SCALE ISSUES INREMOTE SENSING

SCALE ISSUES INREMOTE SENSING

Edited by

QIHAO WENG

Copyright 2014 by John Wiley & Sons, Inc. All rights reserved

Published by John Wiley & Sons, Inc., Hoboken, New JerseyPublished simultaneously in Canada

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any formor by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except aspermitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the priorwritten permission of the Publisher, or authorization through payment of the appropriate per-copy feeto the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400,fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permissionshould be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken,NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permission.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best effortsin preparing this book, they make no representations or warranties with respect to the accuracy orcompleteness of the contents of this book and specifically disclaim any implied warranties ofmerchantability or fitness for a particular purpose. No warranty may be created or extended by salesrepresentatives or written sales materials. The advice and strategies contained herein may not be suitablefor your situation. You should consult with a professional where appropriate. Neither the publisher norauthor shall be liable for any loss of profit or any other commercial damages, including but not limitedto special, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact ourCustomer Care Department within the United States at (800) 762-2974, outside the United Statesat (317) 572-3993 or fax (317) 572-4002.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print maynot be available in electronic formats. For more information about Wiley products, visit our web siteat www.wiley.com.

Library of Congress Cataloging-in-Publication Data:

Scale issues in remote sensing / edited by Qihao Weng.pages cm

ISBN 978-1-118-30504-1 (cloth)1. Remote sensing–Mathematics. 2. Ecology—Mathematical models.

3. Spatial ecology—Mathematical models. I. Weng, Qihao.G70.4.S24 2014621.36 078—dc23

2013031571

Printed in the United States of America

10 9 8 7 6 5 4 3 2 1

CONTENTS

ACKNOWLEDGMENTS ix

CONTRIBUTORS xi

AUTHOR BIOGRAPHY xv

INTRODUCTION 1

1 Characterizing, Measuring, Analyzing, and ModelingScale in Remote Sensing: An Overview 3Qihao Weng

PART I SCALE, MEASUREMENT, MODELING, AND ANALYSIS 11

2 Scale Issues in Multisensor Image Fusion 13Manfred Ehlers and Sascha Klonus

3 Thermal Infrared Remote Sensing for Analysis of LandscapeEcological Processes: Current Insights and Trends 34Dale A. Quattrochi and Jeffrey C. Luvall

4 On the Issue of Scale in Urban Remote Sensing 61Qihao Weng

v

PART II SCALE IN REMOTE SENSING OF PLANTS ANDECOSYSTEMS 79

5 Change Detection Using Vegetation Indices and MultiplatformSatellite Imagery at Multiple Temporal and Spatial Scales 81Edward P. Glenn, Pamela L. Nagler, and Alfredo R. Huete

6 Upscaling with Conditional Cosimulation for MappingAbove-Ground Forest Carbon 108Guangxing Wang and Maozhen Zhang

7 Estimating Grassland Chlorophyll Content from Leafto Landscape Level: Bridging the Gap in Spatial Scales 126Yuhong He

PART III SCALE AND LAND SURFACE PROCESSES 139

8 Visualizing Scale-Domain Manifolds: A Multiscale Geo-Object-BasedApproach 141Geoffrey J. Hay

9 Multiscale Segmentation and Classification of RemoteSensing Imagery with Advanced Edge and Scale-Space Features 170Angelos Tzotsos, Konstantinos Karantzalos, and Demetre Argialas

10 Optimum Scale in Object-Based Image Analysis 197Jungho Im, Lindi J. Quackenbush, Manqi Li, and Fang Fang

PART IV SCALE AND LAND SURFACE PATTERNS 215

11 Scaling Issues in Studying the Relationship BetweenLandscape Pattern and Land Surface Temperature 217Hua Liu and Qihao Weng

12 Multiscale Fractal Characteristics of Urban Landscapein Indianapolis, USA 230Bingqing Liang and Qihao Weng

13 Spatiotemporal Scales of Remote Sensing Precipitation 253Yang Hong and Yu Zhang

vi CONTENTS

PART V NEW FRONTIERS IN EARTH OBSERVATIONTECHNOLOGY 265

14 Multiscale Approach for Ground Filtering from Lidar AltimetryMeasurements 267José L. Silvan-Cárdenas and Le Wang

15 Hyperspectral Remote Sensing with Emphasis on LandCover Mapping: From Ground to Satellite Observations 285George P. Petropoulos, Kiril Manevski, and Toby N. Carlson

INDEX 321

CONTENTS vii

ACKNOWLEDGMENTS

I wish to extend my sincere thanks to all the contributors of this book for making thisendeavor possible. Moreover, I offer my deepest appreciation to all the reviewers,who have taken precious time from their busy schedules to review the chapters.Finally, I am indebted to my family for their enduring love and support. It is myhope that this book will stimulate students and researchers to perform more in-depthanalysis of scale issues in remote sensing and Geographic Information Science.

The reviewers of the chapters are listed here in alphabetical order: Aleksey Boyko,Alexander Buyantuyev, Anatoly Gitelson, Angelos Tzotsos, Benjamin Bechtel,Caiyun Zhang, Cedric Vega, Charles Emerson, Gang Chen, Guangxing Wang,Haibo Yao, Hannes Taubenboeck, Hongbo Su, Hong-lie Qiu, Iryna Dronova, JianjunGe, Lee De Cola, Prasad Thenkabail, Qi Chen, Shelley Meng, Xin Miao, Yuhong He,and Zhixiao Xie.

ix

CONTRIBUTORS

Demetre Argialas, Remote Sensing Laboratory, National Technical University ofAthens, Athens, Greece

Toby N. Carlson, Department of Meteorology, Pennsylvania State University,University Park, PA, USA

Manfred Ehlers, Institute for Geoinformatics and Remote Sensing, University ofOsnabrück, Osnabrück, Germany

Fang Fang, School of Urban and Environmental Engineering, Ulsan NationalInstitute of Science and Technology (UNIST), Ulju-gun, Ulsan, South Korea

Edward P. Glenn, Environmental Research Laboratory of the University ofArizona, Tucson, AZ, USA

Geoffrey J. Hay, Foothills Facility for Remote Sensing and GIScience, Departmentof Geography, University of Calgary, Calgary, Alberta, Canada

Yuhong He, Department of Geography, University of Toronto Mississauga,Mississauga, Ontario, Canada

Yang Hong, School of Civil Engineering and Environmental Science; AdvancedRadar Research Center; Center for Analysis and Prediction of Storms, Universityof Oklahoma, Norman, OK, USA; Water Technology for Emerging Region(WaTER) Center, University of Oklahoma, Norman, OK, USA

Alfredo R. Huete, University of Technology Sydney, Sydney, New South Wales,Australia

xi

Jungho Im, School of Urban and Environmental Engineering, Ulsan NationalInstitute of Science and Technology (UNIST), Ulju-gun, Ulsan, South Korea;Department of Environmental Resources Engineering, State University of NewYork College of Environmental Science and Forestry (SUNY ESF), Syracuse, NY,USA

Konstantinos Karantzalos, Remote Sensing Laboratory, National TechnicalUniversity of Athens, Athens, Greece

Sascha Klonus, Institute for Geoinformatics and Remote Sensing, University ofOsnabrück, Osnabrück, Germany

Manqi Li, School of Urban and Environmental Engineering, Ulsan NationalInstitute of Science and Technology (UNIST), Ulju-gun, Ulsan, South Korea

Bingqing Liang, Department of Geography, University of Northern Iowa, CedarFalls, IA, USA

Hua Liu, Old Dominion University, Norfolk, VA, USA

Jeffrey C. Luvall, NASA Marshall Space Flight Center, Huntsville, AL, USA

Kiril Manevski, Department of Agro-Ecology and Environment, Aarhus University,Blichers Allé, Tjele, Denmark

Pamela L. Nagler, U.S. Geological Survey, Southwest Biological Science Center,Sonoran Desert Research Station, University of Arizona, Tucson, AZ, USA

George P. Petropoulos, Department of Geography and Earth Sciences, Universityof Aberystwyth, Wales, UK

Lindi J. Quackenbush, Department of Environmental Resources Engineering, StateUniversity of New York College of Environmental Science and Forestry (SUNYESF), Syracuse, NY, USA

Dale A. Quattrochi, NASA, Marshall Space Flight Center, Huntsville, AL, USA

José L. Silvan-Cárdenas, Centro de Investigación en Geografía y Geomática “Ing.Jorge L. Tamayo” A.C., Mexico City, Mexico

Angelos Tzotsos, Remote Sensing Laboratory, National Technical University ofAthens, Athens, Greece

Guangxing Wang, Geography and Environmental Resources, Southern IllinoisUniversity, Carbondale, IL, USA

LeWang, Department of Geography University at Buffalo, State University of NewYork, Buffalo, NY, USA

Qihao Weng, Center for Urban and Environmental Change, Department of Earthand Environmental Systems, Indiana State University, Terre Haute, IN, USA

xii CONTRIBUTORS

Maozhen Zhang, College of Environment and Resources, Zhejiang A&FUniversity, Lin-An, ZheJiang, China

Yu Zhang, School of Civil Engineering and Environmental Science; AdvancedRadar Research Center; Center for Analysis and Prediction of Storms, Universityof Oklahoma, Norman, OK, USA

CONTRIBUTORS xiii

AUTHOR BIOGRAPHY

Dr. Qihao Weng is the Director of the Center for Urban andEnvironmental Change and a professor of geography at IndianaState University. He was a visiting NASA senior fellow (2008–2009). Dr. Weng is also a guest/adjunct professor at PekingUniversity, Hong Kong Polytechnic University, Wuhan Univer-sity, and Beijing Normal University, and a guest researchscientist at Beijing Meteorological Bureau, China. He receivedhis Ph.D. in geography from the University of Georgia in 1999.In the same year, he joined the University of Alabama as an

assistant professor. Since 2001, he has been amember of the faculty in the Departmentof Earth and Environmental Systems at Indiana State University, where he has taughtcourses on remote sensing, digital image processing, remote sensing–GIS integration,GIS, and environmental modeling and has mentored 11 doctoral and 10 masterstudents.

Dr. Weng’s research focuses on remote sensing and GIS analysis of urbanecological and environmental systems, land use and land cover change, environ-mental modeling, urbanization impacts, and human–environment interactions. Heis the author of over 150 peer-reviewed journal articles and other publications and 8books. Dr. Weng has worked extensively with optical and thermal remote sensingdata and more recently with lidar data, primarily for urban heat island study, landcover and impervious surface mapping, urban growth detection, image analysisalgorithms, and the integration with socioeconomic characteristics, with financialsupport from U.S. funding agencies that include NSF, NASA, USGS, USAID,NOAA, National Geographic Society, and Indiana Department of NaturalResources. Dr. Weng was the recipient of the Robert E. Altenhofen Memorial

xv

Scholarship Award by the American Society for Photogrammetry and RemoteSensing (1999), the Best Student-Authored Paper Award by the InternationalGeographic Information Foundation (1998), and the 2010 Erdas Award for BestScientific Paper in Remote Sensing by ASPRS (first place). At Indiana StateUniversity, he received the Theodore Dreiser Distinguished Research Award in2006 (the university’s highest research honor) and was selected as a LillyFoundation Faculty Fellow in 2005 (one of the six recipients). In May 2008, hereceived a prestigious NASA senior fellowship. In April 2011, Dr. Weng was therecipient of the Outstanding Contributions Award in Remote Sensing in 2011sponsored by American Association of Geographers (AAG) Remote SensingSpecialty Group. Dr. Weng has given over 70 invited talks (including colloquia,seminars, keynote addresses, and public speeches) and has presented over 100papers at professional conferences (including co-presenting).

Dr. Weng is the Coordinator for GEO’s SB-04, Global Urban Observation andInformation Task (2012–2015). In addition, he serves as an associate editor ofISPRS Journal of Photogrammetry and Remote Sensing and is the series editor forboth the Taylor & Francis Series in Remote Sensing Applications and the McGraw-Hill Series in GIS&T. His past service includes National Director of AmericanSociety for Photogrammetry and Remote Sensing (2007–2010), Chair of AAGChina Geography Specialty Group (2010–2011), and Secretary of ISPRS WorkingGroup VIII/1 (Human Settlement and Impact Analysis, 2004–2008), as well as apanel member of the U.S. DOE’s Cool Roofs Roadmap and Strategy in 2010.

xvi AUTHOR BIOGRAPHY

INTRODUCTION

1CHARACTERIZING, MEASURING,ANALYZING, ANDMODELING SCALEIN REMOTE SENSING: AN OVERVIEW

QIHAO WENG

1.1 SCALE ISSUES IN REMOTE SENSING

Scale is a fundamental and crucial issue in remote sensing studies and image analysis.The University Consortium for Geographic Information Science (UCGIS) identified itas a main research priority area (1996). Scale influences the examination of landscapepatterns in a region. The change of scale is relevant to the issues of data aggregation,information transfer, and the identification of appropriate scales for analysis (Krönertet al., 2001; Wu and Hobbs, 2002). Extrapolation of information across spatial scalesis a needed research task (Turner, 1990). It is suggested that spatial characteristicscould be transferred across scales under specific conditions (Allen et al., 1987).Therefore, we need to know how the information is transferred from a fine scale to abroad scale (Krönert et al., 2001). In remote sensing studies, use of data from varioussatellite sensors may result in different research results, since they usually havedifferent spatial resolutions. Therefore, it is significant to examine changes in spatialconfiguration of any landscape pattern as a result of using different spatial resolutionsof satellite imagery. Moreover, it is always necessary to find the optimal scale for astudy in which the environmental processes operate. Theories, methods, and modelsfor multiscaling are crucial to understand the heterogeneity of landscapes (Wu and Qi,2000; Wu and Hobbs, 2002). Methods and techniques are important for theexamination of spatial arrangements at a wide range of spatial scales. Regionalization

3

Scale Issues in Remote Sensing, First Edition. Edited by Qihao Weng.� 2014 John Wiley & Sons, Inc. Published 2014 by John Wiley & Sons, Inc.

describes a transition from one scale to another, and upscaling or downscaling is anessential protocol in the transition (Krönert et al., 2001).

Characterized by irregularity and scale independence, fractals are recognized asa suitable method to capture the self-similarity property of the spatial structure ofinterest (Zhao, 2001). Self-similarity represents invariance with respect to scale. Ingeoscience, the property of self-similarity is often interpreted as scale independence(Clarke, 1986). However, most environmental phenomena are not pure fractals atall scales. Rather, they only exhibit a certain degree of self-similarity within limitedregions and over limited ranges of scale, which is measurable by using statisticssuch as spatial auto-covariances. The underlying principle of fractals is to use strictor statistical self-similarity to determine the fractal dimension (FD) of an object/surface, which is often used as an indicator of the degree of irregularity orcomplexity of objects. When fractals are applied to remote sensing, an image isviewed as a complex “hilly terrain surface” whose elevations are represented by thedigital numbers. Consequently, FDs are readily computable and can be used todenote how complicated the “image surfaces” are. Remote sensing studies assumethat spatial complexity directly results from spatial processes operating at variouslevels, and higher FD occurs at the scale where more processes operate. With FDs,the spatial processes that occurred at different scales are measurable and compara-ble. Compared to other geospatial algorithms in image analysis such as landscapemetrics, fractals offer a better benefit in that they can be directly applied to rawimages without the need for classification or land cover feature identification, inaddition to their sound mathematic bases. Therefore, it is not surprising to see agrowing number of researches utilize fractals in remote sensing image analysis (DeJong and Burrough, 1995; Emerson et al., 1999, 2005; Lam, 1990; Lam and DeCola, 1993; Myint, 2003; Qiu et al., 1999; Read and Lam, 2002; Weng, 2003).Fractal-derived texture images have also been used as additional layers in imageclassification (Myint, 2003).

Spatial resolution has been another focus in remote sensing studies. It is necessaryto estimate the capability of remote sensing data in landscape mapping since theapplication of remote sensing may be limited by its spatial resolution (Aplin, 2006;Buyantuyev and Wu, 2007; Ludwig et al., 2007). Imagery with finer resolutioncontains greater amount of spatial information, which, in turn, enables the characteri-zation of smaller features better. The proportion of mixed pixels is expected toincrease as spatial resolution becomes coarser (Aplin, 2006). Stefanov and Netzband(2005) identified weak positive and negative correlations between the normalizedvegetation index (NDVI) and landscape structure at three different resolutions (250,500, and 1000m) when they examined the capability of the Moderate ResolutionImaging Spectroradiometer (MODIS) NDVI data in the assessment of arid landscapecharacteristics in Phoenix. Asner et al. (2003) examined the significance of subpixelestimates of biophysical structure with the help of high-resolution remote sensingimagery and found a strong correlation between the senescent and unmixed greenvegetation cover values in a deforested area. Agam et al. (2007) sharpened the coarse-resolution thermal imagery to finer resolution imagery based on the analysis of therelationship between vegetation index and land surface temperature. The results

4 CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE

showed that the vegetation index–based sharpening method provided an effectiveway to improve the spatial resolution of thermal imagery.

Adaptive choice of spatial and categorical scales in landscape mapping wasdemonstrated by Ju et al. (2005). They provided a data-adaptive choice of spatialscale varying by location jointed with categorical scale by the assistance of a statisticalfinite mixture method. Buyantuyev andWu (2007) systematically analyzed the effectsof thematic resolution on landscape pattern analysis. Two problems need to beconsidered in landscape mapping: the multiplicity of classification schemes and thelevel of detail of a particular classification. They found that the thematic resolutionhad obvious effects on most of the landscape metrics, which indicated that changingthematic resolution may significantly affect the detection of landscape changes.However, an increase in spatial resolution may not lead to a better observation sinceobjects may be oversampled and their features may vary and be confusing (Hsiehet al., 2001; Aplin and Atkinson, 2004). Although coarse resolution may includefewer features, imagery with too fine resolution for a specific purpose can be degradedin the process of image resampling (Ju et al., 2005). Remote sensing data may not bealways be sufficient when specific problems were addressed at specific scales and on-ground assessment may be needed, since coarser imagery cannot provide sufficientinformation about the location and connectivity in specific areas (Ludwig et al., 2007).

Substantial researches have previously been conducted on scale-related issues inremote sensing studies, as discussed above. This book intends to revisit andreexamine the scale and related issues. It will also address how new frontiers inEarth observation technology since 1999—such as very high resolution, hyper-spectral, lidar sensing, and their synergy with existing technologies and advancesin remote sensing imaging science such as object-oriented image analysis, data fusion,and artificial neural networks—have impacted the understanding of this basic butpivotal issue. The scale-related issues will be examined from three interrelatedperspectives: in landscape properties, patterns, and processes. These examinationsare preceded by a theoretical exploration of the scale issue by a group of authorities inthe field of remote sensing. The concluding section prospects emerging trends inremote sensing over the next decade(s) and their relationship with scale.

1.2 CHARACTERIZING, MEASURING, ANALYZING,ANDMODELING SCALE

This book consists of 5 parts and 14 chapters, in addition to this introductory chapter.Part I focuses on theoretical aspects of scale and scaling. Part II deals with theestimation and measurement of vegetation parameters and ecosystems across variousspatial and temporal scales. Part III examines the effect of scaling on imagesegmentation and object extraction from remotely sensed imagery. Part IV exem-plifies with case studies on the scale and scaling issues in land cover analysis and inland–atmosphere interactions. Finally, Part V addresses how new frontiers in Earthobservation technology, such as hyperspectral and lidar sensing, have impacted theunderstanding of the scale issue.

CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE 5

Three chapters are included in Part I. In Chapter 2, Ehlers and Klonus examine datafusion results of remote sensing imagery with various spatial scales. The scales arethought to relate to the ground sampling distances (GSDs) of the respective sensors.They find that for electro-optical sensors GSD or scale ratios of 1:10 (e.g., IKONOSand SPOT-5 fusion) can still produce acceptable results if the fusion method is basedon a spectral characteristic-preserving technique such as the Ehlers fusion. Usingradar images as a substitute for high-resolution panchromatic data is possible, but onlyfor scale ratios between 1:6 and 1:20 due to the limited feature recognition in radarimages. In Chapter 3, Quattrochi and Luvall revisit an article published in LandscapeEcology in 1999 by them and examine the direct or indirect uses of thermal infrared(TIR) remote sensing data to analyze landscape biophysical characteristics to offerinsights on how these data can be used more robustly for furthering the understandingand modeling of landscape ecological processes. In Chapter 4, Weng discusses someimportant scale-related issues in urban remote sensing. The requirements for mappingthree interrelated entities or substances in the urban space (i.e., material, land cover,and land use) and their relationships are first examined. Then, the relationshipbetween spatial resolution and the fabric of urban landscapes is assessed. Next,the operational scale/optimal scale for the studies of land surface temperature arereviewed. Finally, the issue of scale dependency of urban phenomena is discussed viareviewing two case studies, one on land surface temperature (LST) variability acrossmultiple census levels and the other on multiscale residential population estimationmodeling.

Part II also contains three chapters. Vegetation indices can be used to separatelandscape components into bare soil, water, and vegetation and, if calibrated withground data, to quantify biophysical variables such as leaf area index and fractionalcover and physiological variables such as evapotranspiration and photosynthesis. InChapter 5, Glenn, Nagler, and Huete use a case study approach to show how remotelysensed vegetation indices collected at different scales can be used in vegetationchange detection studies. The primary sensor systems discussed are digital pheno-cams, Landsat and MODIS, which cover a wide range of spatial (1 cm–250m) andtemporal (15min–16 days) resolutions/scales. Sources of error and uncertaintyassociated with both ground and remote sensing measurements in change studiesare also discussed. In Chapter 6, Wang and Zhang combine plot data and ThematicMapped (TM) images to map above-ground forest carbon at a 990-m pixel resolutionin Lin-An, Zhejiang Province, China, by using two upscaling methods: point simplecokriging point cosimulation and point simple cokriging block cosimulation Theirresults suggest that both methods perform well in scaling up the spatial data as well asin revealing the propagation of input data uncertainties from a finer spatial resolutionto a coarser one. The output uncertainties reflect the spatial variability of theestimation accuracy caused by the locations of the input data and the valuesthemselves. In Chapter 7, Yuhong He intends to bridge the gap in spatial scalesthrough estimating grassland chlorophyll contents from leaf to landscape level using asimple yet effective canopy integration method. Using data collected in a heteroge-neous tall grassland located at Ontario, Canada, Yuhong’s study first scales leaf levelchlorophyll measurements to canopy and landscape levels and then investigates the

6 CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE

relationships between a chlorophyll spectral index and vegetation chlorophyll con-tents at the leaf, canopy, and landscape scales. Significant relationships are found at allthree scales, suggesting that it is feasible to accurately estimate chlorophyll contentsusing both ground and space remote sensing data.

In remote sensing, image segmentation has a longer history and has its roots inindustrial image processing butwas not used extensively in the geospatial community inthe 1980s and 1990s (Blaschke, 2010). Object-oriented image analysis has beenincreasingly used in remote sensing applications due to the advent of high-spatial-resolution image data and the emergence of commercial software such as eCognition(Benz et al., 2004; Wang et al., 2004). In the process of creating objects, a scaledetermines the occurrence or absence of an object class. Thus, the issue of scale andscaling are fundamental considerations in the extraction, representation, modeling, andanalyses of image objects (Hay et al., 2002; Tzotsos et al., 2011).

The three chapters in Part III focus on discussion of these issues. In Chapter 8, Hayintroduces a novel geo-object-based framework that integrates hierarchy theory andlinear scale space (SS) for automatically visualizing and modeling landscape scaledomains over multiple scales. Specifically, this chapter describes a three-tier hierarchi-cal methodology for automatically delineating the dominant structural componentswithin 200 different multiscale representations of a complex agro-forested landscape.By considering scale-space events as critical domain thresholds, Hay further defines anew scale-domain topology that may improve querying and analysis of this complexmultiscale scene. Finally, Hay shows how to spatially model and visualize thehierarchical structure of dominant geo-objects within a scene as “scale-domain mani-folds” and suggests that they may be considered as a multiscale extension to thehierarchical scaling ladder as defined in the hierarchical patch dynamics paradigm.Chapter 9 byTzotsos,Karantzalos, andArgialas introduces amultiscale object-orientedimage analysis framework which incorporates a region-merging segmentation algo-rithm enhanced by advanced edge features and nonlinear scale-space filtering. Initially,edge and line features are extracted from remote sensing imagery at several scales usingscale-space representations. These features are then used by the enhanced segmentationalgorithm as constraints in the growth of image objects at various scales. Throughiterative pairwise objectmerging, the final segmentation can be achieved. Image objectsare then computed at various scales and passed on to a kernel-based learning machinefor classification. This image classification framework was tested on very highresolution imagery acquired by various airborne and spaceborne panchromatic, multi-spectral, hyperspectral, and microwave sensors, and promising experimental resultswere achieved. Chapter 10, by Im, Quackenbush, Li, and Fang, provides a review ofrecent publications on object-based image analysis (OBIA) focusing on determinationof optimum scales for image segmentation and the related trends. Selecting optimumscale is often challenging, since (1) there is no standardized method to identify theoptimality and (2) scales in most segmentation algorithms are arbitrarily selected. Theauthors suggest that there should be transferable guidelines regarding segmentationscales to facilitate the generalization of OBIA in remote sensing applications, to enableefficient comparison of different OBIA approaches, and to select optimum scales for themultitude of different image components.

CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE 7

Part IV introduces three case studies on the scale and scaling issues in analysis ofland cover, landscape metrics, and biophysical parameters. Chapter 11 by Liu andWeng assesses the effect of scaling on the relationship between landscape pattern andland surface temperature with a case study in Indianapolis, Indiana. A set of spatialresolutions were compared by using a landscape metric space. They find that thespatial resolution of 90m is the optimal scale to study the relationship and think that itis the operational scale of the urban thermal landscape in Indianapolis. In Chapter 12,Liang and Weng provide an evaluation of the effectiveness of the triangular prismfractal algorithm for characterizing urban landscape in Indianapolis based on eightsatellite images acquired by five different sensors: Landsat Multispectral Scanner,Landsat Thematic Mapper, Landsat Enhanced Thematic Mapper Plus, AdvancedSpaceborne Thermal Emission and Reflection, and IKONOS. Fractal dimensionscomputed from the selected original, classified, and resampled images are comparedand analyzed. The potential of fractal measurement in the studies of landscape patterncharacterization and the scale/resolution issues are further assessed. Chapter 13 byHong and Zhang provides important insights into the spatiotemporal scales ofremotely sensed precipitation. This chapter first overviews the precipitation measure-ment methods—both traditional rain gauge and advanced remote sensing measure-ments; then develops an uncertainty analysis framework that can systematicallyquantify the remote sensing precipitation estimation error as a function of space, time,and intensity; and finally assesses the spatiotemporal scale-based error propagation inremote sensing precipitation estimates into hydrological prediction.

The last part of this book looks at how new frontiers in Earth observationtechnology have transformed our understanding of this foremost issue in remotesensing. Chapter 14 examines lidar data processing, whereas Chapter 15 exploreshyperspectral remote sensing for land cover mapping. Digital terrain models (DTMs)are basic products required for a number of applications and decision making.Nowadays, high-spatial-resolution DTMs are primarily produced through airbornelaser scanners (ALSs). However, the ALS does not directly deliver DTMs; rather itdelivers a dense point cloud that embeds both terrain elevation and height of naturaland human-made features. Hence, discrimination of above-ground objects fromterrain is a basic processing step. This processing step is termed ground filteringand has proved especially difficult for large areas of varied terrain characteristics. InChapter 14, Silvan-C�ardenas and Wang revise and extend a filtering method based ona multiscale signal decomposition termed the multiscale Hermite transform (MHT).The formal basis of the latter is presented in the context of scale-space theory, a theoryfor representing spatial signals. Through the unique properties of the MHT, namelylocal spatial rotation and scale-space shifting, the original filtering algorithm wasextended to incorporate higher order coefficients in the multiscale erosion operation.Additionally, a linear interpolation was incorporated through a truncated Taylorexpansion which allowed improving the ground filtering performance along sloppyterrain areas. Practical considerations in the operation of the algorithm are discussedand illustrated with examples. In Chapter 15, Petropoulos, Manevski, and Carlsonassess the potential of hyperspectral remote sensing systems for improving discrimi-nation among similar land cover classes at different scales. The chapter provides first

8 CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE

an overview of the current state of the art in the use of field spectroradiometry inexamining the spectral discrimination between different land cover targets. In thisframework, techniques employed today and linked with the most important scalefactors are critically reviewed and examples of recent related studies and spectrallibraries are provided. Then, it focuses on the use of hyperspectral remote sensing forobtaining land use/cover mapping from space. An overview of the different satellitesensors and techniques employed is furnished, providing examples taken from recentstudies. The chapter closes by highlighting the main challenges that need to beaddressed in the future towards a more precise estimation of land cover from spectralinformation acquired from hyperspectral sensing systems at variant spatial scales.

REFERENCES

Agam, N., Kustas, W. P., Anderson, M. C., Li, F., and Neale, C.M. U. 2007. A vegetation indexbased technique for spatial sharpening of thermal imagery. Remote Sensing of Environment107(4):545–558.

Allen, T. F. H., O’Neill, R. V., and Hoekstra, T. W. 1987. Interlevel relations in ecologicalresearch and management: Some working principles from hierarchy. Journal of AppliedSystems Analysis 14:63–79.

Aplin P. 2006. On scales and dynamics in observing the environment. International Journal ofRemote Sensing 27(11):2123–2140.

Aplin, P., and Atkinson, P. M. 2004. Predicting missing field boundaries to increase per-field classification accuracy. Photogrammetric Engineering and Remote Sensing70:141–149.

Asner, G. P., Bustamante, M. M. C., and Townsend, A. R. 2003. Scale dependence ofbiophysical structure in deforested areas bordering the Tapajos National Forest, CentralAmazon. Remote Sensing of Environment 87(4):507–520.

Benz, U. C., Hofmann, P., Willhauck, G., Lingenfelder, I., and Heynen, M. 2004. Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information.ISPRS Journal of Photogrammetry & Remote Sensing 58:239–258.

Blaschke, T. 2010. Object based image analysis for remote sensing. ISPRS Journal ofPhotogrammetry and Remote Sensing 65(1):2–16.

Buyantuyev, A., and Wu, J. 2007. Effects of thematic resolution on landscape pattern analysis.Landscape Ecology 22(1):7–13.

Clarke, K. C. 1986. Computation of the fractal dimension of topographic surfaces using thetriangular prism surface area method. Computers and Geosciences 12:713–722.

De Jong, S. M., and Burrough, P. A. 1995. A fractal approach to the classification ofMediterranean vegetation types in remotely sensed images. Photogrammetric Engineeringand Remote Sensing 61:1041–1053.

Emerson, C.W., Lam, N. S. N., and Quattrochi, D. A. 1999.Multiscale fractal analysis of imagetexture and pattern. Photogrammetric Engineering and Remote Sensing 65:51–61.

Emerson, C. W., Lam, N. S. N., and Quattrochi, D. A. 2005. A comparison of local variance,fractal dimension, and Moran’s I as aids to multispectral image classification. InternationalJournal of Remote Sensing 26:1575–1588.

REFERENCES 9

Hay, G. J., Dube, P., Bouchard, A., andMarceau, D. J. 2002. A scale-space primer for exploringand quantifying complex landscapes. Ecological Modelling 153(1–2):2–49.

Hsieh, P. F., Lee, L. C., and Chen, N. Y. 2001. Effect of spatial resolution on classificationerrors of pure and mixed pixels in remote sensing. IEEE Transactions on Geoscience andRemote Sensing 39(12):2657–2663.

Ju, J. C., Gopal, S., and Kolaczyk, E. D. 2005. On the choice of spatial and categorical scale inremote sensing land-cover classification. Remote Sensing of Environment 96(1):62–77.

Krönert, R., Steinhardt, U., and Volk, M. 2001. Landscape Balance and Landscape Assess-ment. New York: Springer.

Lam, N. S. N. 1990. Description and measurement of Landsat TM images using fractals.Photogrammetric Engineering and Remote Sensing 56:187–195.

Lam, N. S. N., and De Cola, L. 1993. Fractals in Geography. Englewood Cliffs, NJ: PrenticeHall.

Ludwig, J., Bastin, G., Wallace, J., and McVicar, T. 2007. Assessing landscape health byscaling with remote sensing: When is it not enough? Landscape Ecology 22(2):163–169.

Myint, S. W. 2003. Fractal approaches in texture analysis and classification of remotely senseddata: Comparisons with spatial autocorrelation techniques and simple descriptive statistics.International Journal of Remote Sensing 24:1925–1987.

Qiu, H.-L., Lam, N. S. N., Quattrochi, D. A., and Gamon, J. A. 1999. Fractal characterization ofhyperspectral imagery. Photogrammetric Engineering and Remote Sensing 65:63–71.

Read, J. M., and Lam, N. S. N. 2002. Spatial methods for characterizing land cover anddetecting land-cover changes for the tropics. International Journal of Remote Sensing23:2457–2474.

Stefanov,W. L., and Netzband,M. 2005. Assessment of ASTER land-cover andMODIS NDVIdata at multiple scales for ecological characterization of an arid urban center. RemoteSensing of Environment 99(1–2):31–43.

Turner, M. G. 1990. Spatial and temporal analysis of landscape patterns. Landscape Ecology4(1):21–30.

Tzotsos, A., Karantzalos, K., and Argialas, D. 2011. Object-based image analysis throughnonlinear scale-space filtering. ISPRS Journal of Photogrammetry and Remote Sensing66:2–16.

Wang, L., Sousa, W. P., Gong, P., and Biging, G. S. 2004. Comparison of IKONOS andQuickBird images for mapping mangrove species on the Caribbean coast of panama.Remote Sensing of Environment 91:432–440.

Weng, Q. 2003. Fractal analysis of satellite-detected urban heat island effect. PhotogrammetricEngineering and Remote Sensing 69:555–566.

Wu, J., and Hobbs, R. 2002. Key issues and research priorities in landscape ecology: Anidiosyncratic synthesis. Landscape Ecology 17:355–365.

Wu, J., and Qi, Y. 2000. Dealing with scale in landscape analysis: An overview. GeographicInformation Sciences 6(1):1–5.

Zhao, W. 2001. Multiscale Analysis for Characterization of Remotely Sensed Images. Ph.D.Dissertation, Louisiana State University.

10 CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE

PART I

SCALE, MEASUREMENT, MODELING,AND ANALYSIS

2SCALE ISSUES IN MULTISENSORIMAGE FUSION

MANFRED EHLERS AND SASCHA KLONUS

2.1 SCALE IN REMOTE SENSING

Scale is a term that is used in many scientific applications and communities. Typicalwell-known measurement scales are, for example, the Richter scale for the magnitudeof earthquakes and the Beaufort scale for wind speed. We speak of large-scaleoperations if they involve large regions or many people. Cartographers use the termscale for the description of the geometric relationship between a map and real-worldcoordinates. In remote sensing, scale is usually associated with the latter meaning themap scale for typical applications of remote sensors. To a large degree, scale isdependent on the geometric resolution of the sensor which can be measured in groundsampling distance (GSD). The GSD is usually the same or similar to the final pixelsize of the remote sensing data set. In addition to the GSD, scale is also associated withthe level and quality of information that can be extracted from remotely sensed data.

Especially with the launch of the first SPOT satellite with independent panchromaticand multispectral sensors, it became evident that a combined analysis of the high-resolution panchromatic sensor and the lower resolution multispectral images wouldyield better results than any single image alone. Subsequently, most Earth observationsatellites, such as the SPOT and Landsat series, or the very high resolution (VHR)sensors such as IKONOS, QuickBird, or GeoEye acquire image data in two differentmodes, a low-resolution multispectral and a high-resolution panchromatic mode. TheGSD or scale ratio between the panchromatic and the multispectral image can varybetween 1 : 2 and 1 : 8 with 1 : 4 the most common value. This ratio can even becomesmaller when data from different sensors are used, which is, for example, necessary if

13

Scale Issues in Remote Sensing, First Edition. Edited by Qihao Weng. 2014 John Wiley & Sons, Inc. Published 2014 by John Wiley & Sons, Inc.

satellite sensorswith only panchromatic (e.g.,WorldView-1) or onlymultispectral (e.g.,RapidEye) information are involved. Consequently, efforts started in the late 1980s todevelop methods for merging or fusing panchromatic and multispectral image data toform multispectral images of high geometric resolution. In this chapter, we willinvestigate to what degree fusion techniques can be used to form multispectral imagesof larger scale when combined with high-resolution black-and-white images.

2.2 FUSION METHODS

Similar to the term scale, the word fusion has different meanings for differentcommunities. In a special issue on data fusion of the International Journal ofGeographical Information Science(IJGIS), Edwards and Jeansoulin (2004, p. 303)state that “data fusion is a complex process with a wide range of issues that must beaddressed. In addition, data fusion exists in different forms in different scientificcommunities. Hence, for example, the term is used by the image community toembrace the problem of sensor fusion, where images from different sensors arecombined. The term is also used by the database community for parts of theinteroperability problem. The logic community uses the term for knowledge fusion.”

Consequently, it comes as no surprise that several definitions for data fusion can befound in the literature. Pohl and van Genderen (1998, p. 825) proposed that “imagefusion is the combination of two or more different images to form a new image byusing a certain algorithm.” Mangolini (1994) extended data fusion to information ingeneral and also refers to quality. He defined data fusion as a set of methods, tools andmeans using data coming from various sources of different nature, in order to increasethe quality (in a broad sense) of the requested information (Mangolini, 1994).Hall and Llinas (1997, p. 6) proposed that “data fusion techniques combine datafrom multiple sensors, and related information from associated databases.” However,Wald (1999) argued that Pohl and van Genderen’s definition is restricted to images.Mangolini’s definition puts the accent on the methods. It contains the large diversityof tools but is restricted to these. Hall and Llinas refer to information quality in theirdefinition but still focus on the methods.

The Australian Department of Defence defined data fusion as a “multilevel,multifaceted process dealing with the automatic detection, association, correlation,estimation, and combination of data and information from single and multiplesources” (Klein, 2004, p. 52). This definition is more general with respect tothe types of information than can be combined (multilevel process) and very popularin the military community. Notwithstanding the large use of the functional model, thisdefinition is not suitable for the concept of data fusion, since it includes itsfunctionality as well as the processing levels. Its generalities as a definition forthe concept are reduced (Wald, 1999). A search for a more suitable definition waslaunched by the European Association of Remote Sensing Laboratories (EARSeL)and the French Society for Electricity and Electronics (SEE, French affiliate of theInstitute of Electrical and Electronics Engineers) and the following definition wasadopted in January 1998: “Data fusion is a formal framework in which are expressed

14 SCALE ISSUES IN MULTISENSOR IMAGE FUSION

means and tools for the alliance of data originating from different sources. It aims atobtaining information of greater quality; the exact definition of ‘greater quality’ willdepend upon the application” (Wald, 1999, p. 1191).

Image fusion forms a subgroup within this definition, with the objective to generatea single image from multiple image data for the extraction of information of higherquality (Pohl, 1999). Image fusion is used in many fields such as military, medicalimaging, computer vision, the robotics industry, and remote sensing of the environ-ment. The goals of the fusion process are multifold: to sharpen multispectral images,to improve geometric corrections, to provide stereo-viewing capabilities for stereo-photogrammetry, to enhance certain features not visible in either of the single data setsalone, to complement data sets for improved classification, to detect changes usingmultitemporal data, and to replace defective data (Pohl and van Genderen, 1998). Inthis article, we concentrate on the image-sharpening process (iconic fusion) and itsrelationship with image scale.

Many publications have focused on how to fuse high-resolution panchromaticimages with lower resolution multispectral data to obtain high-resolution multi-spectral imagery while retaining the spectral characteristics of the multispectral data(e.g., Cliche et al., 1985; Welch and Ehlers, 1987; Carper et al., 1990; Chavez et al.,1991; Wald et al., 1997; Zhang 1999). It was evident that these methods seem to workwell for many applications, especially for single-sensor, single-date fusion. Mostmethods, however, exhibited significant color distortions for multitemporal andmultisensoral case studies (Ehlers, 2004; Zhang, 2004).

Over the last few years, a number of improved algorithms have been developedwith the promise to minimize color distortion while maintaining the spatial improve-ment of the standard data fusion algorithms. One of these fusion techniques is Ehlersfusion, which was developed for minimizing spectral change in the pan-sharpeningprocess (Ehlers and Klonus, 2004). In a number of comprehensive comparisons, thismethod has tested superior to most of the other pan-sharpening techniques (see, e.g.,Ling et al., 2007a; Ehlers et al., 2010; Klonus, 2011; Yuhendra et al., 2012). For thisreason, we will use Ehlers fusion as the underlying technique for the followingdiscussions. The next section presents a short overview of this fusion technique.

2.3 EHLERS FUSION

Ehlers fusion was developed specifically for spectral characteristic-preserving imagemerging (Klonus and Ehlers, 2007). It is based on an intensity–hue–saturation (IHS)transform coupled with a Fourier domain filtering. The principal idea behind spectralcharacteristic-preserving image fusion is that the high-resolution image has to sharpenthe multispectral image without adding new gray-level information to its spectralcomponents. An ideal fusion algorithm would enhance high-frequency changes suchas edges and gray-level discontinuities in an image without altering the multispectralcomponents in homogeneous regions. To facilitate these demands, two prerequisiteshave to be addressed. First, color information and spatial information have to beseparated. Second, the spatial information content has to be manipulated in a way that

EHLERS FUSION 15

allows an adaptive enhancement of the images. This is achieved by a combination ofcolor and Fourier transforms (Figure 2.1).

For optimal color separation, use is made of an IHS transform. This technique isextended to include more than three bands by using multiple IHS transforms until thenumber of bands is exhausted. If the assumption of spectral characteristic preservationholds true, there is no dependency on the selection or order of bands for the IHStransform. Subsequent Fourier transforms of the intensity component and thepanchromatic image allow an adaptive filter design in the frequency domain. Usingfast Fourier transform (FFT) techniques, the spatial components to be enhanced orsuppressed can be directly accessed. The intensity spectrum is filtered with a low-pass(LP) filter whereas the spectrum of the high-resolution image is filtered with aninverse high-pass (HP) filter. After filtering, the images are transformed back into thespatial domain with an inverse FFT and added together to form a fused intensitycomponent with the low-frequency information from the low-resolution multispectralimage and the high-frequency information from the high-resolution panchromaticimage. This new intensity component and the original hue and saturation componentsof the multispectral image form a new IHS image. As the last step, an inverse IHStransformation produces a fused RGB image that contains the spatial resolution ofthe panchromatic image and the spectral characteristics of the multispectral image.These steps can be repeated with successive three-band selections until all bands arefused with the panchromatic image. The order of bands and the inclusion of spectralbands for more than one IHS transform are not critical because of the colorpreservation of the procedure (Klonus and Ehlers, 2007). In all investigations, theEhlers fusion provides a good compromise between spatial resolution enhancementand spectral characteristics preservation (Yuhendra et al., 2012), which makes it anexcellent tool for the investigation of scale and fusion. Fusion techniques to sharpenmultispectral images have focused primarily on the merging of panchromatic andmultispectral electro-optical (EO) data with emphasis on single-sensor, single-datefusion. Ehlers fusion, however, allows multiple-sensor, multiple-date fusion andhas recently been extended to fuse radar and EO image data (Ehlers et al., 2010).Consequently, we will address scale-fusion issues for multisensor, multitemporalEO image merging as well as the inclusion of radar image data as a “substitute” forpanchromatic images.

FIGURE 2.1 Concept of Ehlers fusion.

16 SCALE ISSUES IN MULTISENSOR IMAGE FUSION

2.4 FUSION OF MULTISCALE ELECTRO-OPTICAL DATA

2.4.1 Data Sets and Study Site

For our investigation, we selected a study area in Germany representing a part of thevillage of Romrod (Figure 2.2). This area was chosen because it contained severalimportant features for fusion quality assessment such as agricultural lands (fields andgrasslands), man-made structures (houses, roads), and natural areas (forest).

It was used as a control site of the Joint Research Centre of the EuropeanCommission (JRC) in the project “Control with Remote Sensing of Area-BasedSubsidies ” (CwRS) (http://agr i fish.jrc.it /marspac/ DCM/). A seri es of mul titemporalmultispectral remote sensing images were available for this site and formed the basisfor our multiscale fusion investigation (Table 2.1).

2.4.2 Multisensor Image Fusion

The multisensor images cover a time frame of almost four years and virtually allseasons and thus pose an excellent challenge for a combined scale–fusion investiga-tion. All multispectral images were fused using the only panchromatic data set, that is,the IKONOS image from June 18, 2005. As a first part of our investigation, allmultispectral images were registered to the panchromatic IKONOS image, resampled

FIGURE 2.2 Study site for EO image fusion (image Google).

FUSION OF MULTISCALE ELECTRO-OPTICAL DATA 17