Managing and communicating

uncertainty in geospatial web service workflows

Richard Jones, Dan Cornford, Lucy Bastin, Matthew Williams

Computer Science, Aston University, Birmingham, United Kingdom

Workshop on Workflows for Earth Observation Systems, Nottingham

The UncertWeb concept

• The “model web”.• When chaining services of limited or unknown

quality, uncertainty must be accounted for if rational decisions are to be made.

ECMWFensembles

GEMSensembles

Emissions data

Air quality forecastmodel

Modelparameters

Uncertaintyenabled air quality

forecast

Air qualitysensor network

Short range airquality forecasts

Meteorologicalsensor network

ECMWF model

GEMS models

Near real-time airquality nowcasts

Land use data

Albatross individualactivity simulator

Historical activitydata

Traffic model

Individualexposure

assessment

Emissions model

GISinfrastructure

data

Demographicdata

“UncertWeb develops mechanisms, standards, tools and test-beds for accountable uncertainty propagation in web service chains.”

Uncertainty quantification

• Even more important when considering service chains.

• Quantification achieved by:– Standards for representing and communicating

uncertainty.– Adding uncertain inputs and outputs to existing

services.– Developing tools to assist users with uncertainty

management.• Computational issues will be non-trivial too!

UncertWeb and UncertML

• Standards for coupling models under uncertainty.– Develops UncertML to provide complete probabilistic

model for uncertainty, and consider other (e.g. Fuzzy, Bayes Linear) representations.

– Provide an API for using UncertML.– Take UncertML through the standardisation process

(where? OGC?, W3C?, IETF?).

• Still some thinking to do as to whether UncertML should deal with encoding and / or be a controlled vocabulary.

Chaining UncertWeb services

• Chaining and discovery services under uncertainty.– Extend existing interoperable services to permit

their use in the uncertainty-enabled model web.– Implement a framework for uncertainty-enabled

model web services.– Develop uncertainty-enabled OGC Web Services.

• One concern is whether to stick to OGC services or move to standard W3C WS.

A prototype UncertWeb chain

• Convert a pressure measurement to sea level:

• Uncertainty is present in the elevation samples, the interpolation, and parameters in the conversion calculation.

• Monte Carlo used for uncertainty propagation.

A prototype UncertWeb chain

A prototype UncertWeb chain

• Output screenshots taken from web demo.• Available at: http://uncertws.aston.ac.uk/client/

Uncertainty enabled services

• All intermediate services conform to the WPS standard.• Current OWS do not explicitly support uncertainty.

– Some recognition in metadata: quality.

• Uncertainty types encoded using UncertML, then used as WPS inputs and outputs.

• UncertWeb will develop profiles of WPS, SOS, CSW and WCS and produce implementations of these that can work with uncertain inputs and outputs.– Intention is to restrict what can be communicated to make it

easier to interoperate within uncertainty-enabled services.

Introduction to BPEL

• Can specify peer-to-peer interaction between a number of services.• Interactions for a workflow are specified in an XML-based script.• Script deployed remotely on an orchestration engine.• The chain is then exposed and can be consumed in a stateless manner.

BPEL advantages

• Orchestration takes place on a server, data does not need to pass through client.

• Deployed workflow available as a service.• Workflow complexity is hidden to the

consumer.• Reproducibility supported as BPEL scripts

define interactions completely.

Chaining under uncertainty?

• Issues raised by prototype:– WPS specification is vague on how WSDL and SOAP

can be integrated into services, making the use of WPS with BPEL non-trivial.

• Can wrap standard requests with SOAP.• Generic WSDL document can be used.

– Chain creation difficult as there is a need for predefined execute documents.

– Exceptions are difficult to catch.– Services accepting different types (of uncertainty)

require an intermediate translation service.– Methods for propagation will be complex.

Other workflow tools

• In addition to BPEL, other tools are available for creating and orchestrating scientific workflows, with Taverna and Kepler being the most widely used.

• These tools differ from BPEL, as chains are orchestrated locally, and they provide a rich graphical user interface.

Other workflow tools

• However, they still suffer from problems due to the lack of detail on support for web service standards in the WPS specification.

• Suitability for the scenario:– Chains aren’t available as services.– Orchestration mainly done locally.– Generic WSDL documents for WPS make use

difficult.

Applying UncertWeb

• Mechanisms applied to several use cases.• Use cases for uncertainty propagation.

– Biodiversity and climate change (links to EuroGEOSS).– Land-use response to climatic and economic change.– Short term uncertainty-enabled forecasts for local air

quality.– Individual activity in the environment.

ECMWFensembles

GEMSensembles

Emissions data

Air quality forecastmodel

Modelparameters

Uncertaintyenabled air quality

forecast

Air qualitysensor network

Short range airquality forecasts

Meteorologicalsensor network

ECMWF model

GEMS models

Real-time airquality nowcasts

The future of UncertWeb

• UncertWeb only started 4 months ago.– Already have prototypes that demonstrate what

can be achieved.• Many challenges still to face.

– Semantics, performance, standards (UncertML), user acceptance...

• Enhanced WPS with automatic generation of WSDL with fully-specified operations and message formats.– Ease service consumption.– Support BPEL graphical clients, Taverna and

Kepler.

Summary

• BPEL was deemed suitable to orchestrate this uncertainty-enabled workflow.– Next challenge is taking these mechanisms to

complex models and making workflow creation more accessible.

• UncertWeb will develop standards, profiles and implementations to create the uncertainty enabled model web.

The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement n° [248488].