differences in the concentration and properties of Chla (Gitelson, 1992). A variety of Chla retrieval algorithms have been developed since then for inland and near-shore waters (e.g. Gower et al., 1999; Gitelson et al., 2007; 2009) employing spectral bands located in the red-NIR regions. Using the dataset described above we found strong dependencies between the position of the peak in the region between 690 and 715 nm and the mean concentration of Chla per wavelength (peak position=679.55*<Chlamean>0.0083, r2=0.95). Relatively low Chla concentrations (5.35±5.19 mg m-3) reveal maximum values of Rrs(λ) at the lower end of this range (690 nm) and then the peak gradually shifts with increasing Chla reaching mean concentrations of 200 mg m-3 at around 713 nm.

Figure 6.2. Zero value frequencies of the first (black line) and second-order (grey line) Rrs derivatives and spectral bands of some past (MERIS), current (MODIS; OLI) and near-future water colour sensors (OLCI). [Note from Erin: I’m not convinced this figure is informative to this chapter. I think it is overly technical and too detailed. Will most readers easily interpret this? Does it communicate the key take home message? Would be better to combine with info in table 6.2 that actually communicates band results and comparison.]

87

Table 6.1. Spectral location (considering the nearest 5nm wavelength) of the inflection points using in-situ hyperspectral data from mainly inland (Spyrakos et al. in preparation) and oceanic and coastal waters (Lee et al. 2007). Optimal bands are indicated by the first (1) and second (2) order derivatives of Rrs. In addition we added a column that specifies what are essential spectral bands to measure a variable due to its absorption maximum as well as adjacent bands required to measure the absorption feature ( -adjacent bands that do not show the absorption features and are thus a reference for the other optically active constituents> The range of spectral bands column shows those variables that do not have a spectraly localised absorption feature such as algal pigments but show a general often a negative exponential relationship with increasing wavelength and therefore show a range over which preferably three bands are required to model that exponential shape.

Notes: From Erin: I’m not convinced this figure is informative to this chapter. I think it is overly technical and too detailed. Will most readers easily interpret this? Does it communicate the key take home message? Would be better to combine with info in table 6.2 that actually communicates band results and comparison. From Menghua: I think we can just use the published results, e.g., from Lee, just one figure, replacing both Fig. 6.2 and Table 6.2. I don’t think it is good idea to use un-published results in the report. From Erin: This table is highly ineffective. Suggest new figure to replace fig 6.2 and this table that encompasses both of the results. Or perhaps a comparison matrix instead.

Nearest to 5nm

wavelength

Essential spectral bands

Range of spectral bands Spyrakos et al. (in preparation)

Lee et al.

(2007)

400 CDOM 2

425 Chl-a non-absorbing reference

CDOM 1,2 2

438 Chl-a a CDOM

440 CDOM 1,2 1,2

445 CHl-a non-absorbing CDOM 1,2 1,2

88

reference

460 CDOM 2 2

465 CDOM 2

475 CDOM 2

480 CDOM 1,2

490 CDOM, NAP 1 1

510 CDOM, NAP 2 2

515 CDOM, NAP 1

520 CDOM, NAP 2 2

540 CDOM, NAP 2

545 CPE non-absorbing reference

CDOM, NAP 2

555 CPE non-absorbingreference

CDOM, NAP 2 2

565 CPE CDOM, NAP 1,2 1

580 CPE non-absorbing reference

CDOM, NAP 2

585 CPE/CPC non-absorbing reference

CDOM, NAP 1,2

615 CPC referencenon-absorbing

NAP 2

625 CPC

635 CPC non-absorbing reference

NAP 1,2 2

645 NAP 2

655 Chl-a non-absorbing reference

NAP 2

665 Chl-a NAP 2 2

89

670 NAP 2

676 Chl-a NAP 2

685 Chl fluorescence NAP 1 1

690 NAP 2

695 NAP 1

700 CHL/NAP

705 CHL/NAP

710 CHL/NAP

715 CHL/NAP

720 CHL/NAP

730 Sun and sky glint/NAP/atmospheric

correction

740 Sun and sky glint/NAP/atmospheric

correction

750 Sun and sky glint/NAP/atmospheric

correction

1,2 NA

770 Sun and sky glint/NAP/atmospheric

correction

1,2 NA

865 Atmospheric correction

1240 MODIS & VIIRS

1600 VIIRS

1640 MODIS

2130 MODIS

2257 VIIRS

90

Inland and coastal waters require more spectral bands at a higher spectral resolution than open ocean conditions due to their optical complexity. In most inland and coastal waters, the optically active constituents can both co-vary and vary independently, sometimes in the same water body (Ampe et al., 2014; 2015). For example, in many inland lakes it is possible to have both high suspended matter and high Chlorophyll-a concentrations. In this case, the effect on reflectance is to change the absorption depth and location of spectral features to longer wavelengths. Thus, optimizing a band position is highly dependent on the inland water quality conditions, which can vary widely across geographic regions and seasons. An optimal inland water quality sensor would have many narrow spectral bands eliminating the need for a priori selection of band positions which may fail to measure all of the optically active constituents.

It is possible to provide a preliminary estimate of the spectral band requirements for inland and coastal waters using the new LIMNADES [comment from Menghua: Will these data be open to public? I don’t feel this is a good idea to use the data.] database (Spyrako et al., in prep-need citation). The LIMNADES collection represents more than 150 aquatic systems from around the world. The data were mainly collected in lakes, but there are also nearshore environments and rivers represented, ranging from oligotrophic to eutrophic conditions (Chlorophyll-a concentration ranges from 0.01 and 10000 mg m-3). There are over 1500 in situ surface radiometric measurements in the database (Figure 6.1), made at a spectral resolution of X to X nm [comment from Erin: Unable to procure info from co-author. Suggest follow up. Arnold or Steve can you apply some pressure?], dependent on the data source and instrument used.

91

Figure 6.1. Mean and standard deviations of water reflectance calculated from radiometric measurements in a wide variety of inland and nearshore environments.

Spectral derivatives were calculated from the LIMNADES radiometric data in order to approximate optimal spectral band requirements. Derivative analysis of hyperspectral data has been used widely (e.g. Lee et al., 2007) to determine the bands that encompass the largest signal variability and therefore to suggest the optimal location of the spectral bands for remote sensing of environment. All data were interpolated to 1-nm resolution and smoothed prior to the analysis. Figure 6.2. shows the results of the derivative analysis. Table 6.1 summarises the bands that embrace the inflections of remote sensing reflectance calculated from measurements in inland, coastal and oceanic waters. Because the LIMNADES database, while large, still represents a limited range of inland and coastal optical conditions, we caution that these bands should be used only as guides, and not as recommendations for future missions for remote sensing of aquatic environments.

The peak wavelengths in Figure 6.2 correspond to 0 values in the first (black line) and second-order derivatives (grey line). Figure 2 also shows the location of MERIS, MODIS, OLI and OLCI spectral bands for comparison. The peaks between 700 and 800nm may reflect variations in the Rrs due to very high load of sediments, presence of cyanobacteria scum or instrument noise in those wavelengths. It should be also noted the peak between 600 and 650 nm for both derivatives as a result of the phycocyanin absorption at the wavelength around 625 nm.

There are spectral regions that are important to both inland and coastal waters, notably, around 440nm, 490, 555, 565, 665 and 685nm. However, there are clear differences in the optimal spectral band position between inland coastal, and oceanic waters as expected. While sensors such as MERIS and OLCI have better placed band positions for inland waters relative to Landsat or MODIS, even these ocean-colour sensors are not optimally designed for inland waters. Thus, an optimal inland water quality sensor would have many narrow spectral bands that could capture the varying reflectance features found in inland and near-coastal waters.

92

Figure 6.2. Zero value frequencies of the first (black line) and second-order (grey line) Rrs derivatives and spectral bands of some past (MERIS), current (MODIS; OLI) and near-future water colour sensors (OLCI). [note from Erin: I’m not convinced this figure is informative to this chapter. I think it is overly technical and too detailed. Will most readers easily interpret this? Does it communicate the key take home message? Would be better to combine with info in table 6.2 that actually communicates band results and comparison.]

Figure 6.2. Spectral location (considering the nearest 5nm wavelength) of the inflection points using in-situ hyperspectral data from mainly inland (Spyrakos et al. in preparation) and oceanic and coastal waters (Lee et al. 2007). Optimal bands are indicated by the first (1) and second (2) order derivatives of Rrs. [comment from Menghua: I think we can just use the published results, e.g., from Lee, just one figure, replacing both Fig. 6.2 and Table 6.2. I don’t think it is good idea to use un-published results in the report; comment from ErinThis table is highly ineffective. Suggest new figure to replace fig 6.2 and this table that encompasses both of the results. Or perhaps a comparison matrix instead.]

6.2 Spectral bands for atmospheric correctionIn remote sensing of water properties over coastal and inland waters, a key step (or data processing procedure) is atmospheric correction (Gordon and Wang, 1994; IOCCG, 2010), i.e., accurately removing atmospheric and water surface radiance contributions and deriving the normalized water-leaving radiances. All other water optical, biological, and biogeochemical

93

properties, e.g., chlorophyll-a, diffuse attenuation coefficients, total suspended sediment, inherent optical properties (IOPs), etc., are derived from satellite-measured normalized water-leaving radiances after atmospheric correction. It should be noted that, since normalized water-leaving radiances usually contribute only about < 10% in the visible wavelengths at the top of the atmosphere (TOA), data quality over inland waters is highly sensitive to the performance of the satellite on-orbit calibrations. It has been shown that for open oceans (or clear inland lakes) ocean color products can be routinely produced using the near-infrared (NIR)-based atmospheric correction (Gordon and Wang, 1994; IOCCG, 2010), e.g., global ocean color products from SeaWiFS, MODIS, MERIS, and VIIRS (McClain, 2009; Wang et al., 2013). However, over turbid coastal and inland waters, atmospheric correction using the shortwave infrared (SWIR)-based and NIR-SWIR combined approaches (Wang, 2007; Wang and Shi, 2007) are useful to derive accurate water property data. Thus, for remote sensing of coastal and inland waters, both the NIR and SWIR bands with the required sensor performance are necessary. MODIS (also VIIRS) [note from Arnold - Menghua: which do you recommend?] has the NIR bands at about 748 and 869 nm (VIIRS at 745 and 862 nm) and SWIR bands at 1240, 1640, and 2130 nm (VIIRS at 1238, 1601, and 2257 nm). However, the SWIR bands have poor performance (Wang and Shi, 2012) and do not meet the sensor performance requirement for ocean color remote sensing (Wang and Shi, 2012). To meet the requirement for atmospheric correction over coastal and inland waters, it needs high performance in sensor signal-to-noise (SNR) requirements, particularly for the SWIR bands (Wang, 2007, Wang and Shi, 2012). In fact, Wang et al. (2012) have shown that the required SNR values for the NIR bands are ~600, while for the SWIR 1240, 1640, and 2130 nm bands are ~250, ~200, and ~100, respectively.

To meet the requirement for atmospheric correction over coastal and inland waters, it needs high performance in sensor signal-to-noise (SNR) requirements, particularly for the SWIR bands (Wang, 2007, Wang and Shi, 2012). In fact, Wang et al. (2012) have shown that the required SNR values for the NIR bands are ~600, while for the SWIR 1240, 1640, and 2130 nm bands are ~250, ~200, and ~100, respectively.

6.3 Spatial Resolution Requirements

[note from Erin: The one thing we are missing are coasts. The PACE science preparatory activities are looking at this, and there are some results and recommendations about how far from the coast you can go before you stop seeing an optical coastal signal (~100 m). I also think Blake Schaeffer has a paper about this too…We should encompass this in our work.]

One of the primary limitations in adapting ocean colour sensors to inland and coastal applications is the spatial resolution of the sensor relative to the size of the water body and the spatial complexity of the land-water boundary. Because of the general need for higher spatial resolution, researchers frequently appropriate land-imaging missions (e.g., Landsat, Worldview-

94

2, MSI-Sentinel-2) for inland and near-coastal water quality applications. However, this comes with significant tradeoffs because most land-imaging missions do not meet the spectral resolution requirements described above.

Recently, several studies have reported the global distribution and spatial aspects of lakes and rivers, allowing us to provide resolution requirements for global measurements of inland waters. For the primary objective of measuring lake surface water quality, a minimum spatial resolution of about 300 m is required. To measure rivers, a minimum of about 20 m is required. The global abundance and size distribution of lakes has recently been reported by Verpoorter et al. (2014) using analysis based on Landsat imagery. They quantified the number and surface areas of the world lakes larger than 0.002 km2 (0.2 ha) (see Fig. 1). Andreadis et al. (2013) reported estimates of global river width, depth and discharge using a combination of Landsat image analysis and hydrologic modeling. Using these datasets, we estimated the total number of lakes and rivers globally resolvable by different sensor spatial resolutions.

Figure 6.3. a) Size distribution of the world’s lakes, b) cumulative and total area according to size class of world’s lakes, c) total and cumulative number of global river widths by size class.

95

Table 6.2. Ground sampling distance requirements showing resolvable size class and total cumulative number and area coverage of the world’s lakes (based on assumptions using Verpoorter et al. (2014) dataset).

Size Class Required GSD* % Total Area Total number

≥ 10 km2 1054 m 44 25,976

≥ 1 km2 333 m 60 353,552

≥ 0.1 km2 105 m 80 4,123,552

≥ 0.01 km2 33 m 90 27,523,552

≥ 0.002 km2 15 m 100 117,423,552

*Calculated using a box of 3 x 3 pixels sufficient to resolve the specified lake size

Table 6.3 Ground sampling distance requirements showing the resolvable river width class and cumulative number of total river reaches of the world’s rivers from Pavelsky et al. (2012) dataset.

River Reach Size Class(width)

Required GSD* Total number of reaches Percent of total reaches

≥ 1.5 km 500 2,877 < 0.1%

≥ 1 km 333 8,483 <1%

≥ 0.5 km 167 35,420 1%

≥ 0.1 km 33 382,466 12%

≥ 0.05 km 17 766,303 24%

≥ 0.01 km 3 2,576,452 81%

*Calculated using a box of 3 x 1 pixels sufficient to resolve the width of the river reach

The vast majority of lakes occur in size classes less than 1 km2. While there are fewer number of lakes in size classes greater than 1 km2, they contribute a large portion of the total surface area covered by lakes. Therefore, in theory, the majority of the world’s surface lake area (~60%) can be resolved with a sensor GSD of ~333 m (Table 1). This is the equivalent of a sensor with

96

MERIS/OLCI-type resolution. If smaller lake classes are considered, nearly 80% of the global lake surface area be viewed with a sensor with a GSD of 105 m, whereas 100% of the global lake area of lakes 0.2 ha or larger can be resolved with 15 m spatial resolution (e.g., Sentinel-2 type sensor resolution). Thus, we estimate that a Landsat resolution sensor (30 m) is capable of detecting approximately 27 million lakes worldwide, or more than 90% of the global lake surface area. The ability to resolve rivers from space requires much higher GSD than lakes. The vast majority of global river reaches are less than 10 m wide, requiring a sensor resolution of ~3 m (Table 2). Less than 1% of total river reaches are resolvable at the MERIS/OLCI sensor type resolution, and only 12% of all river reaches are resolvable using Landsat sensor type resolution. Encouragingly, more than one quarter of global river reaches are wide enough to be resolved from Sentinel-2 type sensors (> 17 m GSD). Due to shoreline complexity of water bodies and cloud cover, the actual number and area of lakes and river widths resolvable in each size class will likely be less than the estimates provided herein. For example Sayers et al. (2015) determined that approximately 80,000 lakes were resolvable globally in the 2011 growing season using MERIS, and only 58,000 lakes globally were theoretically resolvable using MODIS.

Similarly, using an analysis of polygons, Hestir et al. (2015) determined that MERIS resolution is capable of observing more than 50% of the area of inland waters in Europe. However, by comparison, MERIS observed only a few percent of Australia’s water bodies, where a 30 m resolution sensor resolved less than 50% of the area due to differences in topography and water body geometry. This demonstrates that spatial resolution requirements may differ substantially based on geography. However, global requirements should be prioritized over local ones in the design of an inland sensor. Based on these calculations, the minimum spatial resolution requirement for inland water bodies can be categorized in two bins. A GSD of 300 m is satisfactory to observe the majority of the world’s lake surface areas, but this area represents only a small fraction of the total number of lakes on the planet. A sensor with a minimum GSD of 20 m would enable observations for nearly a quarter of global river reaches.

6.4 Radiometric accuracy requirementsBoth IOCCG Report 13 and Report 10 set a goal of 0.5% accuracy for TOA radiance at 443nm. This is necessary to achieve a water-leaving radiance accuracy of 5% (at 443 nm) and a chlorophyll-a product accuracy of ~30%. Dekker et al. (2001) evaluated these requirements for inland chlorophyll product accuracy across a range of oligotrophic to hypertrophic water bodies. They recommended accuracy goals for green and NIR bands rather than just the blue bands of the IOCCG reports due to the spectral properties of inland waters (see spectral resolution requirements above). [note from Menghua: The requirement for nLw at 443 nm is because errors from atmospheric correction are large at the short wavelengths. Indeed, turbid waters red

97

and NIR bands will be useful for deriving Chl-a data. However, there are also plenty very clear inland lakes, which needs blue-green nLw for Chl-a data. Discussions in here are confusing/not quite right.] This is because in inland waters the combined absorption of phytoplankton pigments CDOM and NAP may cause very low Rrs, making green-red and even NIR wavelengths essential to water OAC retrieval. Thus, a goal for inland sensors is 0.5% for the TOA radiance uncertainty at all bands after vicarious calibration (Gordon, 2010). This is achievable using current infrastructure, as the combined uncertainty of MOBY is reported as 2-3% (Brown et al., 2007) (See IOCCG Report 13 for details).

6.4.1 Signal-to-Noise Ratio or radiometric sensitivity requirements

The effect of SNR affects both the accuracy of atmospheric correction on at sensor radiance measurements as well as the subsequent retrieval of optically active constituent retrieval, and can contribute up to one quarter of the optically active constituent retrieval error (Salama and Stein 2009).

Optically active constituent retrieval accuracy is strongly related being able to separate the desired radiance signal from the overall noise in a pixel or image (Wettle et al., 2014) due to both sensor performance other sources of noise. Two primary factors determine the accuracy and precision of retrieval: environmental noise equivalent radiance (NEΔ LE) and the environmental noise equivalent delta reflectance (NEΔ RE) (Brando and Dekker, 20013). While the NEΔ RE is more easily simulated or inverted using bio-optical forward and inverse models, it can vary with illumination conditions. NEΔ LE on the other hand is an absolute measure of the radiance level that a sensor needs to discriminate, remaining constant under all illumination conditions and is thus the most relevant for sensor design. The effects of NEΔ LE are inversely proportional to signal-to-noise ratio (Salama and Stein 2009). [note from Menghua: Essentially, the two parameters discussed are the same. I think the paragraph should be deleted (not quite right).] The effect of SNR on retrieval accuracy varies based on the concentration of the optically active constituent in the environment. Typically, the higher the concentration, the higher the accuracy of retrieval for a given SNR. Hoogenboom & Dekker (1998) performed a sensitivity analysis of retrieval of Chlorophyll-a in coastal and inland waters using an imaging spectrometer (AVIRIS, NEΔ LE =0.0005 W m-2 nm-1 sr-1 at 400 nm to 0.0005 W m-2 nm-1 sr-1 in the range of 500-750 nm). For the inland lake, the signal-to-noise ratio of AVIRIS allowed estimations of Chlorophyll-a with an error of 20% for relative low Chlorophyll concentrations (10 mg m-3) and 11–12% for high concentrations (>30 mg m-3). In the coastal North Sea, the estimated error ranged from 33% for low concentrations (2 mg m-3) and 10–12% for concentrations greater than 12 mg m-3. [note from Menghua: This is in contrast to our experience from satellite remote sensing; high errors with high Chl-a because of errors from atmospheric correction, i.e., dealing with low signals from waters.]

98

Retrieval accuracies can be improved by making radiometric enhancements that increase the number of photons measured at each pixel, through modifying data acquisition strategies, or through image processing techniques. This enhancement can be obtained by increasing the F-number (aperture) of the camera lens, increasing the detector pixel size; increasing integration time (possible from geostationary without consequences for pixel size or shape) but not from polar orbit as the overpass speed needs to remain constant)-unless dwelling is used (focusing the sensor on one target whilst the spacecraft passes over). Using simulations, Moses et al. (2015) found that a 12 times increase in pixel photons resulted in a SNR increase of four times through the VIS and up to 10-12 times higher in NIR (due to significant dark noise and read noise improvements). Another way to increase SNR is to average across multiple pixels (often 3x3, 5x5, etc.), which also results in an increase in product retrieval accuracy (e.g., Brando and Dekker, 2003; Giardino et al., 2007, Vanhellemont and Ruddick, 2014; Moses et al. 2015). Unfortunately, the result of this spatial binning is an effective decrease in spatial resolution (Hestir et al., 2015).

From IOCCG 13:Table 4.1 [Menghua: Simply refer to the report and the corresponding table should be enough. I don’t think this is a good idea to copy and paste the table.] Multispectral band centers, bandwidths, typical top-of-atmosphere clear sky ocean radiances (Ltyp), saturation radiances (Lmax), and minimum SNRs at Ltyp. Radiance units are W m−2 µm−1 sr−1. SNR is measured at Ltyp. Lmin and Lhigh are TOA radiance ranges for valid ocean-colour retrievals derived from a SeaWiFS global one-day data set for the respective SeaWiFS bands, after removing the 0.5% highest and 0.5% lowest radiances. In future, these values should be derived for the remaining bands. Adjustments may be necessary for sensors with different solar and viewing geometries[E2] . [Arnold: We need to redesign this table as it is geared too much to oceans. It misses a cyanophycocyanin band e.g. How do we choose bandwidth? This really requires a sensitivity analysis. I did one in 93 for chlorophyhll and found 12 nm was optimal centered at 676 nm, similar for CPC at 624 nm and CPE at 565. Any more lit refs??] [Erin: Still unaddressed Do we actually need this?]

99

Typical dynamics associated with phenomena in inland standing water bodies (lentic). Rivers should be treated separately as lotic systems with different dynamics and phenomena. [ Lentic / Lotic divide probably should be discussed earlier in report]. Temporal basis of in situ sampling – e.g. automated systems with online near real time data, incorporate papers. Global coverage estimates (IOCCG 1/13). Use revisit / global coverage of Sentinel-3 OLCI as an example of a global constellation of 2 or more satellites in a planned mission? Could also use MODIS as a good example of dual constellation with daily revisit. Need for constellations to improve the revisit for higher resolution sensors (e.g. Landsat-8 with 16d revisit does not meet min requirement without a constellation).

100

Limits of RS – your design resolution versus effective resolution argument is good. Latitudinal effects of cloud cover / glint reducing effective revisit.

Table 2. Observational frequency required to resolve phenomenal occurring on various time scales.

Phenomenon Time-scale of phenomenon variability

Frequency to resolve changes

Diurnal migration and cycle

hourly Geostationary only

the development and movement of a harmful algal bloom

hourly to daily Hourly: geostationary; daily

wetland phenologic change

monthly quarterly

aquatic invasive species response to management

monthly quarterly

phytoplankton community composition

sub-weekly weekly

seasonal responses in biomass

monthly weekly

Trade-offs and solutions (blending) between spectral, spatial, radiometric and temporal requirementsTrade-offs required (perhaps a diagram explaining the constraints within which an inland and near-coastal sensor would have to sit : e.g. cannot have 1000:1 S:N and 1 m resolution and imaging spectrometry with 5 nm bands(just not enough photons).We have seen from the discussion on spectral resolution that for most phytoplankton pigment variables a spectral band resolution of 10 nm is required (with 5 nm for chlorophyll fluorescence ), whereas 20 n wide bands or even wider are sufficient for the

101

OAC’s that show a more gradual (often exponential shaped) relationship with increasing wavelength. From the spatial analysis results we find that for lake (shaped) water bodies globally a 15 m GSD will sample 100% of lakes > 0.002 km2 ; and a GSD of 33m will sample 90 % of lakes. For rivers a GSD of 33 m 12 % of river reaches would be sampled and with a GSD of 17 would image 24 % of rivers and with GSD of 3 m 80 % of river reaches would be sampled.

Therefore it may make sensor for a global mapping system to have spectral bands of 10 nm preferably from about 400 nm to 800 nm with a few additional atmospheric correction bands at longer wavelengths. A GSD of 33 m seems to be adequate for a global mapping mission, with the advice that higher spatial GSD resolution will include more water bodies. The revisit time should be as high as possible so that after subtraction of all images with any form of cloud, cloud shadow, sunglint, smoke, haze, low sun angles etc there is still a sufficient amount of imagery available for detecting, assessing and monitoring water quality. For geostationary sensors the trade-off will focus more on the spatial resolution aspect as the temporal aspect and the radiometric sensitivity are compensated (as compared to polar orbiting sensors).

6.5 The application potential of relevant existing and planned spaceborne sensors (where the specifications are already set) Suitability of past, current and near future sensors (that we cannot change) to measure these direct and derived variables: past SeaWiFS, MERIS, current high spatial resolution MSS (WV-2 &3, RapidEye, SPOT etc.) ; current mid resolution S2-MSS and Landsat * OLI as well as theS3-OLCI, and near future sensonrs SGLI, GOCI-II;(Please see http://www.ioccg.org/sensors_ioccg.html)

6.6 Recommendations on desired specifications on future sensors.Recommendations for (advanced) planned sensors that can still be influenced (e.g. Sentinel-2 C and D, Landsat 9 etc; geostationary......?)

● Sentinel-2 : if all Sentinel 2bands could be 10 m resolution that would be a significant benefit for inland water remote sensing. For S-2 and Landsat some extra bands such as at CHL and CPC suitable wavelengths of 625 and 676 nm would significantly enhance their suitability as global mission for inland water quality.

Recommendations: perfect constellation (within realistic physics and instrumentation constraints) for an ideal sensor-mission approach As mentioned in Hestir et al, (2015) a hyperspectral mission with approximately 10 nm wide bands covering the range of 400-2500 nm is the ideal sensor, where spatial

102

resolution would need to be as high as feasible but at a minimum 33 m GSD, The S:N should be as close as possible to 1000:1 over the 400 to 800 nm wavelength range with 600 :1 for the longer wavelength bands needed for atmospheric correction. The revisit frequency should be as high as is affordable ( either large swath width or multiple sensors). The intended S-2 frequency of on image every 5 days with two Sentinel-2 sensors is highly suited for all water quality assessments except for those features that occur at hourly of daily time frequencies. Sunglint should be avoided wherever possible; if it is not possible to avoid sunglint it is essential that the sensor does not get saturated over sunglinted areas as it than remains possible to remove a significant part of the sunglint from the image.

References

Dekker, A. G., T. J. M. Malthus, et al. (1992). “The effect of spectral band width and positioning on the spectral signature analysis of inland waters.” Remote Sens. Environ. 41(2/3): 211-226. A. G. Dekker, V. E. Brando, J. M. Anstee, N. Pinnel, T. Kutser, H. J.Hoogenboom, R. Pasterkamp, S.W. M. Peters, R. J. Vos, C. Olbert, andT. J. Malthus, (2001) “Imaging spectrometry of water,” in Imaging Spectrometry:Basic Principles and Prospective Applications. Dordrecht, TheNetherlands: Kluwer, Vol. 4, Remote Sensing and Digital ImageProcessing, pp. 307–359. Hestir, E.L., Brando, V.E., Campbell, G., Dekker, A.G. and Malthus T. J. (2015), The relationship between dissolved organic matter absorption and dissolved organic carbon in reservoirs along a temperate to tropical gradient, Remote Sensing of Environment, Volume 156, January 2015, Pages 395-402, http://dx.doi.org/10.1016/j.rse.2014.09.022. Hestir, E.L., Brando, V.E., Bresciani, M., Giardino, C., Matta, E., Villa, P. and, Dekker, A.G.(2015) Measuring freshwater aquatic ecosystems: The need for a hyperspectral global mapping satellite mission, Remote Sensing of Environment, Volume 169(?), November 2015, Pages xxx, doi:10.1016/j.rse.2015.05.023 Lawford, R. The GEOSS water strategy: from observations to decisions (2014); Group on earth Observations, Geneva; published by Japan Aerospace Exploration Agency, Japan; pp 255. Lee, Z., Applying narrow band remote-sensing reflectance models to wideband data (2009); Applied Optics Vol. 48, No. 17: pp 3177-3183. Lee, Z., Carder, K., Arnone, R. and He, MX (2007) Determination of Primary Spectral Bands for Remote Sensing of Aquatic Environments; Sensors 2007, 7:pp 3428-3441. Leijtens, J., Perez-Calero, D., Dekker, A. G., Peters, S.W.M. (2011) SWIMS (hyperSpectral Water Imaging & Monitoring System) a hyperspectral constellation concept for aquatic ecosystems; 34th International Remote Sensing of Environment Conference, Sydney, Australia, 1-15th April 2011.McClain, C.R. & Meister, G., (2012) IOCCG Report nr 13: Mission Requirements for Future Ocean-Colour Sensors: pp. 106.

103

Moses, W.J., Bowles, J.H. & Corson, M.R. (2015) Expected improvements in the quantitative remote sensing of optically complex waters with the use of an optically fast hyperspectral spectrometer- A modeling study. Sensors 2015, (15): p 6152-6173. IOCCG (2010), Atmospheric Correction for Remotely-Sensed Ocean-Colour Products, Wang, M. (Ed.), Reports of International Ocean-Color Coordinating Group, No. 10, IOCCG, Dartmouth, Canada. IOCCG (2012), Mission Requirements for Future Ocean-Colour Sensors, C. R. McClain and G. Meister (Eds.), Reports of International Ocean-Colour Coordinating Group, No. 13, IOCCG, Dartmouth, Canada.

@@@@@@@@@@@@@@@@@@@@

Gordon, H. R. and M. Wang, “Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: A preliminary algorithm,” Appl. Opt., 33, 443–452, 1994. Wang, M., “Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: simulations,” Appl. Opt., 46, 1535–1547, 2007. Wang, M. and W. Shi, “The NIR-SWIR combined atmospheric correction approach for MODIS ocean color data processing,” Opt. Express, 15, 15722–15733, 2007. Wang, M. and W. Shi, “Sensor noise effects of the SWIR bands on MODIS-derived ocean color products,” IEEE Trans. Geosci. Remote Sens., 50, 3280–3292, 2012. C. R. McClain, “A decade of satellite ocean color observations,” Annual Review of Marine Science 1, 19–42 (2009). Gitelson, A. (1992). The peak near 700 nm on the radiance spectra of algae and water: relationship of its magnitude and position with chlorophyll concentration. Int. J. Remote Sensing 13: 17, 3367-3373.

Gower, J.F.R., Doerffer, R. and Borstad, G.A. (1999). Interpretation of the 685 nm peak in water-leaving radiance spectra in terms of fluorescence, absorption and scattering, and its observation by MERIS. Int. J. Remote Sensing 20: 9, 1771–1786.

Gitelson, A., Schalles, J.F. and Hladik, C.M. (2007). Remote chlorophyll-a retrieval in turbid, productive estuaries: Chesapeake Bay case study. Remote Sens. Environ. 109: 464–472.

Gitelson, A. Gurlin, D, Moses, W.J and Barrow T. (2009) A bio-optical algorithm for the remote estimation of the chlorophyll-a concentration in case 2 waters. Environ. Res. Lett. 4: 5pp.

@@@@

104

Chapter 7: Coastal and Inland Water Quality Observing System Assets and Supporting Infrastructure

7.1 IntroductionIn prior chapters, the challenges of making in situ observations in coastal and inland waters and the capabilities and limitations of satellites to observe parameters that can be related to water-quality issues have been discussed. Also, the challenges of linking data (the observations) to information (the conditions in context) and knowledge (transforming information into a format from which decisions can be made and actions can be based) were reviewed. In this chapter, we focus on the assets and supporting infrastructure needed for making routine, fit for purpose observations and producing validated environmental data products relevant to water quality. We consider the needs, targets and goals of managers, decision-makers and other stakeholders and we look at the status and effectiveness of observing systems currently in place. From these endpoints, we identify gaps and make recommendations for future development of satellite-based observing systems that will better serve humanity’s needs for clean water. [put summary here]

7.2 Observing System Assets: Current Status and Future Directions An “observing system” is defined here as the assemblage of assets and processes needed to collect routine, global observations at relevant scales in space and time in order to convert these observations to useful information such as validated environmental data products. That information then can be transformed into knowledge, that is, put into a context from which decisions can be made and actions can be taken. While earth observing systems are in place for a wide range of environmental targets, we will focus on those that are necessary to produce environmental data products from ocean colour satellites.

7.2.1 Satellite assetsIn 1978, one satellite with a visible spectrum radiometer with only a handful of bands demonstrated the dramatic potential of ocean color observations from space (general CZCS citation). To be truly used as the basis for information (and knowledge), however, space

105

observations must be continual, timely and consistent and the products derived from observations must be validated (“ground-truthed”) by in situ observations of the environment. Today, satellites missions are ideally conceived (even if they are not always executed as such) in context with other satellite missions. For example, future missions may be planned to replace aging heritage satellites. Some missions are planned with multiple series of satellites carrying similar suites of sensors with some overlap of active life to assure long term coverage. Multiple and overlapping satellites facilitate agency efforts at transitioning away from a mission-by-mission approach in favor of a mission-agnostic, measurement-based approach. Despite good intentions, potential risks to continuous global satellite ocean colour observations remain due to, for example, the possibilities of mission failures or changes in agency priorities. Current and near-term planned missions with VIS-NIR sensors designed for ocean colour observations are listed in Table 6.1. These include polar-orbiting and geostationary missions. The niches of temporal and spatial scales served by the current near-future suite of sensors show some promise sensor suitability for inland and coastal waters.

Nevertheless, gaps remain to be addressed and these should be considered by agencies as future missions and sensors are planned. As discussed in more detail in IOCCG Report #3, remote-sensing systems for inland and coastal water measurements should meet all the requirements for open ocean measurements plus meet four additional requirements:

● additional spectral bands to discriminate chlorophyll fluorescence, mineral particles and shallow bottom reflectances;

● appropriate channels for atmospheric corrections in turbid and/or shallow waters;● increased dynamic range to avoid sensor saturation due to bright objects typical in

coastal scenes (beaches, for example);● increased number of digitization bits (i.e., to avoid “speckling”; Hu et al 2000).

Future systems should be designed to incorporate these requirements into some part of the overall ocean colour remote-sensing scheme. [synthetic aperature radar; SST/LST; scatterometers;altimeters et al.]

Beyond ocean colour, other satellite-based ocean observations are also routinely accessible. Satellite –based ocean products such as those listed below are or may be used along with ocean colour in generating information about coastal water quality. In cases where the spatial resolution is currently too coarse to be useful near the coasts (e.g., systems originally designed for open ocean), thought should be given as to the potential for development of instruments which might increase spatial resolution.

● Multi-purpose imagery (reflective visible; i.e. “true color”)● Sea (Land) Surface Temperature (infrared)● Ocean Salinity (open ocean, coarse spatial resolution; microwave interferometric

radiometry)● Ocean Surface Winds (from synthetic aperature radar or scatterometry)

106

● Ocean topography and currents (from altimetry)● Ocean wave height and spectrum (from altimetry or synthetic aperature radar)

In addition to ocean observations, satellite-based meteorological parameters are also used in water quality prediction and forecasting efforts. Analogous satellite-based products for land applications may also be useful, especially for inland water quality applications. Because the focus of this IOCCG report is on ocean colour observations, we mention these complementary satellite observations only briefly here. In-depth documentation about these earth environment products and their derivatives is available elsewhere.

7.2.2 In situ assetsIn situ observations for the purpose of understanding the water quality of inland and coastal waters have strengths and weaknesses, as discussed earlier in Chapter 3. Satellite data augment in situ data by increasing spatial and temporal coverage, but satellite data also have inherent limitations. For example, ocean colour remote sensing cannot “see” through clouds nor into the depths of the water column below the optical “surface”. Furthermore, for satellite observations to be useful, they must be validated (“ground-truthed”) with rigorous and consistent in situ measurements. In situ observations are an absolutely essential component of earth observing systems.

Remotely-sensed ocean colour requires high quality, sustained in situ optical monitoring for the purpose of vicarious calibration of instrument gains (see Chapter X for more on vicarious gains). The Marine Optical BuoY (MOBY) was established operationally in 1997 offshore of Hawaii in blue, open ocean water originally for the vicarious validation of SeaWiFs and has been maintained almost continuously (with a break in 2010-2011) over the past 18 years. MOBY has been the primary source for ocean colour satellite vicarious calibration of water leaving radiances. The continuity of MOBY (with the collaboration of the National Institute of Technology and Standards) as an in situ reference remains imperative, even for missions that focus on coastal and inland waters (ocean studies board report; IOCCG report #).

For validation of remotely sensed ocean colour in coastal and inland waters, additional high quality, in situ monitoring is also essential. The Aerosol Robotic Network for Ocean Color (AERONET-OC) program has SeaPRISM sun photometer instruments on fixed platforms in many coastal regions to observe water leaving radiances in coastal waters. The sites are operated by different sponsoring entities but instruments are calibrated with an identical reference source and deployed using standardized protocols, and the data are processed with the same code. Currently, 23 sites are operated in various locations, globally (maybe a Zibordi reference here). However the number and distribution of high quality monitoring sites (like AERONET-OC) for coastal and inland waters needs to be improved and additional locations should be chosen strategically.

107

To facilitate in situ measurements on a variety of time and space scales, and to accommodate more comprehensive observation studies, instruments and technical personnel can be deployed on multiple types of platforms. Some existing in situ observation platforms include:

● moorings (surface and profiling)● bottom tripods● piers/docks● vessels (dedicated for research, commercial vessels such as cruise ships and ferries,

even private “ships of opportunity”)● offshore platforms (research-dedicated and commercial non-dedicated sharing

arrangements)● autonomous platforms (underwater autonomous vehicles (AUVs), floats, gliders, and

drifters) Emerging platforms may involve sensor-webs, citizen science involving private boats, hand-held devices such as smart phones, swarms of miniature neutrally buoyant drifters with inter-communication. In situ validation data is collected primarily by federal and local government agencies with smaller entities, such as Universities and private research groups, also making contributions. International efforts for marine in situ data collection include the Global Ocean Observing System (GOOS) which includes regional alliances and implementation programs aimed at data collection and distribution. These programs include Argo profiling floats, drifting buoys, and hydrography efforts. Regional GOOS alliances (such as U.S. IOOS and EuroGOOS) operate and collect in situ data from local networks of automated sensors, gliders, buoys, and manual sampling campaigns. In situ data collection for freshwater sources is conducted by widespread and disparate organizations. In most developed countries, major inland water monitoring efforts are conducted by federal, regional, local agencies, and public treatment and wastewater discharging facilities. Many developing countries lack the policy mandates, infrastructure, and resources for large scale in situ data networks and rely on universities and private research groups for in situ data collection.

7.2.3 Modeling and data assimilation capabilities (in context of observing system; include this briefly here as prep for further discussion in Chapter 8? At least from OCR data assimilation perspective? It does impact observing system design and implementation of course. I think at least brief introduction/coverage is needed here in this context)

1. Need for complementary and supporting observations OSSEs to guide/strategically deploy observing system assets?

108

7.2.4 Needs, issues and gaps (relative to existing capabilities)

7.2.5 Future directions, solutions and priorities for ocean observing assets● Continuity ● New and improved capabilities ● Capacity building for developing and developed nations

7.3 Constellations and Integration of Observing Assets

7.3.1 Current integration approach across observing system assets● Existing satellite constellations; strengths and weaknesses of existing virtual

constellation approach ● Complementarities and integration between space-based and in situ observing system

assets (not from cal/val perspective) ● Incorporation relative to modeling/data assimilation activities● Opportunities for near-term integration of existing/broader observing system capabilities

7.3.2 Future constellation and integration approach● Space segment: virtual vice truly integrated/functional constellation; cascading down

(up) across relevant spatio-temporal scales● Integration with in situ observing assets● Adaptive sensor webs● Modeling/assimilation for effective integration; support for nowcasts, forecasts and

predictions

7.4 Supporting observing system infrastructure Remote sensing of ocean colour in coastal, inland waters is a non-linear, multivariate problem (IOCCG report 3). Ocean Color in the open ocean is primarily due to the presence of chlorophyll, the pigment responsible for photosynthesis. Coastal and inland waters, however, are often more spectrally complex where three major components are responsible for the water leaving radiance: chlorophyll, mineral particles and dissolved colored matter. To be effective in coastal and inland waters, satellite sensors must have bands that can account for and differentiate among these constituents at the minimum (ioccg report 3). Also, the atmosphere above the water may also be more complex and more variable than over open ocean (i.e., dust, pollution) therefore atmospheric corrections play a substantial role in accurate ocean color retrievals. However, turbid coastal and inland waters violate the “black at NIR” assumption that forms the basis for most open ocean atmospheric correction algorithms. Therefore, satellite sensors for coastal and inland ocean colour need well-behaving, well-calibrated NIR and SWIR

109

bands to account for atmospheric correction needs. See Chapter 5 and IOCCG report 3 for a more detailed, technical discussion about measurements from satellites for inland and coastal waters.

7.4.1 Calibration of satellite observationsBoth solar and lunar on-orbit calibrations are required to produce high quality (accuracy and precision) Level 1a (i.e., sensor data record) top-of-atmosphere radiance data suitable for generating ocean color environmental products. Currently, the best approach to solar on-orbit calibrations utilizes a solar diffuser and a solar diffuser stability monitor. These must be tested and well-characterized before launch. Lunar calibrations require the satellite to maneuver to look at the moon. Planning and commitment to execute these maneuvers on a routine basis is required from flight programs. On-orbit sensor cross calibrations are sometimes possible, however, these introduce other complexities (see xxx) .

After solar and lunar calibration corrections are made to achieve the best TOA radiances, and the best atmospheric correction approaches have been applied, there remains some discrepancy between the water leaving radiances retrieved by satellite and those measured in situ. To account for this difference a gain correction factor must be applied to the satellite data. The method developed for determining this gain is termed a vicarious calibration and this involves making in situ measurements at “stable” sites in the open ocean (rpt #14). MOBY (supported by NOAA in the US) is the radiometric buoy stationed in the Pacific Ocean off Lanai, Hawaii. Since 1996, is has been the primary ocean observatory for vicarious calibration of satellite ocean color sensors [Clark et al., 1997]. The MOBY in situ program supports the international effort to develop a global, multi-year time series of consistently calibrated ocean color data products. BOUSSOLE is an optical buoy site maintained in the Mediterranean (supporting observing system infrastructure calibration of satellite obesrvations - cite IOCCG report 14).

● Needs, Issues and Gaps● Future directions, solutions and priorities● Specific recommendations

7.4.2 Validation of satellite data productsTo validate the environmental data products retrieved from satellites, they must be compared with in situ measurements of optical properties (including atmosphere for atmospheric corrections) and key environmental water quality parameters in multiple coastal regions and inland waterways. The criteria for these “matchups” in open ocean waters will most likely need to be adjusted for coastal and inland waters. Interestingly, higher spatial resolution satellite observations come at a cost of longer (in time) repeat cycles. Therefore, opportunities for “matchups” may be greater in space, but fewer in time. Multiple platforms for making in situ observations across spatial scales should be considered to exploit the benefits of higher spatial resolution satellite data for coastal regions and inland waters.

110

In situ validation observations for coastal and inland waters have particular challenges but also some advantages compared with open ocean validation observations. In considering the advantages, generally, coastal and inland waters are logistically more accessible than offshore blue water. For example: smaller boats may be appropriate as sampling platforms; travel costs may be reduced (i.e., ~local); autonomous sampling systems may be easier to access for maintenance; aircraft overflights can be coordinated from nearby land bases).

Optical complexity is often greater in coastal waters and more variable across inland water bodies than in the open ocean. Variabilities may be associated with season, land use, precipitation patterns, or geological origins, for example. Think of “pristine” mountain lakes versus “polluted” urban rivers or of watersheds impacted by various agricultural practices versus industrial development. For waters with greater variability over a given space and time, more observations will be required to “capture” trends in matchups than would be needed for open ocean. Coastal and inland waters are managed at more local levels than international waters. Local resources (funding, labor, expertise) may be strained as they often must carry a large percentage of an in situ sampling program. Common sampling techniques and common databases must be developed so that local observations are available for validations of satellites with global coverage. Instrumentation for appropriate in situ optical measurements may need to be optimized for each region of interest. In optically shallow waters, the bottom must be characterized in addition to the water column. A greater number of types of observations are needed. Optical observations need to be done across a greater wavelength range (i.e., farther into the infrared).

Comparisons with heritage ocean color products is an additional validation technique which has the benefit of more data and often the ability to validate against climatology (long term trends). The drawback to this approach with respect to inland and coastal waters is that heritage sensors with the bands needed for coastal and inland waters are no longer in service. As time goes on, the assumption of validation against climatology becomes less secure.As we prepare to bring satellite aquatic colour observations to bear on understanding and addressing global water-quality problems, space agencies must understand and commit to the requirements of well-calibrated, high-quality sensor data and the support of well-validated environmental products. This includes committing to lunar maneuvers and vicarious validation sites (i.e., MOBY) in order to calibrate sensors over the long term and facilitate multiple sensors or sensor transitions for climate quality decadal scale trend analyses.

● Needs, Issues and Gaps● Future directions, solutions and priorities● Specific recommendations

7.4.3 Quality monitoring and assessment of data and product utility (metrics of success?)

111

7.4.4 Capacity building for developing and developed nations

Update existing traditional inland /coastal observing protocols to reflect the needs of satellite validationUpdate exiting oceanographic protocols, techniques, instruments to reflect complexities of inland, coastal watersConsolidate , vet protocols, collect data make available for validation and algortimhn development.

7.5 Recommendations to agencies and programmes

7.5.1 CEOS

7.5.2 Global Observing Systems: GOOS, GCOS and GTOS

7.5.3 GODAE OceanView

7.5.4 GEOSS: Blue Planet and Water Cycle Initiatives

7.6 Summary and conclusions

112

Chapter 8: Ground and User SegmentCo-Leads: Steve Groom*, Chris, Carsten, ArnoldContributing authors: Steve Groom + whoever wrote the original page of text

8.1 Introduction Next to the challenging task of building and launching satellites, the success of an Earth observation (EO) mission relies on being able to operate the satellite from the ground and ensure that the data gathered are of good quality and made available readily to users. Furthermore, for the data to be exploited by scientific, management and commercial end-users a suitable processing chain is needed to generate variables of interest, such as chlorophyll-a or suspended particulate matter concentration. These may be produced and delivered in near-real time for operational monitoring applications, like harmful algal blooms, or as long time-series of archived data to look at natural variability, and if the time-series is of sufficient duration, trends. These activities are termed the Ground Segment and may be split into activities undertaken by the space agency that operates the sensor and deliver “upstream” data (the mission or satellite ground segment) and a “downstream” service that takes “upstream” data and produces services and products for the end-users (the user ground segment). This chapter briefly describes the mission ground segment and then focusses on the user ground segment. A number of issues such as data dissemination are common to both types of GS.

8.2 Mission or satellite ground segment

8.2.1 OverviewThe satellite ground segment typically comprises a flight operations and control centre with an antenna to send commands to the satellite, together with at least one ground station to receive data stored on-board the spacecraft as they are 'downlinked' from the satellite as it passes overhead. This may be complemented by direct-broadcast stations that typically only receive data recorded whilst the satellite is within line-of-sight. Direct broadcast can provide rapid access to regional data (ie the geographical coverage of the station) but requires a separate processing chain to that operated by the satellite agency. Facilities for data processing, storage and distribution are also needed.

Last but not least, the quality of the data and the performance of the satellite sensor has to be monitored continuously. For water quality applications based on ocean colour data calibration is crucial, since the water signal is usually only a small percentage of the top-of-atmosphere signal.

113

Most ocean-colour satellite has on-board calibration system, which may use on-board lamps or the moon and sun as calibration targets. The calibration has to be carried out promptly, routinely with regular re-assessments to guarantee the overall performance of the mission. Calibration is achieved by a combination of approaches including: pre-launch characterisation; in-orbit calibration that may involve viewing internal lamps, the sun or moon as reference sources of radiance. However, for ocean colour these approaches have been found insufficient to deliver radiances of the quality needed to retrieve water quality parameters [reference]. Hence, so-called System Vicarious Calibration (SVC) is undertaken which compares the expected top-of-atmosphere signal as detected by the sensor with a simulated system of in-water (Measured) radiance together with an atmospheric model. Since this is a complete system calibration it needs to be performed separately different atmospheric correction models; hence, SVC is undertaken as part of the mission GS or the user GS. NASA undertake SVC on ocean colour sensors using the MOBY system [reference] located off Hawaii in clear oceanic case 2 waters. SVC is also performed in the ESA Ocean Colour CCI project on SeaWiFS, MODIS, MERIS and VIIRS data using a variety of atmospheric corrections and in 2015-16 includes approaches improved for use in case 2 waters. However, it should be recognised that the use of novel atmospheric corrections for inland waters would require an SVC to be performed over MOBY or other reference site: this may be problematic if the AC uses concepts unique to enclosed water bodies like lakes or are optimized for continental aerosols.

8.2.2 Mission ground segmentsIt should first be noted that, arguably, there are no operational mission ground segments dedicated to coastal and, in particular, inland water quality, at least from one agency. An end-to-end demonstrator of a user ground segment was established in support of the GEO 2012-15 work plan within the ChloroGIN environment (see sub-section 4) and the new European Copernicus Land project will use the “Calimnos” processor to produce operational water-quality data and these are discussed in section 8.3.

8.2.2.1 Nasa/gsfc (oceancolor)The NASA GSFC Ocean Biology Processing Group (OBGP http://oceancolor.gsfc.nasa.gov) produce ocean colour data from NASA sensors such as SeaWiFS MODIS and CZCS as well as third-party missions including ESA MERIS (both reduced and full resolution) and NOAA VIIRS. Data are provided at Level 1 and level 2 [note from Steve Groom - I presume that L1, L2 etc are defined earlier ] processed using the “l2gen” atmospheric correction with standard products. These data can provide the starting point to a user ground segment: for example, users interested in standard chl-a algorithms can use the L2 data and re-map to areas of interest. Alternatively, L2 Rrs data can be used with specific regional algorithms to produce user-specific products. Level 3 data are also provided at daily, weekly, monthly, seasonal and annual resolutions, 4 or 9-km (depending on sensor) with a variety of algorithms as standard, evaluation and “test” products. Individual images can be downloaded as PNG or NetCDF files and subsets can be ordered. These L2 and L3 products are focussed on marine applications and no inland water specific products are produced.

114

8.2.2.2 ESA/Sentinel (2,3) collaborative GS ;ESA Sentinel data are freely available for any use academic or commercial and a number of different approaches are envisaged including rolling archives, national collaborative ground segments and multicast using EUMETCast (see below). It is important to note that European Copernicus services will provide products derived from the agency ground segments and as noted below for inland waters this is the responsibility of Copernicus Land. For inland and costal water quality applications the two key instruments are Sentinel 2 MSI and Sentinel 3 OLCI: each has a different ground segment structure. For Sentinel 3, L1 and L2 data will be produced by Eumetsat and will be distributed by GEONETCast as well as via ftp. The Copernicus Marine Environment Monitoring Service (CMEMS) led by Mercator Ocean has an Ocean Colour Thematic Assembly Centre (OCTAC) that will obtain Sentinel 3 data from Eumetsat and produce regional products based on regional algorithms (e.g. for the Mediterranean, Volpe et al 2007) as well as regional and global L3 mapped products, in both near-real time and reprocessing.

The situation with Sentinel 2 MSI is less clear. The sensor is designed primarily for land applications but it is likely to have some WQ observation capability, at least for turbid waters. There is no formal ground segment currently planned, but efforts are in place for user GS to provide products [note from Steve Groom - This is very woolly - Carsten can you comment on this?]

8.2.2.3 NOAA STAR and CoastWatchThe NOAA NESDIS Center for Satellite Applications & Research (STAR) undertakes processing of ocean colour data from MODIS and VIIRS from Level 0 up to level 3 products (see http://www.star.nesdis.noaa.gov/sod/mecb/color/) whilst mapped data for standard user regions of interest are available from NOAA CoastWatch that links through to different regional nodes.

8.2.3 Data distribution mechanismsIn general it has been shown that the easier it is to access to data then the more those data are used. This applies to all levels of users from experts (or operators of user ground segments) easily gaining access to data via ftp to end-users who may want an EXCEL spreadsheet of numbers showing variability with a location or with time. There are a number of dissemination systems that can be described here ranging from traditional FTP, Tape or Web download to more complex and novel web-based services using OGC standards or otherwise, and multicast through such approaches as GEONETCast.

8.2.3.1 FTP, tape and web downloadsData access by file transfer protocol and tape are well established in the community and are not described further herein. Web download is also well established and probably represents the most popular way of obtaining water quality data such as through the NASA ocean colour web site or the ESA Ocean colour climate change initiative site.

115

8.2.3.2 GEONETCast near real time dissemination system (GEO)GEONETCast is a distribution system based on Digital Video Broadcast technology with multicasting environmental data via commercial satellites. It is particularly valuable for institutes or universities with poor internet access due to their location, such as marine institutes located at the coast, or due to limited national infrastructure. For Africa a number of projects have attempted to build capacity amongst marine institute. The European Commission funded DevCoCast and EAMNet project installed low-cost GEONETCast receivers at approximately eight sites across the continent, including one focussed on inland waters on Lake Victoria. The low-cost systems had mixed success, not least due to the exposed locations at the coast. As a legacy of the EAMNet project water quality data at 1-km resolution for user regions of interest covering the entire coast of Africa are produced by PML and UCT and transmitted via GEONETCast. The on-going Monitoring of Environment and Security in Africa programme (MESA: http://rea.au.int/mesa/) has installed further reception capacity based around Regional Implementation Centres including two involved in Marine and Coastal Management operated by the Mauritius Institute for Oceanography (in the Indian Ocean) and the University of Ghana (western Africa).

8.2.3.3 Web-based analysis systemsFor exploitation of water quality data by non-specialist end-users it is important to provide easy-to-use tools, ideally web-based, that support simple analysis of datasets provided by the ground segments. This ensures that EO data have an impact beyond specialist academic communities.

There is a variety ranging from simple tools to analysis systems like NASA GIOVANNI, OGC-based systems.NASA GIOVANNIDescription of GIOVANNI….any takers (I don’t use it)OGC-based systems

The Open Geospatial Consortium (OGC) specifies standards for accessing data by the web. There are a number of standards including Web Mapping Service (WMS) whereby images (such as GIF) are provided, Web Coverage Services (WCS) which serves arrays of data, and Web Feature Service (WFS) that specifies point or other vector data such as ship-tracks or region of interest boundaries . In principle web-based portals using these standards can communicate with other data portals, request and visualise data. Many of the portals have built in analysis capabilities allowing time-series analyses and these offer great potential for the future exploitation of EO data since delivery is only required of the results of the analysis not the full data-set.An example of a web—vis system is the Operational Ecology portal developed in the part EC FP7 OpEc project (see Fig 8.1). This system allows selection of regions of interest either square or arbitrary polygons then time plots of Hovmoller plots of this data. Figure 8.1 shows an extracted time series of chl-a from the ESA ocean colour climate change initiative data set within a box in Lake Michigan. A version of thissystem is being demonstrated in the UK GloboLakes project that will include data analysis in user-specified regions of interest defined by

116

shapefiles, additional analyses such as scatterplots between two variables, and data extraction in time and space, such as along a ship track.

Fig 8.1 Example OGC based web-vis analysis system with time series of monthly chl-a for a box in Lake Michigan extracted from the ESA ocean colour climate change initiative data set

8.3 User Ground SegmentsThis section gives a brief description of a user ground segment to produce operationally water quality observations from satellite. Essentially combines a number of steps described earlier including water-quality algorithms (chapter 6) and delivery from earlier in this chapter.ChloroGIN-Lakes: an end-to-end demonstrator

In support of the Task Sheet for the "GEO 2012-2015 Work Plan: Global Inland and Near-Coastal Water Quality Information System" an end-to-end demonstrator was established at the start of 2012 based on ESA MERIS full resolution data. Unfortunately, due to the demise of Envisat in April 2012 the demonstrator only functioned briefly but it served its purpose to give an overview of how a complete system can be produced based on a combination of Agency and user ground segments.

ChloroGIN-lakes started with MERIS FR data obtained at Plymouth Marine Laboratory in near-real time at L1 and L2 from the ESA rolling archive via ftp. The data products comprised the standard ocean algal_1 and algal_2 chl-a, yellow substance and SPM: it was accepted that

117

these were not optimised for inland water bodes but the aim of the exercise was to show that an end to end demonstrator was possible. The level 1 data were also obtained and were processed to level 2 through SeaDAS showing that local processing with potentially lake-specific algorithms was possible.

The data were then mapped to individual regions of interest (e.g. see Figure 8.2 below) using the same system as used at PML for UK and international projects such as Copernicus Marine Environment Monitoring Service. Each region was stored in a database and data were automatically processed when covering all or part of the region of interest.Finally, the images were made accessible via the standard ChloroGIN portal where users can “click” on an AOI, select a date and then view individual scenes using the regional data producers’ web site (in this case the PML MultiView web viewer).

Fig 8.2: a) ChloroGIN-Lakes front-end showing regions processed; b) example MERIS algal_2 (chl-a) image for Lake Victoria, 16 February 2012

Subsequent to the failure of Envisat, data from MODIS were processed at 500m for various lakes; however, the quality of these data in terms of spatial resolution and number of bands was inferior to MERIS but nevertheless demonstrated the system.

CaLimnos: a future approach to inland water quality data production

Calimnos is an inland water-quality data production framework based on the ESA funded Diversity II processing chain, developed by Brockmann Consult, together with updates developed at Plymouth Marine Laboratory in the context of the UK GloboLakes project. Initially it was designed to process archived ESA Envisat MERIS data but will also produce near-real time and archived products from Sentinel 3 OLCI data. The current instance (see Fig 8.3) applies a wide variety of atmospheric correction and in water retrieval algorithms / processors to generate a range of products that can then be aggregated

118

Calimnos is being updated to include the outputs of research on-going in GloboLakes notably the ensemble approach proposed in GloboLakes to apply the most appropriate combination of processors for an individual lake.

The system has also been selected for operational production of water quality data in the Copernicus Land project for an initial period of 2016 – 2019 and will be run on the CEMS facility at Harwell, UK (see chapter 9). The products will likely comprise 10-day composites (decads) of trophic status, suspended particulate matter and reflectance both from archived data and in “near-real time”.

Figure 8.3. on Calimnos showing Envisat MERIS data flow from satellite through to final products

8.4 SummaryThe success of a satellite mission is critically dependent upon the ground segment activities undertaken by the space agency and if separate an operational data production centre. Mission ground segments exist for oceans and some coastal regions but to date there has been no dedicated inland and near-coastal ground segment. In projects such as ESA’s Diversity II and the UK GloboLakes methods have been developed to process EO data from large numbers of lakes worldwide using a variety of algorithms and so produce a user-ground segment. With the advent, notably, of the European Sentinel missions including Sentinel 2 and 3 along with the archive of MERIS it is now becoming possible to contemplate an operational inland and near coastal water quality ground segment. In Europe the Copernicus Land project will produce, we believe, the first operational service for inland water quality products.

119

References· Volpe et al 2007· NASA SVC· ESA OC CCI SVC

120

Chapter 9: New Concepts (Virtual Laboratory)(document version 5, 12.01.2015)

Carsten Brockmann, Steve Groom, Paul DiGiacomo, Blake Schaeffer

9.1 IntroductionWith the advent of new missions providing Earth Observation (EO) data at high and very high spatial resolution over a large swath (e.g. Sentinel 2 multi-spectral instrument) or in hyperspectral mode (e.g. Enmap [note from Steve Groom - which platform is this carried on?]) the amount of data will increase by an order of magnitude. This expanding operational capability of global monitoring from space, combined with data from long-term EO archives (e.g. Landsat, ERS, Envisat), in-situ networks and models, will provide users with unprecedented insight into how our oceans, coasts and inland waters interact with the atmosphere, land and ice, and are influenced by anthropogenic pressure. While the availability of the growing volume of environmental data from space represents a unique opportunity for science and applications, it also poses a major challenge to achieve its full potential in terms of data exploitation. Firstly, because the emergence of large volumes of data (Petabytes era [Steve Groom comment- maybe “petrabytes per year” or month?]) raises new issues in terms of discovery, access, exploitation, and visualization of “Big Data”, with profound implications on how users do “data-intensive” Earth Science [1]. Secondly, because the inherent growing diversity and complexity of data and users, whereby different communities - having different needs, methods, languages and protocols – need to cooperate to make sense of a wealth of data of different nature (e.g. EO, in-situ, model), structure, format and error budgets. Evolution in information technology and the consequent shifts in user behaviour and expectations provide new opportunities to provide more significant support to EO data exploitation. Examples of such new concepts can be found in Europe (ESA Thematic Exploitation Platform[2]), US (Virtual Laboratory), Australia (Data Cube[3], Biodiversity & Climate Change Virtual Laboratory (BCCVL) [4]), or at international level (Geohazards Supersite [5])

This chapter will discuss the challenges and opportunities of making available to the user communities the large amount of EO data in combination with in-situ data and models. The basic idea is to provide a platform to users with data and tools, and to provide means to bring users’ tools and processing software to the data, instead of pre-defined and processed Level 2 and Level 3 data. This discussion will be complemented by examples of existing and planned platform and virtual laboratories.

121

9.2 “Big data” from Space

9.2.1 Historic archives and future data volumesSpace-borne Earth Observation has been providing data since the early 1970’s data when the MSS sensor on Landsat 1 was the first to deliver systematic observation of the Earth’s surface, including coastal and inland waters, in 4 spectral bands. However, on-board data storage and data transmission capabilities were very limited. A significant increase in data volume per acquired scene took place with the Thematic Mapper on Landsat 4 in 1982, shortly followed by Landsat 5 in 1984, which provided more spectral bands (6, resp.[SG - what is resp?] 7) and higher spatial resolution. The 160MB per scene required three 9’’ [SG2] tapes and dedicated computer hardware to process them. Today such data volumes look small; however, due to the 25 years of Landsat 5 operations, the total volume of products available at the USGS is over 3.6 million scenes corresponding to 576 TB (status 2008, http://earthobservatory.nasa.gov/Features/LandsatBigData/). The USGS archive has been complemented recently by scenes acquired in Kiruna, Sweden, by the European Space Agency. This complementary archive adds another 450TB of data.

With every new generation of satellite missions the generated data volumes increase significantly. ESA’s first series of EO satellites, ERS, produced about 100TB during their 10 year mission life time. This corresponds to the amount of data produced by the subsequent mission ENVISAT in only 1 year. The ENVISAT archive now amounts to 1.2PB, not counting derived products e.g. for climate studies.

The Sentinel 1-2-3 A satellites will provide a data stream of 500MB/s (Bagellini, 2015) equal to 43TB / day which need to be processed and kept under stewardship. The resulting core user products are estimated as 4TB/day (ESA, 2015). It should be noted that ESA will operate two satellites (A and B series) in parallel for most of the time, which will duplicate the data volumes.

Of course one needs to differentiate between the data acquired by a mission and the data actually required for inland water applications, which is only a portion of the complete data archive. Verpoorter et al (2014) estimated the number of lakes that can be monitored with different spatial resolutions: for example, with a ground sampling distance of 15m (L8, S2) lakes of size > 0.002 km² can be mapped at 100% of their size. There are about 120 Million of these lakes worldwide.

There seems to be a crude rule that the data generated by one satellite mission during a typical 10-year lifetime is produced by the next generation within 1 year. [SG-four generate/generations in a short sentence!] The good news of this rule is that the storage (and processing time) of all historic missions is never a problem because these data add a comparable small amount to those data generated by the current mission.

122

9.1.2 Challenges for data access and data processingI have started to rechereche on this. Any references or information on data volumes you are aware of are welcome! Steve, Paul. See reference below (Lawrence et al 2013)

9.3 Moving software to the data (Carsten)

9.3.1 Data harmonization (“Data Cube”)

9.3.2 Exploratory tools (e.g. Product Feature Extraction[1],[2])

9.3.3 Service models

9.3.3.1 Infrastructure as a Service (IaaS)

9.3.3.2 Software as a Service (SaaS)

9.3.4 Community tools

9.3.5 Mobile TechnologyAs of 2014, 96.3% of people had some form of mobile cellular subscription (World Bank, 2016). Mobile technology with internet connectivity was 47% of the population in 2015 and quickly growing, with third-generation networks (3G) coverage available to 69% of the world (Telecommunication Development Bureau, 2015). Nearly ubiquitous global access to mobile technology provides new opportunities both for citizen scientists collecting data (Dickinson et al 2012; Newman et al, 2012) and as a new platform for environmental data dissemination to stakeholders (Jonoski et al 2013; Mattas-Curry et al, 2015). Interactive web service portals using Open Geospatial Consortium formats, combined with advances in cloud-based infrastructure will allow for coordinated data sharing through online applications (Mouw et al, 2015).

9.4 Examples

9.4.1 The Australian Geoscience Data CubeThe effort and cost required to convert satellite EO data into meaningful geophysical variables has prevented the systematic analysis of all available observations. To overcome these problems, in 2014, Geoscience Australia, CSIRO and the NCI [SG1] established the Australian Geoscience Data Cube, building on earlier work of Geoscience Australia and expanding it to include additional EO and other gridded data collections (e.g. MODIS, DEM) in order to expand the range of integrated data analysis capabilities that were available. They utilise an integrated

123

High Performance Computing and Data environment to rapidly process, restructure and analyse the Australian Landsat data archive. In this approach, the EO data are assigned to a common grid framework that spans the full geospatial and temporal extent of the observations – the EO Data Cube. This approach is pixel-based and incorporates geometric and spectral calibration and quality assurance of each Earth surface reflectance measurement. The utility of the approach has been demonstrated with rapid time-series mapping of surface water across the entire Australian continent using 27 years of continuous, 25 m resolution observations. The preliminary analysis of the Landsat archive has proven how the EO Data Cube can effectively liberate high-resolution EO data from their complex sensor-specific data structures and revolutionise the ability to measure environmental change.

The development of the software behind the Data Cube is done transparently to the users. The current implementation of Australian Geoscience Data Cube v2 concept is intended as a working prototype for a cohesive, sustainable framework for large-scale multidimensional data management for geoscientific data. This public development release is intended to foster broader collaboration on the design and implementation. It is not intended for operational use. The operational version of the original AGDC code is available from github.com/GeoscienceAustralia

9.4.2.The ESA Thematic Exploitation PlatformIn the traditional workflow for the analysis ofEO data, users download the data to their local site and then process, using their available software and computing resources. With the increasing volume of data available from missions such as Sentinel, and the resulting need for powerful computing resources for processing, the existing methods of working are inefficient and restrict the use of EO data. ESA’s Thematic Exploitation Platform (TEP) concept aims to provide a working environment where users can access algorithms and data remotely, providing them with computing resources and tools that they might not otherwise have, and avoiding the need to download and store large volumes of data. This new way of working will encourage wider exploitation of EO data.

As an example, the Urban TEP is briefly described here. Although its focus seems to be very different from the inland water perspective, thematic and technical requirement are quite similar. Sustainable urban and rural development require knowledge about status, properties, cross-linking and dynamics of settlements and their hinterland. Key elements are effective and efficient instruments for the observation, analysis and control of the built environment. In this context the TEP Urban platform aims at opening up new opportunities to enable the creation and safeguarding of liveable cities by exploring:

● Unique EO capabilities in Europe;● Big data perspective;● High-level IT-infrastructures;● Massive processing power;● Vast expert knowledge;● New media and ways of communication;● Increasing connectivity and networks.

124

By exploiting these opportunities TEP Urban will contribute to initiate step changes [SG2] regarding:

● Remote processing (users-to-the-data);● Enabling technology (large-scale exploitation, timeliness);● Distribution of expertise (increase assets);● Sharing data and technology (innovation, benchmarking);● Open, integrative, participatory, collaborative concepts (community stimulation,

outreach).

With these objectives and perspective, TEP Urban is finally expected to function as:● Enabling technology,

○ Technical: Linking big data, IT-infrastructures, processing and analysis solutions;○ Thematic: Provision of standardised, new, and tailored products and services○ Societal: Improving access to and distribution of data, methods and information.

● Instrument to gain of knowledge on the urban system:○ Contribution to close gaps in Earth system science;○ Increased efficiency, effectiveness and sustainability of functions and services in

policy, planning, economy, and science).● Market place of ideas and driver of innovation;● Access point for and network of stakeholders and experts;● Seed point for the animation of new user communities outside EO/geo-sector.

The TEP Urban technical concept is based on a generic, modular, multi-purpose design facilitating maximum flexibility with respect to the adaptation to and integration of user requirements, application scenarios, processing and analysis technologies, and IT infrastructures. A schematic view of the basic TEP Urban elements is given in the figure below. Considering the thematic and technical objectives of TEP Urban, the key technical functionalities of the architecture are:

● Web-based, modular platform employing generic, distributed high-level computing infrastructures (Platform as a Service – PaaS);

● High-performance access to thematic data (Information as a Service – InaaS);● State-of-the art pre-processing, analysis, and visualisation techniques (Software as a

Service – SaaS);● Customising functionalities for transfer of algorithms/products into services;● Functionalities to support networking, communication, and dissemination activities.

The TEP Urban platform is not designed to provide direct access or distribute EOdata.

For a sustainable development and the certainty to be used after the lifecycle of the project, TEP Urban follows an Open Source Strategy:

● The project is committed to Open Source components and modules;● It will have an online presence at https://github.com/urban-tep TEP-QuickWin

components are already Open Source and some will be reused and evolved;

125

· Components and enhancements developed will be released as Open Source;

Interfaces between our integrated software APIs are based on open standard specifications (OCCI, OGC).

9.4.3 Climate, Environment and Monitoring from Space (CEMS)The CEMS facility, located at Harwell, UK, comprises academic and commercial components with an aim of transferring academic applications to the private sector; the latter under the auspices of the UK Satellite Applications Catapult (SAC). Academic CEMS is integrated with the JASMIN[1] infrastructure (Bennett pers. comm.) funded by the Natural Environment Research Council (NERC) and the UK Space Agency (UKSA) and managed by the Science and Technology Facilities Council (STFC) Centre for Environmental Data Archival (CEDA; www.ceda.ac.uk). The physical infrastructure consists of a parallel-NFS data cluster and compute nodes, which currently has 4000 cores for data processing, 17 PB of high-performance disk storage and is accompanied by tape libraries. The facility provides services to the UK and European climate and Earth system science communities to support data analysis and processing, storage, batch computing, hosted computing and cloud computing. Users can provide a pre-configured Virtual Machine based on CEMS-provided templates and run this in parallel on the CEMS infrastructure, benefitting from the high-speed storage and extensible computing resources.

Current water quality applications on JASMIN-CEMS include production of lake water surface temperature (LSWT) for approximately 1000 lakes worldwide by the University of Reading, led by Prof Chris Merchant, based on AATSR data from Envisat. This work is undertaken as part of the UK GloboLakes project (see chapter 8). As part of the European Copernicus Land service processing of lake water quality data is also planned at CEMS using elements of the PML – Brockmann Consult “Calimnos” processor (see chapter 8) via virtual machines. This will initially be based on Envisat MERIS but will also use Sentinel 3 OLCI and possibly Sentinel 2 MSI data taking advantage of access to the data through the UK Collaborative Ground Segment. It is also expected that Copernicus Land will produce LSWT on CEMS.

JASMIN is connected to the commercial CEMS infrastructure in the SAC via a dedicated link, and is connected via multiple high performance network links to the wider community including to PML where a local supercomputer undertakes water quality data processing. JASMIN/CEMS also provides co-location with a number of other important climate and environmental data sets, and the CEDA archives for long-term data curation. (V Bennett, pers. Comm.)

9.4.4 Something from US Paul

126

9.5 Summary and Conclusions: Benefits and drawback from inland water points of viewKey points/messages from Steve:The key points and messages to me:· first, a system should have access to all the data that the user may wish to process, so for the inland water community this would be MERIS at FR, maybe Landsat, MODIS and SeaWiFS from the archive and going forward should be Sentinel 3 at 300m (OLCI) and 1-km (SLSTR) and Sentinel 2.· Second, there should be sufficient processing available to do long runs of data in a reasonable amount of time albeit with checks to ensure that users do not inadvertently waste a huge amount of CPU cycles.· Third, there should be richness in data processing modules that should be optimized for the environment on which they run. Richness here refers to a range of atmospheric correction algorithms and approaches, and in water algorithms and models.· Fourth, there should be a range of tools available to view, analyse and further process samples of data.· Fifth there should be appropriate download tools to obtain the final results for local analysis, processing and incorporation into manuscripts and reports.· Next, there should be appropriate archiving for results, though these might be highly specialized; on the other hand a complete run of, say, the Great Lakes, would undoubtedly be of interest to many users if only so that alternative approaches can be followed and avoiding repeats.· It would also be useful if there were easy-to-use mechanisms for agencies and other end-users of water quality information to access the data in an abstracted form.· Key points/messages from Paul:Key points/messages from Carsten:@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@

References [1] Hey, T., S. Tansley, and K. Tolle (Eds.), The Fourth Paradigm: Data-Intensive Scientific Discovery, 286 pp., Microsoft Res., Redmond, Wash, 2009. http://research.microsoft.com/en-us/collaboration/fourthparadigm/4th_paradigm_book_complete_lr.pdf

[2] F. Bitto et al., 2014, The Earth Observation Exploitation Platform (EOXP) Prototype, Technical Note, March 2014. https://earth.esa.int/documents/10174/1157689/EOXP-ACS-ESA-IIM

[3] http://nci.org.au/virtual-laboratories/earth-observation/australian-geoscience-data-cube/

[4] http://nci.org.au/virtual-laboratories/earth-observation/, http://www.bccvl.org.au/

127

[5] http://www.helix-nebula.eu/usecases/esa-ssep-use-case

[6] N. Fomferra, F. Bovolo, L. Bruzzone, O. Danne, Olaf, B. Demir, M. Iapaolo, Michele, R. Jha, J. Lu, P.G. Marchetti, R. Quast, L. Veci, "ESA LTDP PFA - Product Feature Extraction and Analysis" in Living Planet Symposium 2013, ESA: ESA, 2013. Proceedings of: Edinburgh, United Kingdom, Edinburgh, United Kingdom, 09-13 September 2013

[7] Lorenzo Bruzzone, Begüm Demir, Francesca Bovolo, Carsten Brockmann, Norman Fomferra, Michele Iapaolo, Rajesh Jha, Jun Lu, Ralf Quast, Kerstin Stelzer, Luis Veci : ANALYZING AND RETRIEVING REMOTE SENSING IMAGES FROM LARGE DATA ARCHIVES. In : Proceedings of ESA-EUSC-JRC 9th conference on Image Information Mining – The Sentinel Era. March 2014, pp . https://www.google.de/url?sa=t&rct=j&q=&esrc=s&source=web&cd=9&ved=0CFkQFjAI&url=http%3A%2F%2Fpublications.jrc.ec.europa.eu%2Frepository%2Fbitstream%2F111111111%2F30757%2F1%2Flb-na-26548-en-n_soille.pdf&ei=I0v-U9jMBcTqaPzAgcAO&usg=AFQjCNGafNLqkywerkHUbwl5sTt3p8BoNA&sig2=sDcBilOz2ZJ0sU0ZxSTlOw&cad=rja

[8] Lawrence B.N., V.L. Bennett, J. Churchill, M. Juckes, P. Kershaw, S. Pascoe, S. Pepler, M. Pritchard and A. Stephens. Storing and manipulating environmental big data with JASMIN, in: IEEE International Conference on Big Data, 2013, doi:10.1109/BigData.2013.6691556

Dickson, J. L., J. Shirk, D. Bonter, R. Bonney, R. L. Crain, J. Martin, T. Phillips, & K. Purcell. 2012. The current state of citizen science as a tool for ecological research and public engagement. Frontiers in Ecology and the Environment. 10(6):291-297.

Jonoski, A., A. Almoradie, K. Kahn, I. Popescu, & S. J. van Andel. 2013. Journal of Hydroinformatics. 15.4: 1137-1149.

Mattas-Curry, L., B. A. Schaeffer, R. N. Conmy, & D. J. Keith. 2015. A satellite perspective to monitor water quality using your mobile phone. Lake Line. 35(2): 28-31.

Mouw, C. B., S. Greb, D. Aurin, P. M. DiGiacomo, Z. Lee, M. Twardowski, C. Binding, C. Hu, R. Ma, T. Moore, W. Moses, & S. E. Craig. 2015. Aquatic color radiometry remote sensing of coastal and inland waters: Challenges and recommendations for future satellite missions. Remote Sensing of Environment. 160: 15-30.

Newman, G., A. Wiggins, A. Crall, E. Graham, S. Newman, K. Crowston. 2012. The future of citizen science: emerging technologies and shifting paradigms. Frontiers in Ecology and the Environment. 10(6):298-304.

Telecommunication Development Bureau. 2015. ICT facts and figures: The world in 2015. Geneva, Switzerland.

128

World Bank. 2016. Mobile cellular subscriptions. http://data.worldbank.org/indicator/IT.CEL.SETS.P2/countries?display=graph. Accessed January 14, 2016.

129

Chapter 10: Data Exploitation Programmes and User SupportCarsten Brockmann, Paul Digiacomo, Rick Stumpf, Blake Schaeffer

10.1 IntroductionThe Space Agencies design, build, launch and operate earth observation satellites based on requirements of the scientific / technical user communities, and to some extend driven by technology push. After successful launch of a space craft and during the operational phase of the mission, further federation and support of the user community through data exploitation programs is important not only for the users to obtain funding for data exploitation, but at least equally important, it is on the critical path from Space Agencies’ point of view for gathering feedback on data quality, to stimulate evolution of algorithms and to drive the development of new applications. In this sense, data exploitation programs and user support close the loop and contribute to the requirements for improved or new earth observation missions.

10.2 User support programsThe aim of a user support program is to stimulate the data exploitation, specifically of new mission concepts, as well as training users in efficient data access and exploitation of the information content of the raw data. New algorithmic ideas can be tested, or the transfer of applications into an operational or commercial market can be initiated. Users can come from the scientific community, but can and should be also non-Earth Observation end-users, such as public administrations or private companies providing any kind of service.

All space agencies have user support programs but with different emphasis and different structure. For example, within its Earth Observation Directorate the European Space Agency has 4 dedicated programs for “Working with Users – ESA EO Exploitation Programs”. They follow a logic from basic research through application development up to developing operational services and markets:

The scientific community is served through two programmatic elements. The “Support to Science Element” (STSE) aims at providing scientific support for both future and on-going missions, by taking a proactive role in the formulation of new mission concepts and of the related scientific agenda, by offering a multi-mission support to the scientific use of ESA Earth Observation missions data and to the promotion of the achieved results. Under STSE small R&D studies are funded for testing novel algorithmic ideas, as well as programs to support young researchers with funds and research stay at ESA. The “Scientific Exploitation of Operational Missions” (SEOM) has the prime objective to federate, support and expand the

130

large international research community. It aims to further strengthen the strong international role of European Earth Observation research community by enabling them to extensively exploit observations from future European operational EO missions. SEOM enables the science community to address many new avenues of scientific research that will be opened by free and open access to data from operational EO missions. SEOM funds R&D studies to develop new algorithms and products not included in the operational products, provides tools for data exploitation and conducts thematic workshops (e.g. ESA Living Planet Symposium or the Sentinel 2 Science Workshop).

Institutional users (like national and continental met services, or monitoring agencies), thematic user communities (e.g. GEO communities of practice) or value adding companies interested in developing new applications (center yellow box in Figure 1) are supported through the “Data User Element” (DUE). The mission of this element is to favor the establishment of a long-term relationship between the User communities and Earth Observation. DUE funds projects that implement algorithms, processing systems and generate products for specific user communities. For example, the DIVERSITY II project which was the first project providing a global dataset of inland water products processed with a consistent processing chain and service the EO as well as Biodiversity user communities, has been funded through the DUE element. Thematic User Consultation workshops and a user consultation process is always a preparatory activity in order to shape the Statement of Work of a DUE project.

Finally, service industry and its clients are served through the “Value Adding Element” (VAE). This program provides support to the geo-spatial information industry and its customers to develop and grow the prospects of EO services being used in business and organizations for their operations.A cross cutting recent activity are the development of Thematic Exploitation Platforms.Another concept is employed by Jaxa which supports users through R&D opportunities through their “Institute of Space and Aeronautical Science”, including a Young Fellowship program, or dedicated development projects. Under the “Utilization for Earth and Environment Observation” heading data exploitation support tools have been developed and made available to users, such as the G-Portal providing access to data of spaceborne sensors that Japan Aerospace Exploration Agency (JAXA) has developed/involved. Under Conquering the Global Warming Jaxa supports scientists to contribute to the GEO Global Carbon project.All these programs and projects are very useful in general terms, helping users to familiarize with the data of specific Space Agencies and to provide feedback to the Space Agencies. However, no program is currently in place to address inland waters. Among all current and near-future missions, Landsat 8 and Sentinel 2 have the greatest potential to significantly advance our understanding of inland waters and programs should be established, coordinated across the individual space agencies.

10.3 Thematic workshopsThematic Workshops organized by space agencies are very beneficial to foster exchange between scientists, as well as between scientists and space agencies. Various kind of themes

131

can be addressed by such workshop, ranging from broad scope such as climate change (e.g. “Climate Research and Earth Observations from Space”, 2014, Eumetsat) to focused one like the International Ocean Colour Science Meeting, co-organised by ESA, NASA, CNES, NOAA and Eumetsat down to very specific aspects like “Mapping Water Bodies from Space” (ESA).These workshops provide a platform for users to demonstrate the value of Earth Observation data which is important for the Space Agencies to justify their investment against their funding bodies and to support requests for new funds for new missions or data exploitation programs. Thematic workshops are also an important instrument to shape specific application programs or projects. Such meetings should be conducted before a specific work program is drafted, in order to optimally shape the program to user needs.

User consultation meetings should be conducted as part of a user support program for inland waters. Such user consultation meetings should be linked and coordinated with IOCCG as well as the GEO Water Quality Community of Practice.

10.4 Dedicated R&D studiesThere are numerous general funding opportunities for research and development in all countries. However, these usually define a broad perimeter of potential activities, and researchers propose project with specific objectives and aims as defined by them. This is a good mechanism which gives maximum room for creativity. However, with this it is not possible to address very specific problems, which have already been identified as a methodological gap or missing understanding of bio-geophysical processes.

In the case of inland water quality a lot of scientific and methodological progress has been achieved over the last years. The previous chapters summarize the state of the art and identify gaps that need to be addressed in the near future. To recall here a few of them:

● The atmospheric correction above lakes is currently the largest obstacle and should be addressed with highest priority

● Specific inherent optical properties of lakes vary largely on a global scale. Methodologies are available to model the light field depending on SIOPs, to invert the satellite signal if at least a potential set of SIOPs is known, but currently it has not been achieved to generate a global water quality product and being able to cope with an unknown set of SIOPs.

● Deriving other products than Chlorophyll content, TSM, yellow substance and cyanobacteria; in particular products that require the link of EO data with a lake ecosystem model.

Space Agencies often have programs in place to contract dedicated R&D studies to address such problems. The tender precisely defines the required research to be undertaken, and deliverables in form of scientific documentation and prototype code, to be used then in their ground segments for operational processing, possibly after a re-engineering of the scientific code into operational robust software. It is important that the scientists in charge of the development stay responsible for the operational version, and to ensure a long term

132

maintenance of the algorithm by the scientists in order to benefit from further scientific developments. This is particularly relevant for inland water quality where a lot of feedback will be gained during the coming years from Landsat 8 and Sentinel 2, which will lead to improvements of existing algorithms.

Each R&D study should include proper validation of the proposed method. This will be addressed in the next section.

10.5 ValidationValidation is an essential element of all earth observation algorithm development as well as operational product quality control. The expert knowledge is available in the user community and should be used; in addition, participating in validation contributes significantly to the acceptance of the products by the user community.

Validation takes place at two different levels:● Algorithm validation is part of the algorithm development. It confirms the applicability of

an algorithm within its defined scope. Algorithm validation is often documented in the Algorithm Theoretical Baseline Document. In many cases algorithms are developed before launch of a mission, and algorithm validation is performed using simulated data, ground measurements and data from other sensors, including airborne missions and precursor missions

● Product validation uses the product generated by the space mission using the ground segment algorithms. Typically, it requires reference measurements (in-situ data) of high quality, nowadays called fiducial reference measurements, and a well-defined protocol on the comparison method of the satellite and the reference measurement.

Reference data for inland waters should include radiometric data, concentrations of water constituents and SIOPs characterising the water type. Today, harmonized protocols as well as intercalibration of instruments used for lake radiometry is missing or at least not covering the global expert community.

In order to obtain best quality algorithms and to maintain best product quality throughout a mission lifetime, the Space Agencies should include in their user support programs measures to

● Include sufficient funding for algorithm validation as part of the R&D studies (see above)● Establish and fund a minimum set of reference sites equipped with suitable instruments

to provide fiducial reference measurements, and maintain the scientific expertise to operate the instruments and evaluate the data

10.6 Summary of recommendations from inland water point of view

● User support programmes are of critical importance for users and space agencies

133

○ No programme is currently in place to address inland waters. Among all current and near-future missions, Landsat 8 and Sentinel 2 have the greatest potential to significantly advance our understanding of inland waters and programmes should be established, coordinated across ESA and NOAA (NASA?)

● Thematic workshops are very beneficial to foster exchange between scientists, and between scientists and space agencies

○ User consultation meetings should be conducted as part of the user support programme for inland waters

● Dedicated R&D studies are important to close algorithmic gaps and to improve quality of products

○ The atmospheric correction above lakes is currently the largest obstacle and should be addressed with highest priority.

○ Other themes that need to be studies with simulated and real data, and which are not covered by any other programme include: … (SIOPs?)

● Validation is an essential element of earth observation and should be better included in the user programmes.

○ In-situ data for inland waters should include radiometric data, concentrations of water constituents and SIOPs characterising the water type

○ Harmonised protocols or intercalibration is currently missing

134