Measuring and Modelling

Biodiversity from Space

Thomas W. Gillespie Giles M. Foody Duccio Rocchini Ana Paula Giorgi Sassan Saatchi

CCPR-063-08

December 2008 Latest Revised: December 2008

California Center for Population Research On-Line Working Paper Series

Progress in Physical Geography http://ppg.sagepub.com

Measuring and modelling biodiversity from space Thomas W. Gillespie, Giles M. Foody, Duccio Rocchini, Ana Paula Giorgi and Sassan Saatchi Progress in Physical Geography 2008; 32; 203 DOI: 10.1177/0309133308093606

The online version of this article can be found at: http://ppg.sagepub.com/cgi/content/abstract/32/2/203

Published by:

http://www.sagepublications.com

Additional services and information for Progress in Physical Geography can be found at:

Email Alerts: http://ppg.sagepub.com/cgi/alerts

Subscriptions: http://ppg.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.co.uk/journalsPermissions.nav

Citations (this article cites 138 articles hosted on the SAGE Journals Online and HighWire Press platforms): http://ppg.sagepub.com/cgi/content/refs/32/2/203

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Progress in Physical Geography 32(2) (2008) pp. 203–221

Measuring and modelling biodiversity from space

Thomas W. Gillespie,1* Giles M. Foody,2 Duccio Rocchini,3 Ana Paula Giorgi1 and Sassan Saatchi4

1Department of Geography, University of California Los Angeles, Los Angeles, CA 90095-1524, USA 2School of Geography, The University of Nottingham, University Park, Nottingham NG7 2RD, UK 3Dipartimento di Scienze Ambientali ‘G. Sarfatti’, Università degli Studi di Siena, Via Mattioli 4, 53100 Siena, Italy 4Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA, 91109, USA

Abstract: The Earth is undergoing an accelerated rate of native ecosystem conversion and degradation and there is increased interest in measuring and modelling biodiversity from space. Biogeographers have a long-standing interest in measuring patterns of species occurrence and distributional movements and an interest in modelling species distributions and patterns of diversity. Much progress has been made in identifying plant species from space using high-resolution satellites (QuickBird, IKONOS), while the measurement of species movements has become commonplace with the ARGOS satellite tracking system which has been used to track the movements of thousands of individual animals. There have been signifi cant advances in land-cover classifi cations by combining data from multi-passive and active sensors, and new classifi cation techniques. Species distribution modelling has been growing at a striking rate and the incorporation of spaceborne data on climate, topography, land cover, and vegetation structure has great potential to improve models. There have been signifi cant advances in modelling species richness, alpha diversity, and beta diversity using multisensors to quantify land-cover classifi cations and landscape metrics, measures of productivity, and measures of heterogeneity. Remote sensing of nature reserves can provide natural resources managers with near real-time data within and around reserves that can be used to support conservation efforts anywhere in the world. Future research should focus on incorporating recent spaceborne sensors, more extensive integration of available spaceborne imagery, and the collection and dissemination of high-quality fi eld data. This will improve our understanding of the distribution of life on earth.

Key words: biogeography, conservation planning, diversity modelling, remote sensing, species distribution modelling.

*Author for correspondence. Email: [email protected]

© 2008 SAGE Publications DOI: 10.1177/0309133308093606

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 204 Progress in Physical Geography 32(2)

I Introduction Thus, there is currently a lack of high- The Earth is undergoing an accelerated resolution data and maps for a number of re- rate of native ecosystem conversion and gions and biogeographers are continuing to degradation (Nepstad et al., 1999; Myers research ways to map species distributions et al., 2000; Achard et al., 2002) and there is and diversity that could have signifi cant appli- increased interest in measuring and model- cations for conservation planning (Foody, ling biodiversity from space (Nagendra, 2003; Whittaker et al., 2005). 2001; Kerr and Ostrovsky, 2003; Turner Remote sensing has considerable poten- et al., 2003). Biodiversity can be defined tial as a source of information on biodiversity as the variation of life forms within a given at landscape, regional, continental, and global ecosystem, region or the entire earth. How- spatial scales (Nagendra, 2001; Willis and ever, biodiversity is a multifaceted variable Whittaker, 2002; Turner et al., 2003). The and so one that can be diffi cult to measure main attractions of remote sensing as a and express simply (Duro et al., 2007). Bio- source of information on biodiversity are that geographers have long-standing interest in it offers an inexpensive means of deriving the distribution of biodiversity over different complete spatial coverage of environmental spatial and temporal scales (Whittaker et al., information for large areas in a consistent 2001; Lomolino et al., 2004). In particular, bio- manner that may be updated regularly geographers are interested in measuring or (Muldavin et al., 2001; Duro et al., 2007). quantifying patterns of species occurrence, Despite its well-established attractions and distribution, and distributional movements. potential, historically, remote sensing has Biogeographers are also interested in model- been relatively underused in studies of bio- ling or providing probability maps of species diversity (Innes and Koch, 1998; Trisurat distributions and patterns of diversity. et al., 2000). Recently, however, there has The most accurate ways to collect bio- been an increase in studies and reviews of geographical data on species distributions are bio-diversity taking advantage of advances in intensive ground surveys or inventories of sensor technology or focusing on broad pat- species in the fi eld. High-resolution maps of terns in variables related to biodiversity (Kerr species are available in the United Kingdom et al., 2001; Turner et al., 2003; Rocchini where inventories of plants and birds have et al., 2007; Saatchi et al., 2008). These been undertaken for over a decade at a 10 × 10 advances in remote sensing are generally km resolution (Gibbons et al., 1993). Plant and divided into direct and indirect approaches animal distribution data are also available at (Nagendra, 2001; Turner et al., 2003; Duro a 50 × 50 km resolution in Europe, Australia, et al., 2007). Direct approaches use space- the USA, Canada, and South Africa (Kidd borne sensors to identify either species, such and Ritchie, 2006; Finnie et al., 2007). How- as the identifi cation of tree species, or land- ever, these inventories require skilled indi- cover types, and directly map the distribution viduals, a signifi cant amount of time in the of species assemblages. Indirect approaches fi eld, and can be extremely expensive. Even use spaceborne sensors to model species dis- in relatively well-studied areas, different fi eld tributions and the distributions of diversity. data sources can lead to dissimilar or biased Both approaches have signifi cant applications maps of species distributions and diversity for species and ecosystem conservation that (Graham and Hijmans, 2006; Moerman and have still not been completely developed to Estabrook, 2006; Pautasso and McKinney, their full utility. 2007), and in areas such as the tropics spe- This research reviews recent and future cies occurrence and distribution data are advances in remote sensing that can be used relatively coarse and not well collected by biogeographers to measure and model bio- (Phillips et al., 2003; Schulman et al., 2007b). diversity patterns from spaceborne sensors.

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 205

First, we examine satellites currently being visible and infrared bands used in species used to measure and model biodiversity mapping. The NASA Landsat series is the from space. Second, we examine advances most widely used sensor for biodiversity in direct approaches for measuring species studies due to the ease in which the data can and land-cover classifications. Third, we be obtained, long time series, and low cost. examine advances in modelling patterns of The Landsat series has been used extensively species and diversity. Finally, we examine the in land-cover classifi cations, diversity models, applications of remote sensing methods for and conservation studies. However, Landsat conservation planning. ETM+ began to malfunction on 31 May 2003, ending 31 years of continuous Landsat II Spaceborne sensors series data. Other satellites and sensors such There has been a dramatic increase in earth as IRS, SPOT, and ASTER are becoming observation satellites and sensors over the more common; however, the lower number last seven years that have been used to of studies may reflect the higher cost and measure and model biodiversity from space availability of the data. The MODIS and (Table 1). Passive sensors, which record AVHRR sensors have provided extremely refl ected (visible and infrared wavelengths) useful data for regional, continental, and and emitted energy (thermal wavelengths), global studies of land-cover classification are most frequently used in biodiversity and diversity models. These sensors also studies. The highest spatial resolution data provide data on temperature, precipitation, comes from commercial satellites, such as and fi re that have been incorporated into bio- QuickBird and IKONOS, which contain diversity studies.

Table 1 Satellites with passive or active sensors that can be used to measure and model biodiversity from space

Satellite (sensor) Pixel size (m) Bands Cited in this review Passive sensors Spectral bands QuickBird 2 0.6, 2.5 5 7 IKONOS 2 1, 4 5 6 OrbView 3 1, 4 5 0 Landsat (TM, ETM+) 15, 30, 60, 120 7–8 42 IRS (LISS III) 5, 23, 70 5 4 EOS (ASTER) 15, 30, 90 14 3 SPOT 2.5, 10, 1150 5 2 EOS (Hyperion) 30 220 2 ALOS 2.5, 10 4 0 NOAA (AVHRR) 1100 5 8 EOS (MODIS) 250, 500, 1000 36 6 Active sensors Bands SRTM 30, 90 X, C 5 QSCAT 2500 Ku 2 Radarsat 9–100 C 1 SIR-C 10–200 X, C, L 1 TRMM (TMI) 18000 X, K, Ka, W 1 ERS-2 26 C 0 Envisat (ASAR) 30 C 0

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 206 Progress in Physical Geography 32(2)

Radar is the most common active space- phenology (Ramsey et al., 2005). There has borne sensor used in biodiversity studies. also been signifi cant progress in identifying Radar sensors send and receive a microwave tree canopies within forest ecosystems. For pulse in different wavelengths (ie, X-, C, instance, high-resolution data has been used L- bands) to create an image based on radar to identify mangrove species (Dahdouh- backscatter or interferometric radar can Guebas et al., 2004; Wang et al., 2004) and be used to provide high-resolution data on seven species of tree were classifi ed with an elevation and topography. Unlike passive overall accuracy of 86% in temperate forests sensors, radar can penetrate cloud cover, in Belgium (Carleer and Wolff, 2004). providing imagery both day and night re- Fine spatial resolution imagery (QuickBird, gardless of weather conditions. The Shuttle IKONOS, OrbView) from space has also Radar Topography Mission (SRTM) provides allowed researchers to address questions that 30–90 m resolution data on elevation and previously were impractical to study from topography that has been used in species and space or on the ground. It is now possible, diversity models. Radar backscatter from for instance, for studies to be undertaken at QSCAT, Radarsat-1 and SIR-C has been the scale of individual tree crowns over large used in land-cover classifi cation and diversity areas (Hurtt et al., 2003; Clark et al., 2004b). models. Such data have been used to quantify tree mortality in a tropical rainforest (Clark et al., III Measuring species and land-cover 2004a) and so may contribute usefully to classifi cations contentious debates on the issue. Moreover, it may sometimes be possible to achieve high 1 Species mapping levels of accuracy for some species from Early studies of species mapping used large- satellite as well as airborne sensor data scale aerial photography to identify individual (Carleer and Wolff, 2004). There is great plants, especially trees, to species. However, potential manually or digitally to identify tree there is an increasing desire to identify and species and canopy attributes from high- map species within landscapes from high- resolution imagery. High-resolution imagery resolution spaceborne sensors that have is collected primarily from commercial been launched in recent years (Sanchez- satellites that are still expensive to acquire Azofelfa et al., 2003; Turner et al., 2003; (US$3000–5000 for 10 km2). However, the Goodwin et al., 2005). From fine spatial cost should decrease with the competition resolution imagery it has been possible and an increasing number of archived images. to accurately identify some plant species Thus, it should be possible in the near future (Martin et al., 1998; Haara and Haarala, 2002; to identify and map temperate trees to a Carleer and Wolff, 2004; Foody et al., 2005). high degree of accuracy within a landscape Much progress has been made in identifying and selected tree canopies within stands of single species of plants, such as non-natives, tropical forest. that are of particular interest in natural The identifi cation of animals from space is resource management. QuickBird was used currently difficult because most of the to map giant reed (Arundo donax) in southern Earth’s species are smaller than the largest Texas with 86–100% accuracy (Everitt pixel of current public access satellites (0.6 m) et al., 2006). The spaceborne hyperspectral and revisit times are too infrequent for sensor Hyperion has shown potential for meaningful comparisons. However, meas- identifying the occurrence of select invasive urement of species movements has become species in the southeastern United States, commonplace with the advent of the such as Chinese tallow (Triadica sebifera), ARGOS satellite tracking system (Gillespie, to within 78% accuracy due to distinct leaf 2001). This tracking system uses polar

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 207 orbiting satellites and transmitters that are turtles) that are nearly impossible to track as small as 5 cm and weigh 49 g to provide in the fi eld (Deutsch et al., 2003; Ferraroli location data on the movement of species et al., 2004; Hawkes et al., 2007) (Figure 1). for over 500 days (Hawkes et al., 2007; As the costs and transmitters’ size continue Argos, 2008). It has been used to track the to decrease, this technology will become movements of thousands of individual more available and there is still great poten- animals. Between 2001 and 2007, over 70 tial to identify processes associated with peer-reviewed publications used this remote species movements by combining remote sensing tracking technology to improve our sensing data. knowledge of biogeography (Argos, 2008). Most terrestrial animal research has been 2 Land-cover classifi cation undertaken on raptors (ie, Steppe eagles) and The production of thematic maps of species large mammals (ie, Mongolian gazelles) in assemblages is one of the most common regions where it is difficult to track their applications of spaceborne remote sensing movements in the fi eld (Meyburg et al., 2003; (Foody, 2002). In particular, plant species as- Ito et al., 2005). There have also been rapid semblages and distributional patterns within advances in the study of marine mammals the landscapes, regions, and continents have (West Indian manatees) and reptiles (sea long been of interest to biogeographers (von

Figure 1 Reconstructed movements of 12 leatherback turtles (A–L) nesting in French Guiana and Suriname Source: Ferraroli et al. (2004).

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 208 Progress in Physical Geography 32(2)

Humbolt and Bonpland, 1805). Numerous the information provided may be more ac- large-area, multi-image-based, multiple- curate than suggested in the map’s summary sensor land-cover mapping programmes exist accuracy statement (DeFries and Los, 1999) that have resulted in robust and repeatable and some applications may require quite large-area land-cover classifi cations (Franklin modest levels of accuracy (Foody, 2008). and Wulder, 2002; Duro et al., 2007). There is also much to be gained by moving Franklin and Wulder (2002) undertook an away from conventional thematic mapping excellent review of large-scale land-cover practices. For example, one great advantage of classifi cations, such as CORINE and GAP, remote sensing is that the analysts can defi ne that generally seek to attain 85% accuracy and map the classes of interest to the appli- across all mapping classes using a variety cation in hand. There is, therefore, no need to of passive sensors (TM, SPOT, AVHRR, be constrained by the map legends. Similarly, MODIS) and to a lesser extent active sensors there is no need to be constrained to follow (RADARSAT, JERS). These land-cover clas- the standard image processing approaches to sifi cations provide direct measurements on mapping. Finally, there is considerable scope the distribution of species assemblages. Re- for different types of classifi cation analysis cently, there have been a number of advances for mapping. In particular, soft or fuzzy clas- in methods that can improve the resolution sifi cations have considerable potential. These and accuracy of land-cover classification. allow the study of environmental gradients Increased integration of radar data may sig- and transition zones and subpixel land cover nificantly improve classification accuracy (Foody, 1996; Rocchini and Ricotta, 2007). In (Saatchi et al., 2001; Boyd and Danson, addition, the use of soft classifi cations in post- 2005; Li and Chen, 2005). There have also classification change detection allows the been increased use of new classification study of land-cover modifi cations as well as techniques such as decision tree- and support conversions (Foody, 2001). This is particularly vector machine-based approaches and the valuable, as remote sensing has focused use of multilayer perception and radial basis on conversions, with little attention paid function neutral networks that signifi cantly to the severity of change limiting environ- improve accuracy (Foody, 2004a; Boyd mental applications (Nepstad et al., 1999; et al., 2006). Foody, 2001). There is a need for further research on information extraction techniques. This IV Modelling biodiversity includes continued development of image classifi ers for the derivation of accurate the- 1 Species distribution modelling matic maps. Contemporary approaches, such Species distribution modelling, also known as as those based on support vector machines ecological niche modelling, has been growing (Pal and Mather, 2005) appear to offer many at a striking rate in the last 20 years (Guisan attractions, especially if resources for training and Thuiller, 2005). Species distribution the classifi er are limited (Foody et al., 2006). models are based on presence, absence, or Attention is also needed on methodological abundance data from museum vouchers or issues such as accuracy assessment, a topic fi eld surveys and environmental predictors recognized as a major priority area for to create probability models of species dis- research (Rindfuss et al., 2004). The validity tributions within landscapes, regions, and of the maps derived from remote sensing continents (Guisan and Thuiller, 2005). A is a critical issue but is fraught with diffi culty review of 60 publications between 2001 (Foody, 2002). Critically, however, the re- and 2007 showed a majority developed and quired level of accuracy should be defi ned explained an approach or technique, evalu- for an application because in some instances ated an approach or compared modelling

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 209 approaches (ie, Maxent versus GARP), or increasingly being used in species distribution developed new ideas to improve the existing models, especially in the tropics (Chaves models. Most environmental predictors used et al., 2007; Buermann et al., 2008; Saatchi in these species distribution models have et al., 2008). Land-cover classifi cations col- been based on geographical information lected from spaceborne sensors have long system data over different scales (Figure 2). been used to link species distributions with However, there has been an increase in the vegetation types and associated habitat pre- incorporation of spaceborne remote sensing ference (Nagendra, 2001; Gottschalk et al., data on climate, topography, and land cover 2005; Leyequien et al., 2007). The greatest that has great potential to improve models of accuracy was found with non-mobile species species over different spatial scales (Turner such as plants (Pearson et al., 2004). How- et al., 2003). ever, vegetation maps as a surrogate for Climatic variables using geographical habitat preference have provided insights information system data sets (ie, WorldClim, into the distributions of birds (Peterson et al., BIOCLIM) are the primary environmental 2006), herpetofauna (Raxworthy et al., variables used in species distribution models, 2003), and insects (Luoto et al., 2002). especially for regions and continents (Elith Although the inclusion of suggested remote et al., 2006; Pearson et al., 2007). However, sensing indices or metrics can offer a great recently remote sensing data on precipi- amount of data to improve ecological studies, tation at 0.1 degree from NOAA satellites very few publications used remote sensing (Pearson et al., 2007) and 0.25 degrees from data (Turner et al., 2003; Pearson et al., 2004). Tropical Rainfall Mapping Mission (Saatchi Recently, there has been an increase in the et al., 2008) have been used in conjunction utility of spaceborne passive sensors data with ground-based measurements. This may such as leaf area index (Chaves et al., 2007) be superior to traditional GIS estimates of and percentage tree cover (Buermann et al., precipitation based on interpolation among 2008) for species distribution models widely dispersed climate stations in isolated (Figure 3). Active airborne sensors such as regions. Topography data has also been an airborne lidar have been used to improve important component of species distribution species distribution models by quantifying models (Pearson et al., 2004; Eltih et al., vegetation structure within a landscape 2006). Topography data is usually collected (Goetz et al., 2007). However, a number of from digitized elevation maps, but 90 m recent studies have used radar backscatter elevation and topography data are available from QSCAT (Buermann et al., 2008; at a near global extent due to the Shuttle Saatchi et al., 2008) and SIR (Bergen et al., Radar Topography Mission. This data is 2007) to improve species distribution models

Scale Domain Global Continental Regional Landscape Local Site Micro > 10000 km 2000-10000 km 200-2000 km 10-200 km 1-10 km 10-1000 m < 10 m Climate Topography Environmental Land-use Variable Soil Type Biotic Interaction Figure 2 Modelling and environmental variables by spatial scale Source: Pearson and Dawson (2003).

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 210 Progress in Physical Geography 32(2)

Figure 3 Maxent model of Carpornis melanocephala (Black-Headed Berryeater) in Brazil using remote sensing data, climate data, and a combination of both remote sensing and climate data by providing information on vegetation struc- diversity, which is the species diversity in ture. In the future, remote sensing data and one area, community, or ecosystem. Beta their derived indices should receive increas- diversity refers to the amount of turnover in ing attention from researchers applying species composition from one site to another species distribution modelling techniques. or identifi es taxa unique to each area, com- The inclusion of multiscale remote sensing munity, or ecosystem. Beta diversity is more data should allow researchers to improve closely related to changes in species simi- predictions over different scales, especially at larity or turnover with space. Typically, the landscape and regional scales. studies have focused on assessments of spe- cies richness with limited attention to other 2 Diversity models aspects such as species abundance and com- There have been a number of advances in position that are difficult to detect from modelling or predicting species richness, alpha spaceborne sensors (Foody and Cutler, diversity and beta diversity using multisensors 2003; Schmidtlein and Sassin, 2004). that examine relationships over different Information on species richness or diversity temporal and spatial scales with increasingly may be extracted from remotely sensed sophisticated methods to improve accuracy. data in a variety of ways such as land-cover The simplest measure of diversity is species classifications, measures of productivity, richness or the number of species per unit area and measures of heterogeneity (Nagendra, (ie, trees per hectare, birds per km2). The term 2001; Kerr and Ostrovsky, 2003; Leyequien diversity is more complex and technically et al., 2007). refers to a combination of species richness Many studies have related species richness and weighted abundance or evenness data or diversity to information on the land-cover and is generally quantified as an index mosaic of test sites derived from satellite (Simpson index, Shannon index or Fisher imagery (Nagendra and Gadgil, 1999a; alpha). These indices are used to defi ne alpha 1999b; Gould, 2000; Griffi ths et al., 2000;

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 211

Kerr et al., 2001; Oindo et al., 2003;Gottschalk diversity with remotely sensed data has been et al., 2005; Leyequien et al., 2007). Through sought. Most attention has focused on the use relationships with land-cover and habitat of the popular normalized difference vege- suitability, it is possible to assess the diversity tation index (NDVI) from passive sensors of species and assess impacts associated with because it is easy to calculate using the red changes in the habitat mosaic such as frag- and near infrared bands common to almost mentation based on landscape metrics (ie, all passive spaceborne sensors (Oindo and area and isolation) (Kerr et al., 2001; Luoto Skidmore, 2002; Seto et al., 2004; Gillespie, et al., 2002; 2004; Cohen and Goward, 2005; Lassau and Hochuli, 2007). NDVI has 2004; Fuller et al., 2007; Lassau and Hochuli, been associated with net primary productivity 2007). With such indirect approaches to bio- and has been hypothesized to quantify species diversity assessment, spatial resolution still richness and diversity based on the species- has an infl uence on a study as it impacts land- energy theory (Currie, 1991; Evans et al., cover classifi cation accuracy and indices of 2005). An increasing number of studies and landscape pattern (Foody, 2002; Millington reviews have found signifi cant associations et al., 2003; Saura, 2004) as well as the esti- between NDVI and diversity (Nagendra, mation of summary indices of biodiversity and 2001; Kerr and Ostrovsky, 2003; Leyequien estimates of composition (Kerr et al., 2001; et al., 2007). Many studies have reported Oindo et al., 2003). Nonetheless, even with significant positive correlations between relatively coarse spatial resolution imagery plant species richness or diversity from plot it is possible to derive useful information on or regions data and NDVI in both temperate diversity (Kerr et al., 2001; Foody and Cutler, (Fairbanks and McGwire, 2004; Levin 2003; Foody, 2004b; Cohen and Goward, et al., 2007; Rocchini, 2007a) and tropical 2004). ecosystems (Bawa et al., 2002; Gillespie, Alternatively, a direct relationship be- 2005; Feeley et al., 2005; Cayuela et al., tween measures of species richness and 2006) (Figure 4). NDVI can explain between

Figure 4 Predicted values of á tree diversity (Fisher’s alpha) in the Highlands of Chiapas, Mexico, and prioritization of areas for conservation based on identifi cation of high predicted á tree diversity within each fl oristic region Source: Cayuela et al. (2006).

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 212 Progress in Physical Geography 32(2)

30% and 87% of the variation in species While knowledge of species richness and richness or diversity within a vegetation type, alpha diversity represents crucial components landscape, or region. Results for terrestrial in diversity studies, the concept of beta diver- fauna are more complicated given the mo- sity (ie, the amount of species turnover) is also bility of faunal species and because NDVI important since it adds to the simpler con- does not directly quantify animal species but cept of alpha diversity the capability of species habitats (Leyequien et al., 2007). detecting spatial gradients that functionally Similar relationships between NDVI and di- act in determining the spatial variation in versity have been noted for animal taxa such species composition (Koleff et al., 2003; as birds and butterflies within landscapes Nekola and Brown, 2007). To date, few (Seto et al., 2004; Goetz et al., 2007) and efforts have been made to relate species regions (Hurlbert and Haskell, 2003; Foody, turnover to spectral variability, substantially 2004b; Ding et al., 2006; Bino et al., 2008). confining spectral variation hypothesis to However, NDVI does not always have a species richness prediction (Chust et al., positive relationship with animal species 2006; Cayuela et al., 2006). Tuomisto et al. richness and there is no consensus as to (2003) and Rocchini (2007a) built distance which scale results in the greatest accuracy. decay models replacing spatial distance Heterogeneity in land-cover types, by spectral ones, on the strength of the ex- spectral indices, and spectral variability de- pected high species turnover at high eco- rived from satellite imagery has also been cor- logical and thus spectral distance. Rocchini related with species richness (Gould, 2000; et al. (2005) derived species accumulation Rocchini, 2007b). This is largely based on the curves by ordering plots according to their hypothesis that heterogeneity in land cover, maximum spectral distance, thus accumu- spectral indices, or spectral variability within lating a higher number of species than an area or landscape is an indicator of habitat random curves given the same sampling heterogeneity which allows more species to effort (Figure 5) and promoting spectral coexist and hence greater species richness (Simpson, 1949; Palmer et al., 2002; Carlson Spectral Based Procedure et al., 2007; Rocchini et al., 2007). The vari- 500 ation in land-cover types within an area has been associated with species richness for a 480 number of taxa (Gould, 2000; Kerr et al., 460 Random Procedure 2001; Leyequien et al., 2007). Variation in 440 spectral indices has been shown to be pos- 420 itively associated with species richness and 400 diversity for a number of taxa in different 380 regions (Gould, 2000; Oindo and Skidmore, 360 2002; Fairbanks and McGwire, 2004; Levin 340 et al., 2007). More advanced techniques have Number of Species Accumulated 1 9 17 25 33 41 49 57 65 73 81 89 97 examined the variability of spectral signals in Number of Sampling Units satellite imagery which has been demon- strated to have an intrinsic power in evalu- Figure 5 Species accumulation curves. ating species diversity (ie, Spectral Variation Ordering plots on the strength of their Hypothesis; Palmer et al., 2002), since it is maximum spectral distance should expected that the higher the spectral vari- result in a higher number of species than ability is, the higher the habitat and species random curves, thus promoting spectral variability will be (Carlson et al., 2007; variability as a straightforward tool for Rocchini et al., 2007). inventorying species in a lower timelag

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 213 variability as a straightforward tool for inven- lichens (Waser et al., 2004), and mammals torying species in a lower timelag. Both (Oindo and Skidmore, 2002). While these examples demonstrated the powerfulness approaches provide a basic understanding of of using spectral distance between sites for patterns and can be used to create predictive beta diversity estimates and species inventory diversity maps for a landscape, region, or con- maximization. tinent, more sophisticated techniques are Most recently, there has been a move being examined and developed to model towards the use of multiple remote sensing patterns of diversity (Foody, 2004a; 2005). sensors over different time periods and General linear models and general additive increasingly sophisticated approaches to models have become increasingly important modelling diversity over different spatial in the spatial prediction of biodiversity pat- scales. Many remote sensing studies of di- terns; however, they have been poorly used versity to date have employed the use of one considering remote sensing data (Luoto et al., sensor at one period in time (ie, Gillespie, 2005; 2002; Schwarz and Zimmermann, 2005). Feeley et al., 2005; Gottschalk et al., 2005). Spatial statistics such as geographically However, increasingly diversity studies are weighted regression analyses have also undertaken using multiple passive sensors resulted in improved models of diversity (ie, Landsat, ASTER, QuickBird) (Levin et al., (Foody, 2005). Furthermore, increased ac- 2007; Rocchini, 2007b) or examine rela- curacy of predictions can be obtained using tionships with diversity over different time more complex approaches such as neural periods (Fairbanks and McGwire, 2004; networks (Foody and Cutler, 2006). Foody, 2005; Levin et al., 2007; Leyequien Finally, the effects of scale have long been et al., 2007). These studies are important in recognized as needing to be accounted for in the assessment of individual sensors and the biodiversity studies, but this remains a major effects of seasonality. There has also been an challenge (Whittaker et al., 2001; Willis and increasing interest in the combination of pas- Whittaker, 2002). Given the importance sive and active sensors to improve species of the spatial dimension to biogeographical diversity models. Active spaceborne sensors research (Millington et al., 2003) such scale- can provide data on the vegetation structure related issues are likely to be a major com- that has been associated with diversity, ponent of future research especially for especially avian diversity, across a number biogeographers interested in creating pre- of spatial scales (Imhoff et al., 1997; Bergen dictive diversity maps. While the ability to et al., 2007; Goetz et al., 2007; Leyequien provide complete data coverage for large et al., 2007). Recent advances in the model- areas is often seen as a major advantage of ling of species diversity with a combination of remote sensing, some problems of working passive sensors (MODIS) and active sensors with large areas have not been addressed. (QSCAT, SRTM) from satellites has also It is generally assumed that relationships be- been used to model tree diversity for the tween the biodiversity variable of interest and entire Amazon Basin (Saatchi et al., 2008). the remotely sensed response are spatially There has also been an increase in sophi- stationary and hence transferable between sticated statistical and spatial analyses to sites within the region of study. The spatial study diversity. The prediction of diversity resolution and scale dependence of relation- has substantially relied on simple univariate ships noted in the literature, however, indi- regression or multiple regression models cate that the relationships assessed may be appropriately scaling sensor imagery to spatially non-stationary (Foody, 2004b). fi eld data on vascular plants (Gould, 2000; The commonly made assumption that rela- Fairbanks and McGwire, 2004; Carter et al., tionships will remain spatially stationary may 2005; Rocchini, 2007b; Levin et al., 2007), be untenable and have a negative impact

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 214 Progress in Physical Geography 32(2) on the generalizabilty of remote sensing an exhaustively defi ned set of classes that is methods. Various methods may be used to commonly made in a standard classifi cation model non-stationary relationships and have analysis (Foody, 2004a). been applied in the modelling of wildlife dis- In recognition of the need to conserve bio- tributions from remote sensing (Foody, 2005; diversity, reserves and other such protected Osborne et al., 2007). Critically, however, areas have been formed. Remote sensing remote sensing offers the ability to obtain may have a major role to play in helping to pri- multiscale observations and data to explore oritize candidate locations for new reserves non-stationary relationships. (Schulman et al., 2007a). The conservation of biodiversity needs accurate and up-to-date V Conservation planning information (Knudby et al., 2007). Methods It is well established that biodiversity is to identify priority areas for conservation threatened greatly by human activity (Myers have generally focused on biological variables et al., 2000). In particular, land-cover (Shi et al., 2005) and often only relatively changes such as those linked to human- coarse biological information is needed to induced habitat loss, fragmentation, and identify priorities for conservation (Harris degradation represent the largest current et al., 2005). Frequently, what is required threat to biodiversity (Chapin et al., 2000; in conservation assessments is a quick but Menon et al., 2001; Gaston, 2005). Remote rigorous method to identify where human- sensing can be used to derive information induced threats and high biodiversity coin- on fragmentation, often in the form of land- cide (Ricketts and Imhoff, 2003). Remote scape pattern and shape indices calculated sensing offers a repeatable, systematic, and from a thematic map produced with an spatially exhaustive source of information image classifi cation analysis (Gillespie, 2005; on key variables such as productivity, dis- Lung and Schaab, 2006). Although valuable, turbance, and land cover that impact bio- the approach clearly requires an accurate diversity (Duro et al., 2007; Wright et al., classifi cation and the relationship between 2007). Moreover, the provision of data for classifi cation accuracy and landscape pattern large areas is especially attractive in remote index accuracy is not necessarily a simple and often inaccessible regions (Cayuela one (Foody, 2002; Langford et al., 2006). et al., 2006; Saatchi et al., 2008). As such, However, it is possible to tailor the process remote sensing is often a cost-effective to suit the circumstances of a particular con- data source (Luoto et al., 2004) and enables servation application such as certain land- rapid biodiversity assessments (Lassau and cover types. It is possible to focus attention Hochuli, 2007). on just these classes, saving time, effort and Remote sensing may also be valuable after resources that would otherwise be directed the establishment of reserves, not least be- on the classes of no interest. This is often cause competing pressures, such as those valuable in resource-limited conservation ap- associated with economic development and plications. As an example, the European population growth, place great stress on Union’s Habitats Directive seeks to maintain reserves and the surrounding lands (Nagendra the extent of valuable habitats on a no-net- et al., 2004). The spatial coverage provided by loss policy. Remote sensing may be used to remote sensing offers, however, the poten- monitor a habitat of interest with a one-class tial to monitor the effectiveness of protected classification approach adopted to focus areas, allowing comparisons of changes effort and resources on the class of interest inside and outside of reserves to be evalu- (Boyd et al., 2006; Sanchez-Hernandez et al., ated (Southworth et al., 2006; Wright 2007). This can also reduce problems asso- et al., 2007). The ability to monitor the areas ciated with not satisfying the assumptions of outside formally protected reserves is also

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 215 attractive as these may have a major role to from space. The increase in high-resolution play in conserving biodiversity (Putz et al., spectral satellites will make it possible to 2001). For example, even relatively severely acquire data at enhanced spatial (1 m), spec- logged forest outside a reserve may represent tral (visible, infrared, thermal), and radio- a signifi cant resource for biodiversity con- metric resolutions (11 bit) that can be used servation (Cannon et al., 1998) and secondary to map individual species. Indeed, Google forests are an often overlooked resource that Earth has led the way by providing QuickBird may be managed to help reduce pressures imagery (Loarie et al., 2008). Future, radar elsewhere (Bawa and Seidler, 1998). Thus, satellites may be ideal for studying species actions inside and outside the protected distributions and diversity patterns, especially areas are important, supporting the view that in regions with high cloud cover like the biodiversity conservation activities should tropics. There will be ten satellites (SAR- be undertaken at the level or scale of the Lupe, COSMO- SkyMed, TerraSAR-X) landscape (Nagendra and Gadgil, 1999b; launched by 2009 that provide elevation Margules and Pressey, 2000; Potvin et al., and radar backscatter data to 1 m pixel re- 2000; Hannah et al., 2002). This activity may solution (Gillespie et al., 2007). This will benefi t from remote sensing as its synoptic provide valuable multidimensional data sets overview provides information on the entire (vegetation structure, biomass, land-cover landscape. classifi cations) that should result in a richer Remote sensing may be a useful com- characterization of the environment than ponent to general biodiversity assessments, conventional passive image data sets. especially in providing data at appropriate The full information content of existing spatial and temporal scales. For example, the data sets is often not used in biodiversity biodiversity intactness index was proposed studies. There should perhaps be a move recently as a general indicator of the overall away from analyses based upon simple state of biodiversity to aid monitoring and summary indices that commonly underuse decision-making (Scholes and Biggs, 2005). spectral regions and are undertaken at a Although there are concerns for its use, single spatial scale (Asner et al., 2004). Bio- notably with the impacts of land degradation, geographers are perfectly positioned to take remote sensing may be an important source advantage of the different satellite data sets of data for its derivation (Rouget et al., that integrate climate, topography, spectral, 2006). and radar data over a landscape, regional, continental, and global spatial scale. This VI Conclusions would allow an increased understanding There can be no question that spaceborne of species distributions, land-cover classifi - imagery has made signifi cant contributions to cations, diversity models, and near real-time the science of biogeography and biodiversity conservation planning data across multi- over the last seven years. Future research spatial scales. should focus on incorporating recent and Finally, even if satellite imagery has been new spaceborne sensors, more extensive enthusiastically advocated as the resource of integration of available data from passive and the future for directly and indirectly investi- active imagery that can be used across spatial gating biodiversity from space, it is worth scales, and the collection and dissemination remembering that it should aim at sustaining of high-quality fi eld data. rather than replacing field-based meth- The recent developments in satellite odologies. Biogeographers should continue and sensor technology will further improve to collect and share high-quality data on our abilities directly and indirectly to study plants and animals including high-resolution biogeographical patterns of biodiversity location data that can be used in the future

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 216 Progress in Physical Geography 32(2) to test or validate models. There should be the Amazonian and Andean regions using remote an increasing number of data sets such as sensing data. Journal of Biogeography, in press. Synthesis and Analysis of Local Vegetation Cannon, C.H., Peart, D.R. and Leighton, M. 1998: Tree species diversity in commercially logged Inventories Across Scales (SALVIAS) where Bornean rainforest. Science 281, 1366–68. scientists can store and share data with the Carleer, A. and Wolff, E. 2004: Exploitation of scientifi c community. very high resolution satellite data for tree species For these reasons, it appears that bio- identification. Photogrammetric Engineering and geography as a discipline has a secure place Remote Sensing 70, 135–40. in science and should continue to improve Carlson, K.M., Asner, G.P., Hughes, R.F., Ostertag, R. and Martin, R.E. 2007: Hyperspectral our understanding of the distributions of life remote sensing of canopy biodiversity in Hawaiian on earth. lowland rainforests. Ecosystems 10, 536–49. Carter, G.A., Knapp, A.K., Anderson, J.E., References Hoch, G.A. and Smith, M.D. 2005: Indicators of Achard, F., Eva, H.D., Stibig, H.J., Mayaux, plant species richness in AVIRIS spectra of a mesic P., Gallego, J., Richards, T. and Malingreau, grassland. Remote Sensing of Environment 98, 304–16. J.P. 2002: Determination of deforestation rates Cayuela, L., Benayas, J.M., Justel, A. and Salas- of the world’s humid tropical forests. Science 297, Rey, J. 2006: Modelling tree diversity in a highly 999–1002. fragmented tropical montane landscape. Global Argos 2008: Tracking and monitoring. https:// Ecology and Biogeography 15, 602–13. www.argos-system.org/html/services/tracking- Chapin, F.S. III, Zavaleta, E.S., Eviner, V.T., monitoring_en.html (last accessed 28 April 2008). Naylor, R.L., Vitousek, P.M., Reynolds, H.L., Asner, G.P., Nepstad, D., Cardinot, G. and Ray, Hooper, D.U., Lavorel, S., Sala, O.E., Hobbie, D. 2004: Drought stress and carbon uptake in an S.E., Mack, M.C. and Diaz, S. 2000: Con- Amazon forest measured with spaceborne imaging sequences of changing biodiversity. Nature 405, spectroscopy, Proceedings of the National Academy 234–42. of Sciences of the United States of America 101, Chaves, J.A., Pollinger, J.P., Smith, T.B. and 6039–44. LeBuhn, G. 2007: The role of geography and Bawa, K., Rose, J., Ganeshaiah, K.N., Barve, N., ecology in shaping the phylogeography of the speckled Kiran, M.C. and Umashaanker, R. 2002: Asses- hummingbird (Adelomyia melanogenys) in Ecuador. sing biodiversity from space: an example from the Molecular Phylogenetics and Evolution 43, 795–807. Western Ghats, India. Conservation Ecology 6(2), 7. Chust, G., Chave, J., Condit, R., Aquilar, S., Bawa, K.S. and Seidler, R. 1998: Natural forest Lao, S. and Pérez, R. 2006: Determinants and management and conservation of biodiversity in spatial modeling of ß-diversity in a tropical forest tropical forests. Conservation Biology 12, 46–55. landscape in Panama. Journal of Vegetation Science Bergen, K.M., Gilboy, A.M. and Brown, D.G. 2007: 17, 83–92. Multi-dimensional vegetation structure in modeling Clark, D.B., Castro, C.S., Alvarado, L.D.A. and avian habitat. Ecological Informatics 2, 9–22. Read, J.M. 2004a: Quantifying mortality of tropical Bino, G., Levin, N., Darawshi, S., van der Hal, N., rain forest trees using high-spatial-resolution satellite Reich-Solomon, A. and Kark, S. 2008: Landsat data. Ecology Letters 7, 52–59. derived NDVI and spectral unmixing accurately Clark, D.B., Read, J.M., Clark, M.L., Cruz, A.M., predict bird species richness patterns in an urban Dotti, M.F. and Clark, D.A. 2004b: Application of landscape. International Journal of Remote Sensing, 1-M and 4-M resolution satellite data to ecological in press. studies of tropical rain forests. Ecological Applications Boyd, D.S. and Danson, F.M. 2005: Satellite remote 14, 61–74. sensing of forest resources: three decades of research Cohen, W.B. and Goward, S.N. 2004: Landsat’s development. Progress in Physical Geography 29, role in ecological applications of remote sensing. 1–26. Bioscience 54, 535–45. Boyd, D.S., Sanchez-Hernandez, C. and Foody, Colwell, R.K. and Coddington, J.A. 1994: Estimating G.M. 2006: Mapping a specific class for priority terrestrial biodiversity through extrapolation. habitats monitoring from satellite sensor data. Philosophical Transactions of the Royal Society of International Journal of Remote Sensing 27, 2631–44. London, Series B 345, 101–18. Buermann, W., Saatchi, S., Smith, T.B., Zutta, Currie, D.J. 1991: Energy and large-scale patterns B.R., Chaves, J.A., Mila, B. and Graham, of animal-and plant-species richness. American C.H. 2008: Predicting species distributions across Naturalist 137, 27–49.

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 217

Dahdouh-Guebas, F., van Hiel, E., Chan, J.C.-W., Finnie, T.J.R., Preston, C.D., Hill, M.O., Uotila, Jayatissa, L.P. and Koedam, N. 2004: Qualitative P. and Crawley, M.J. 2007: Floristic elements distinction of congeneric and introgressive mangrove in European vascular plants: an analysis based on species in mixed patchy forest assemblages using Atlas Florae Europaeae. Journal of Biogeography 34, high spatial resolution remotely sensed imagery 1848–72. (IKONOS). Systematics and Biodiversity 2, 113–19. Foody, G.M. 1996: Fuzzy modelling of vegetation from DeFries, R.S. and Los, S.O. 1999: Implications of remotely sensed imagery. Ecological Modelling 85, 3–12. land-cover misclassifi cation for parameter estimates — 2001: Monitoring the magnitude of land-cover in global land-surface models: an example from the change around the southern limits of the Sahara. simple biosphere model (SiB2). Photogrammetric Photogrammetric Engineering and Remote Sensing 67, Engineering and Remote Sensing 65, 1083–88. 841–47. Deutsch, C.J., Reid, J.P., Bonde, R.K., Easton, — 2002: Status of land cover classification accuracy D.E., Kochman, H.I. and O’Shea, T.J. 2003: assessment. Remote Sensing of Environment 80, Seasonal movements, migratory behavior, and site 185–201. fi delity of West Indian manatees along the Atlantic — 2003: Remote sensing of tropical forest environments: coast of the United States. Wildlife Monographs 151, towards the monitoring of environmental resources 1–77. for sustainable development. International Journal of Ding, T., Yuan, H., Geng, S., Koh, C. and Lee, P. Remote Sensing 24, 4035–46. 2006: Macro-scale bird species richness patterns of — 2004a: Supervised image classification by MLP East Asian mainland and islands: energy, area, and and RBF neural networks with and without an isolation. Journal of Biogeography 33, 683–93. exhaustively defined set of classes. International Duro, D., Coops, N.C., Wulder, M.A. and Han, Journal of Remote Sensing 25, 3091–104. T. 2007: Development of a large area biodiversity — 2004b: Spatial nonstationarity and scale-dependency monitoring system driven by remote sensing. Progress in the relationship between species richness and in Physical Geography 31, 235–60. environmental determinants for the sub-Saharan Elith, J., Graham, C.H., Anderson, R.P., Dudik, endemic avifauna. Global Ecology and Biogeography M., Ferrier, S., Guisan, A., Hijmans, R.J., 13, 315–20. Huettmann, F., Leathwick, J.R., Lehmann, A., — 2005: Mapping the richness and composition of Li, J., Lohmann, L.G., Loizelle, B.A., Manion, British breeding birds from coarse spatial resolution G., Moritz, C., Nakamura, M., Nakazawa, Y., satellite sensor imagery. International Journal of Overton, J., Townsend Peterson, A., Phillips, Remote Sensing 26, 3943–56. S.J., Richardson, K., Scachetti-Pereira, R., — 2008: Harshness in image classification accuracy Schapire, R.E., Soberon, J., Williams, S., assessment. International Journal of Remote Sensing, Wisz, M.S. and Zimmermann, N.E. 2006: Novel in press. methods improve prediction of species’ distributions Foody, G.M. and Cutler, M.E.J. 2003: Tree from occurrence data. Ecography 29, 129–51. biodiversity in protected and logged Bornean tropical Evans, K.L., Greenwood, J.J.D. and Gaston, K.J. rain forests and its measurement by satellite remote 2005: Dissecting the species-energy relationship. sensing. Journal of Biogeography 30, 1053–66. Proceedings of the Royal Society B –Biological Sciences — 2006: Mapping the species richness and composition 272, 2155–63. of tropical forests from remotely sensed data with Everitt, J.H., Yang, C. and Deloach, C.J. Jr 2006: neural networks. Ecological Modelling 195, 37–42. Remote sensing of giant reed with QuickBird satellite Foody, G.M., Atkinson, P.M., Gething, P.W., imagery. Journal of Aquatic Plant Management 43, Ravenhill, N.A. and Kelly, C.K. 2005: Identifi - 81–85. cation of specific tree species in ancient semi- Fairbanks, D.H.K. and McGwire, K.C. 2004: natural woodland from digital aerial sensor imagery. Patterns of floristic richness in vegetation com- Ecological Applications 15, 1233–44. munities of California: regional scale analysis with Foody, G.M., Mathur, A., Sanchez-Hernandez, multi-temporal NDVI. Global Ecology and Bio- C. and Boyd, D.S. 2006: Training set size require- geography 13, 221–35. ments for the classifi cation of a specifi c class. Remote Feeley, K.J., Gillespie, T.W. and Terborgh, J.W. Sensing of Environment 104, 1–14. 2005: The utility of spectral indices from Landsat Franklin, S.E. and Wulder, M.A. 2002: Remote ETM+ for measuring the structure and composition sensing methods in medium spatial resolution satellite of tropical dry forests. Biotropica 37, 508–19. data land cover classifi cation of large areas. Progress Ferraroli, S., Georges, J.Y., Gaspar, P. and Le in Physical Geography 26, 173–205. Maho, Y. 2004: Where leatherback turtles meet Fuller, R.M., Devereux, B.J., Gillings, S., Hill, fi sheries. Nature 249, 521–22. R.A. and Amable, G.S. 2007: Bird distributions

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 218 Progress in Physical Geography 32(2)

relative to remotely sensed habitats in Great Britain: Hawkes, L.A., Broderick, A.C., Coyne, M.S., towards a framework for national modeling. Journal Godfrey, M.H. and Godley, B.J. 2007: Only of Environmental Management 84, 586–605. some like it hot – quantifying the environmental Gaston, K.J. 2005: Biodiversity and extinction: species niche of the loggerhead sea turtle. Diversity and and people. Progress in Physical Geography 29, Distributions 13, 447–57. 239–47. Hurlbert, A.H. and Haskell, J.P. 2003: The effect Gibbons, D.W., Reid, J.B. and Chapman, R.A. of energy and seasonality on avian species richness 1993: The new atlas of breeding birds in Britain and and community composition. American Naturalist Ireland: 1988–1991. London: Poyser. 161, 83–97. Gillespie, T.W. 2001: Remote sensing of animals. Hurtt, G., Xiao, X.M., Keller, M., Palace, M., Progress in Physical Geography 25, 355–62. Asner, G.P., Braswell, R., Brondizio, E.S., — 2005: Predicting woody-plant species richness in Cardoso, M., Carvalho, C.J.R., Fearon, M.G., tropical dry forests: a case study from south Florida, Guild, L., Hagen, S., Hetrick, S., Moore, B., USA. Ecological Applications 15, 27–37. Nobre, C., Read, J.M., Sa, T., Schloss, A., Gillespie, T.W., Chu, J., Frankenberg, E. and Vourlitis, G. and Wickel, A.J. 2003: IKONOS Thomas, D. 2007: Assessment and prediction of imagery for the Large Scale Biosphere-Atmosphere natural hazards from satellite imagery. Progress in Experiment in Amazonia (LBA). Remote Sensing of Physical Geography 31, 459–70. Environment 88, 111–27. Goetz, S., Steinberg, D., Dubayah, R. and Blair, Imhoff, M.L., Sisk, T.D., Milne, A., Morgan, G. B. 2007: Laser remote sensing of canopy habitat and Orr, T. 1997: Remotely sensed indicators of heterogeneity as a predictor of bird species richness habitat heterogeneity: use of synthetic aperture in an eastern temperate forest, USA. Remote Sensing radar in mapping vegetation structure and bird of Environment 108, 254–63. habitat. Remote Sensing of Environment 60, 217–27. Goodwin, N., Turner, R. and Merton, R. 2005: Innes, J.L. and Koch, B. 1998: Forest biodiversity and Classifying Eucalyptus forests with high spatial its assessment by remote sensing. Global Ecology and and spectral resolution imagery: an investigation Biogeography 7, 397–419. of individual species and vegetation communities. Ito, T.Y., Miura, N., Lhagvasuren, B., Enkhbileg, Australian Journal of Botany 53, 337–45. D., Takatsuki, S., Tsunekawa, A. and Jiang, Gottschalk, T.K., Huettmann, F. and Ehler, M. Z. 2005: Preliminary evidence of a barrier effect of 2005: Thirty years of analyzing and modeling avian a railroad on the migration of Mongolian gazelles. habitat relationships using satellite imagery: a review. Conservation Biology 19, 945–48. International Journal of Remote Sensing 26, 2631–56. Kerr, J.T. and Ostrovsky, M. 2003: From space to Gould, W. 2000: Remote sensing of vegetation, plant species: ecological applications for remote sensing. species richness, and regional biodiversity hot spots. TRENDS in Ecology and Evolution 18, 299–305. Ecological Applications 10, 1861–70. Kerr, J.T., Southwood, T.R.E. and Cihlar, J. 2001: Graham, C.H. and Hijmans, R.J. 2006: A comparison Remotely sensed habitat diversity predicts butterfl y of methods for mapping species ranges and spe- species richness and community similarity in Canada. cies richness. Global Ecology and Biogeography 15, Proceedings of The National Academy of Sciences of 578–87. the United States of America 98, 11365–70. Griffi ths, G.H., Lee, J. and Eversham, B.C. 2000: Kidd, D.M. and Ritchie, M.G. 2006: Phylogeographic Landscape pattern and species richness: regional information systems: putting the geography into scale analysis from remote sensing. International phylogeography. Journal of Biogeography 33, 1851–65. Journal of Remote Sensing 21, 2685–704. Knudby, A., LeDrew, E. and Newman, C. 2007: Guisan, A. and Thuiller, W. 2005: Predicting species Progress in the use of remote sensing for coral reef distribution: offering more than simple habitat biodiversity studies. Progress in Physical Geography models. Ecology Letters 8, 993–1009. 31, 421–34. Haara, A. and Haarala, M. 2002: Tree species clas- Koleff, P., Gaston, K.J. and Lennon, J.J. 2003: sifi cation using semi-automatic delineation of trees Measuring beta diversity for presence-absence data. on aerial images. Scandinavian Journal of Forest Journal of Animal Ecology 72, 367–82. Research 17, 556–65. Langford, W.T., Gergel, S.E., Dietterich, T.G. Hannah, L., Midgley, G.F. and Millar, D. 2002: and Cohen, W. 2006: Map misclassifi cation can Climate change-integrated conservation strategies. cause large errors in landscape pattern indices: ex- Global Ecology and Biogeography 11, 485–95. amples from habitat fragmentation. Ecosystems 9, Harris, G.M., Jenkins, C.N. and Pimm, S.L. 474–88. 2005: Refi ning biodiversity conservation priorities. Lassau, S.A. and Hochuli, D.F. 2007: Associations Conservation Biology 19, 1957–68. between wasp communities and forest structure:

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 219

do strong local patterns hold across landscapes? conservation. ISPRS Journal of Photogrammetry and Austral Ecology 32, 656–62. Remote Sensing 57, 289–99. Levin, N., Shimida, A., Levanoni, O., Tamari, Moerman, D.E. and Estabrook, G.F. 2006: The H. and Kark, S. 2007: Predicting mountain plant botanist effect: counties with maximal species richness and rarity from space using satellite-derived richness tend to be home to universities and botanists. vegetation indices. Diversity and Distribution 13, 1–12. Journal of Biogeography 33, 1969–74. Leyequien, E., Verrelst, J., Slot, M., Schaepman- Muldavin, E.H., Neville, P. and Harper, G. 2001: Strub, G., Heitkonig, I.M.A. and Skidmore, Indices of grassland biodiversity in the Chihuahuan A. 2007: Capturing the fugitive: applying remote Desert ecoregion derived from remote sensing. Con- sensing to terrestrial animal distribution and diversity. servation Biology 15, 844–55. International Journal of Applied Earth Observation Myers, N., Mittermeier, R.A., Mittermeier, C.G., and Geoinformation 9, 1–20. Da Fonseca, G.A. and Kent, J. 2000: Biodiversity Li, J. and Chen, W. 2005: A rule-based method for hotspots for conservation priorities. Nature 403, mapping Canada’s wetlands using optical, radar, and 853–58. DEM data. International Journal of Remote Sensing Nagendra, H. 2001: Using remote sensing to assess 26, 5051–69. biodiversity. International Journal of Remote Sensing Loarie, S.R., Joppa, L.N. and Pimm, S.L. 2008: 22, 2377–400. Satellites miss environmental priorities. TRENDS in Nagendra, H. and Gadgil, M. 1999a: Biodiversity Ecology and Evolution, in press. assessment at multiple scales: linking remotely Lomolino, M.V., Sax, D.F. and Brown, J.H. 2004: sensed data with fi eld information. Proceedings of the Foundations of biogeography: classic papers with National Academy of Sciences of the United States of commentaries. Chicago: University of Chicago Press. America 96, 9154–58. Lung, T. and Schaab, G. 2006: Assessing fragmentation — 1999b: Satellite imagery as a tool for monitoring and disturbance of west Kenyan rainforests by means species diversity: an assessment. Journal of Applied of remotely sensed time series data and landscape Ecology 36, 388–97. metrics. African Journal of Ecology 44, 491–506. Nagendra, H., Tucker, C., Carlson, L., Luoto, M., Kuussaari, M. and Toivonen, T. 2002: Southworth, J., Karmacharya, M. and Karna, Modelling butterfly distribution based on remote B. 2004: Monitoring parks through remote sensing: sensing data. Journal of Biogeography 29, 1027–37. Studies in Nepal and Honduras. Environmental Luoto, M., Toivonen, T. and Heikkinen, R.K. 2002: Management 34, 748–60. Prediction of total and rare plant species richness in Nekola, J.C. and Brown, J.H. 2007: The wealth of agricultural landscapes from satellite images and species: ecological communities, complex systems topographic data. Landscape Ecology 17, 195–217. and the legacy of Frank Preston. Ecology Letters 10, Luoto, M., Virkkala, R., Heikkinen, R.K. and 188–96. Rainio, K. 2004: Predicting bird species richness Nepstad, D.C., Verissimo, A., Alencar, A., using remote sensing in boreal agricultural-forest Nobre, C., Lima, E., Lefebvre, P., Schlesinger, mosaics. Ecological Applications 14, 1946–62. P., Potter, C., Moutinho, P., Mendoza, E., Margules, C.R. and Pressey, R.L. 2000: Systematic Cochrane, M. and Brooks, V. 1999: Large-scale conservation planning. Nature 405, 243–53. impoverishment of Amazonian forests by logging Martin, M.E., Newman, S.D., Aber, J.D. and and fi re. Nature 398, 505–508. Congalton, R.G. 1998: Determining forest species Oindo, B.O. and Skidmore, A.K. 2002: Interannual composition using high spectral resolution remote variability of NDVI and species richness in Kenya. sensing data. Remote Sensing of Environment 65, International Journal of Remote Sensing 23, 285–98. 249–54. Oindo, B.O., Skidmore, A.K. and De Salvo, P. Menon, S., Pontius, R.G., Rose, J., Khan, M.L. 2003: Mapping habitat and biological diversity in the and Bawa, K.S. 2001: Identifying conservation- Maasai Mara ecosystem. International Journal of priority areas in the tropics: a land-use change Remote Sensing 24, 1053–69. modeling approach. Conservation Biology 15, 501–12. Osborne, P.E., Foody, G.M. and Suarez-Seoane, Meyburg, B.U., Paillat, P. and Meyburg, C. 2003: S. 2007: Non-stationarity and local approaches to Migration routes of Steppe Eagles between Asia modelling the distributions of wildlife. Diversity and and Africa: a study by means of satellite telemetry. Distributions 13, 313–23. Condor 105, 219–27. Pal, M. and Mather, P.M. 2005: Support vector Millington, A.C., Velez-Liendo, X.M. and machines for classification in remote sensing. Bradley, A.V. 2003: Scale dependence in multi- International Journal of Remote Sensing 26, 1007–11. temporal mapping of forest fragmentation in Bolivia: Palmer, M.W., Earls, P., Hoagland, B.W., White, implications for explaining temporal trends in P.S. and Wohlgemuth, T. 2002: Quantitative tools landscape ecology and applications to biodiversity for perfecting species lists. Environmetrics 13, 121–37.

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 220 Progress in Physical Geography 32(2)

Pautasso, M. and McKinney, M.L. 2007: The Rocchini, D. 2007b: Effects of spatial and spectral botanist effect revisited: plant species richness, resolution in estimating ecosystem á-diversity by county area, and human population size in the United satellite imagery. Remote Sensing of Environment 111, States. Conservation Biology 21, 1333–40. 423–34. Pearson, R.G. and Dawson, T.P. 2003: Predicting Rocchini, D. and Ricotta, C. 2007: Are landscapes the impacts of climate change on the distribution as crisp as we may think? Ecological Modelling 204, of species: are bioclimate envelope models useful? 535–39. Global Ecology and Biogeography 12, 361–71. Rocchini, D., Andreini Butini, S. and Chiarucci, A. Pearson, R.G., Dawson, T.P. and Liu, C. 2004: 2005: Maximizing plant species inventory effi ciency Modelling species distribution in Britain: a hierarchical by means of remotely sensed spectral distances. integration of climate and land-cover data. Ecography Global Ecology and Biogeography 14, 431–37. 27, 285–98. Rocchini, D., Ricotta, C. and Chiarucci, A. 2007: Pearson, R.G., Raxworthy, C.J., Nakamura, M. Using satellite imagery to assess plant species and Townsend Peterson, A. 2007: Predicting richness: the role of multispectral systems. Applied species distributions from small numbers of occur- Vegetation Science 10, 325–31. rence records: a test case using cryptic geckos in Rouget, M., Cowling, R.M., Vlok, J., Thompson, Madagascar. Journal of Biogeography 34, 102–17. M. and Balmford, A. 2006: Getting the biodiversity Peterson, A.T., Sánchez-Cordero, V., intactness index right: the importance of habitat Martínez-Meyer, E. and Navarro-Sigüenza, degradation data. Global Change Biology 12, A.G. 2006: Tracking population extirpations via 2032–36. melding ecological niche modeling with land-cover Saatchi, S., Agosti, D., Alger, K., Delabie, J. and information. Ecological Modelling 195, 229–36. Musinsky, J. 2001: Examining fragmentation and Phillips, O.L., Martínez, R.V., Vargas, P.N., loss of primary forest in Southern Bahian Atlantic Monteagudo, A.L., Zans, M.E.C., Galiano, forest of Brazil with radar imagery. Conservation W.G., Pena Cruz, A., Timana, M., Yli-Halla, Biology 15, 867–75. M. and Rose, S. 2003: Effi cient plot-based fl oristic Saatchi, S., Buermann, W., Mori, S., ter Steege, assessment of tropical forests. Journal of Tropical H. and Smith, T. 2008: Modeling distribution of Ecology 19, 629–45. Amazonian tree species and diversity using remote Potvin, F., Belanger, L. and Lowell, K. 2000: Marten sensing measurements. Remote Sensing of the Envir- habitat selection in a clearcut boreal landscape. onment, in press. Conservation Biology 14, 844–57. Sanchez-Azofelfa, G.A., Castro, K.L., Rivard, B., Putz, F.E., Blate, G.M., Redford, K.H., Fimbel, R. Kalascka, M.R. and Harriss, R.C. 2003: Remote and Robinson, J. 2001: Tropical forest management sensing research priorities in tropical dry forest envir- and conservation of biodiversity: an overview. Con- onments. Biotropica 35, 134–42. servation Biology 15, 7–20. Sanchez-Hernandez, C., Boyd, D.S. and Foody, Ramsey, E., Rangoonwala, A., Nelson, G., G.M. 2007: One-class classification for mapping Ehrlich, R. and Martella, K. 2005: Generation a specifi c land-cover class: SVDD classifi cation of and validation of characteristics spectra from EO1 fenland. IEEE Transactions on Geoscience and Remote Hyperion image data for detecting the occurrence Sensing 45, 1061–73. of the invasive species, Chinese tallow. International Saura, S. 2004: Effects of remote sensor spatial Journal of Remote Sensing 26, 1611–36. resolution and data aggregation on selected frag- Raxworthy, C.J., Martinez-Meyer, E., Horning, mentation indices. Landscape Ecology 19, 197–209. N., Nussbaum, R.A., Schneider, G.E., Ortega- Schmidtlein, S. and Sassin, J. 2004: Mapping of Huerta, M.A. and Peterson, A.T. 2003: Pre- continuous floristic gradients in grasslands using dicting distributions of known and unknown reptile hyperspectral imagery. Remote Sensing of Environ- species in Madagascar. Science 426, 837–41. ment 92, 126–38. Ricketts, T. and Imhoff, M. 2003: Biodiversity, urban Scholes, R.J. and Biggs, R. 2005: A biodiversity areas, and agriculture: locating priority ecoregions intactness index. Nature 434, 45–49. for conservation. Conservation Ecology 8(2), 1. Schulman, L., Ruokolainen, K., Junikka, L., Rindfuss, R.R., Walsh, S.J., Turner, B.L., Fox. Saaksjarvi, IE, Salo, M., Juvonen, S.K., Salo, J. and Mishra, V. 2004: Developing a science of J. and Higgins, M. 2007a: Amazonian biodiversity land change: challenges and methodological issues. and protected areas: do they meet? Biodiversity and Proceedings of the National Academy of Sciences of the Conservation 16, 3011–51. United States of America 101, 13976–81. Schulman, L., Toivonen, T. and Ruokolainen, Rocchini, D. 2007a: Distance decay in spectral space K. 2007b: Analysing botanical collecting efforts in analyzing ecosystem β-diversity. International in Amazonia and correcting for it in species range Journal of Remote Sensing 28, 2635–44. estimation. Journal of Biogeography 34, 1388–99.

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008 Thomas W. Gillespie et al.: Measuring and modelling biodiversity from space 221

Schwarz, M. and Zimmermann, N.E. 2005: A Turner, W., Spector, S., Gardiner, N., Fladeland, new GLM-based method for mapping tree cover M., Sterling, E. and Steininger, M. 2003: Remote continuous fi elds using regional MODIS refl ectance sensing for biodiversity science and conservation. data. Remote Sensing of Environment 95, 428–43. Trends in Ecology and Evolution 18, 306–14. Seto, K.C., Fleishman, E., Fay, J.P. and Betrus, von Humbolt, A. and Bonpland, A.J.A. 1805: Essai C.J. 2004: Linking spatial patterns of bird and sur la geographie des plantes. Paris: Levrault, Schoell butterfl y species richness with Landsat TM derived and Compagnie. NDVI. International Journal of Remote Sensing 25, Wang, L., Sousab, W.P., Gong, P. and Biging, 4309–24. G.S. 2004: Comparison of IKONOS and QuickBird Shi, H., Singh, A., Kant, S., Zhu, Z.L. and Waller, images for mapping mangrove species on the E. 2005: Integrating habitat status, human population Caribbean coast of Panama. Remote Sensing of the pressure, and protection status into biodiversity Environment 91, 432–40. conservation priority setting. Conservation Biology Waser, L.T., Stofer, S., Schwarz, M., Küchler, 19, 1273–85. M., Ivits, E. and Scheidegger, C.H. 2004: Simpson, E.H. 1949: Measurement of diversity. Prediction of biodiversity: regression of lichen species Nature 163, 688. richness on remote sensing data. Community Ecology Southworth, J., Nasendra, H. and Munroe, 5, 121–34. D.K. 2006: Are parks working? Exploring human- Whittaker, R.J., Araújo, M.B., Jepson, P., Ladle, environment tradeoffs in protected area conservation. R.J., Watson, J.E.M. and Willis, K.J. 2005: Applied Geography 26, 87–95. Conservation biogeography: Assessment and Trisurat, Y., Eiumnoh, A., Murai, S., Hussain, prospect. Diversity and Distribution 11, 3–23. M.Z. and Shrestha R.P. 2000: Improvement Whittaker, R.J., Willis, K.J. and Field, R. 2001: of tropical vegetation mapping using a remote Scale and species richness: towards a general, sensing technique: a case of Khao Yai National Park, hierarchical theory of species diversity. Journal of Thailand. International Journal of Remote Sensing 21, Biogeography 28, 453–70. 2031–42. Willis, K.J. and Whittaker, R.J. 2002: Species Tuomisto, H., Poulsen, A.D., Ruokolainen, diversity-scale matters. Science 295, 1245–48. K., Moran, R.C., Quintana, C., Celi, J. and Wright, S.J., Sanchez-Azofeifa, G.A., Portillo- Cañas, G. 2003: Linking fl oristic patterns with soil Quintero, C. and Davies, D. 2007: Poverty and heterogeneity and satellite imagery in Ecuadorian corruption compromise tropical forest reserves. Amazonia. Ecological Applications 13, 352–71. Ecological Applications 17, 1259–66.

Downloaded from http://ppg.sagepub.com at UCLA on September 22, 2008