J.B. Adams, A.R. Gillespie, ,Remote Sensing of Landscapes with Spectral Images (2006) Cambridge...

2
Computers & Geosciences 34 (2008) 422–423 Book review Remote Sensing of Landscapes with Spectral Images, J.B. Adams, A.R. Gillespie. Cambridge University Press, Cambridge, UK (2006). 362pp., US$85, Hardback, ISBN: 0-521-66221-4 Thematic maps of landscapes prepared from high altitude or satellite imagery are extremely useful for determining the spatial distributions of geologic features. This new book by John Adams and Alan Gillespie, both professors of Earth and Space Sciences at the University of Washington, was written as a textbook for graduate students and professionals who desire to use spectral images to identify materials on the ground. The authors rely heavily on just a handful of well-chosen examples that are examined with a variety of techniques. They emphasize the practical uses of the spectral images rather than the theory and mathematical algorithms behind the analysis processes, which the authors note have already been covered so thoroughly elsewhere that they saw no reason to repeat the work. The scarcity of equations in this text is adequately compensated with many clear graphs and derivative images. For those interested in gaining a deeper understanding, the authors frequently refer the reader to seminal books and journal articles for the details of spectral image analysis techniques and algorithms (indeed, the book contains 6 full pages of references). I did several times wish the book had an Appendix with a few of the more interesting or useful algorithms written out. However, every chapter makes extensive use of boxed margin notes that provide definitions, reminders, and useful tips to keep in mind. The book’s figure captions are very clearly written, providing detailed descriptions that nicely supplement the main text. The majority of images in this book are black and white, with only 9 color plates bound into the center of the book. Color versions of the majority of these black and white figures are available for viewing on a Cambridge University Press web site dedicated to this publication. It is ideal if the reader has an internet connection conveniently available to view these color images while reading the book. Chapter 1, ‘‘Extracting information from spectral images’’ (38 pages), stresses the importance of invoking physical models that relate the remote sensing measurements with the materials and processes on the ground. The authors start with the simplest physical process, shadows and shading, as an introduction to the complexities of spectral analysis. The chapter includes a discussion of testing and validation of results, including why results go wrong, and the tradeoffs between accuracy, spatial scale, and detectability. The chapter closes with a very useful section summarizing the steps for extracting information from spectral images. Chapter 2, ‘‘Spectroscopy of landscapes’’ (26 pages), introduces the basics of reflectance and emittance spectroscopy, including a discussion of the differences between full-spectrum laboratory reference measurements and the limited channels of remote sensing equipment. In the section on spectral mixing, the reader is introduced to Clark’s Law, ‘‘First find out what is there, then worry about how much’’, a theme that repeats throughout this text- book. Another recurring theme is that spectral modeling should begin with a theoretical physical model of field observations (the hypothesis), and then the hypothesis should be tested for agreement with the remote sensing data. Chapter 3, ‘‘Standard methods for analyzing spectral images’’ (61 pages), discusses which algo- rithms work under what circumstances and illus- trates several common pitfalls. Most of these algorithms serve to enhance certain characteristics of interest, for instance photosynthetic vegetation, but these algorithms are mostly empirical and do not actually invoke physical models. In this chapter, we are introduced to a synthetic image designed by the authors to provide a ‘‘truth’’ dataset against which the various algorithms can be tested. It was enlightening indeed to see how badly many of the ARTICLE IN PRESS www.elsevier.com/locate/cageo doi:10.1016/j.cageo.2007.08.001

Transcript of J.B. Adams, A.R. Gillespie, ,Remote Sensing of Landscapes with Spectral Images (2006) Cambridge...

Page 1: J.B. Adams, A.R. Gillespie, ,Remote Sensing of Landscapes with Spectral Images (2006) Cambridge University Press,Cambridge, UK 0-521-66221-4 362pp., US$85, Hardback.

ARTICLE IN PRESS

doi:10.1016/j.ca

Computers & Geosciences 34 (2008) 422–423

www.elsevier.com/locate/cageo

Book review

Remote Sensing of Landscapes with Spectral Images,

J.B. Adams, A.R. Gillespie. Cambridge University

Press, Cambridge, UK (2006). 362pp., US$85,

Hardback, ISBN: 0-521-66221-4

Thematic maps of landscapes prepared from highaltitude or satellite imagery are extremely useful fordetermining the spatial distributions of geologicfeatures. This new book by John Adams and AlanGillespie, both professors of Earth and SpaceSciences at the University of Washington, waswritten as a textbook for graduate students andprofessionals who desire to use spectral images toidentify materials on the ground. The authors relyheavily on just a handful of well-chosen examplesthat are examined with a variety of techniques. Theyemphasize the practical uses of the spectral imagesrather than the theory and mathematical algorithmsbehind the analysis processes, which the authorsnote have already been covered so thoroughlyelsewhere that they saw no reason to repeat thework.

The scarcity of equations in this text is adequatelycompensated with many clear graphs and derivativeimages. For those interested in gaining a deeperunderstanding, the authors frequently refer thereader to seminal books and journal articles forthe details of spectral image analysis techniques andalgorithms (indeed, the book contains 6 full pages ofreferences). I did several times wish the book had anAppendix with a few of the more interesting oruseful algorithms written out. However, everychapter makes extensive use of boxed margin notesthat provide definitions, reminders, and useful tipsto keep in mind. The book’s figure captions are veryclearly written, providing detailed descriptions thatnicely supplement the main text.

The majority of images in this book are black andwhite, with only 9 color plates bound into the centerof the book. Color versions of the majority of theseblack and white figures are available for viewing ona Cambridge University Press web site dedicated to

geo.2007.08.001

this publication. It is ideal if the reader has aninternet connection conveniently available to viewthese color images while reading the book.

Chapter 1, ‘‘Extracting information from spectralimages’’ (38 pages), stresses the importance ofinvoking physical models that relate the remotesensing measurements with the materials andprocesses on the ground. The authors start withthe simplest physical process, shadows and shading,as an introduction to the complexities of spectralanalysis. The chapter includes a discussion of testingand validation of results, including why results gowrong, and the tradeoffs between accuracy, spatialscale, and detectability. The chapter closes with avery useful section summarizing the steps forextracting information from spectral images.

Chapter 2, ‘‘Spectroscopy of landscapes’’ (26pages), introduces the basics of reflectance andemittance spectroscopy, including a discussion ofthe differences between full-spectrum laboratoryreference measurements and the limited channels ofremote sensing equipment. In the section on spectralmixing, the reader is introduced to Clark’s Law,‘‘First find out what is there, then worry about howmuch’’, a theme that repeats throughout this text-book. Another recurring theme is that spectralmodeling should begin with a theoretical physicalmodel of field observations (the hypothesis), andthen the hypothesis should be tested for agreementwith the remote sensing data.

Chapter 3, ‘‘Standard methods for analyzingspectral images’’ (61 pages), discusses which algo-rithms work under what circumstances and illus-trates several common pitfalls. Most of thesealgorithms serve to enhance certain characteristicsof interest, for instance photosynthetic vegetation,but these algorithms are mostly empirical and donot actually invoke physical models. In this chapter,we are introduced to a synthetic image designed bythe authors to provide a ‘‘truth’’ dataset againstwhich the various algorithms can be tested. It wasenlightening indeed to see how badly many of the

Page 2: J.B. Adams, A.R. Gillespie, ,Remote Sensing of Landscapes with Spectral Images (2006) Cambridge University Press,Cambridge, UK 0-521-66221-4 362pp., US$85, Hardback.

ARTICLE IN PRESSBook review / Computers & Geosciences 34 (2008) 422–423 423

popular classification schemes failed on this syn-thetic dataset, primarily due to the effects of spectralmixing at sub-pixel scales. In this chapter, my abilityto keep track of 3-letter acronyms was challenged:PCT, TCT, MSS, RMS, NPV, SMA, NIR, TIR,TOA, RGB, HSI, DEM, CCD, GIS, among otherTLAs.

Chapter 4, ‘‘Spectral mixture analysis’’ (42pages), is an excellent tutorial on the fundamentalsof SMA. The explanation of selection criteria forendmembers was particularly good, as was theexplanation of the causes and significance of over-flow fractions (like ‘‘negative shade’’ when data falloutside a convex hull scatterplot). Their advice forfinding the best SMA endmember candidates wassobering, since the mathematical methods fordetermining endmember fractions become unstablefor more than a few endmembers, and we geologistsoften wish to create maps with dozens of discreteunits. They explain why not all analytically correctendmembers make sense for interpreting an image,while ‘‘virtual’’ or imaginary endmembers mayresult in better interpretations.

Chapter 5 is titled ‘‘Fraction images of land-scapes’’ (24 pages) and discusses how to link SMAcomponent fraction image data to landscape proper-ties on the ground. The authors use Landsat imagesof the Seattle area, the Gifford-Pinchot NationalForest, and Manaus (Brazil), plus Viking LanderMars images to illustrate the power of contraststretching, mapping overflow fractions, normaliza-tion, and false-color compositing to bring outfeatures of interest. In the section on classificationusing endmember fractions, they again stress that itmay not be a good idea to apply standard classifiersto fraction images because the resulting groupingsmay not make sense in terms of ground-truth data.

Chapter 6, ‘‘Target detection’’ (52 pages), ex-plores the factors that inhibit our ability to identifythe features we wish to map, especially the similarityof our target material to ubiquitous backgroundmaterials. The authors first illustrate why spectralcontrast between a target and its background isrequired for detection. The rest of the chapterexplores the use of several techniques for high-lighting spectral contrast, plus a section on detec-tion limits due to noise, pixel-to-pixel ‘‘clutter’’, andthe effects of pixel footprint size.

Chapter 7, ‘‘Thematic mapping of landscapes’’(54 pages), discusses the process of defining thespatial extents of target materials in spectral images

and creating thematic maps of those features. Theymake a clear distinction between ‘‘attribute maps’’,which show continuous properties like temperatureand which must be interpreted by the personviewing the image, and ‘‘unit maps’’ in which themap’s author has distilled the image into cleanboundaries between different materials, eliminatingthe need for the persons viewing the image tointerpret spectral attributes. The authors also stressthat the classes of materials that are detectable inspectral images may not actually be what we wish tofind on the ground, and for this reason compromisesin unit class labels will almost always be required:good to know, but again somewhat sobering. Thischapter steps the reader through several methods ofcreating thematic maps using the now-familiarLandsat images, with special attention to how thealgorithms can be tailored to extract the features ofinterest (or their close approximations, given theneed for compromises).

Chapter 8, ‘‘Processes and change’’ (39 pages),covers the concepts of landscape change throughtime using either multiple images spaced outthrough time or using theoretical/physical modelsof processes to predict where a pixel falls within anongoing process cycle (like reforestation or seasonalcrop growth).

The book also contains a 12-page glossary and 6pages of references. The Cambridge Press web sitelisted in the book’s preface is incorrect. The correctsite address is www.cambridge.org/9780521662215,and this is where the online resources can be found,including color figures with captions for eachchapter, the color reference images, a web pagecontaining reader questions and replies from theauthors, plus a short note on errata.

The authors based the content and organizationof this textbook on their combined teachingexperiences over the past 25 years. They haveclearly tried to include examples of interest to awide range of disciplines, including geology, for-estry, botany, urban planning, and planetarysciences. The authors assume a basic familiaritywith the fundamentals of remote sensing, and theyalso assume an undergraduate college-level back-ground in physics and algebra.

Gregory Benson3210 Highland Laurels, Kingwood, TX 77345, USA

E-mail address: [email protected]