conservation applications of astronaut photographs of ......national aeronautics and space...

876

Conservation Biology, Pages 876–884Volume 15, No. 4, August 2001

Conservation Applications of Astronaut Photographs ofEarth: Tidal-Flat Loss (Japan), Elephant Effects onVegetation (Botswana), and Seagrass andMangrove Monitoring (Australia)

JULIE A. ROBINSON,*†† KAMLESH P. LULLA,* MINORU KASHIWAGI,†MAGGIE SUZUKI,† M. DUANE NELLIS,‡ CHARLES E. BUSSING,§

WARREN J. LEE LONG,** AND LEN J. M

C

KENZIE**

*Earth Sciences and Image Analysis, Johnson Space Center, National Aeronautics and Space Administration,2400 NASA Rd 1, C23, Houston, TX 77058, U.S.A.†Tokyo Office, Japan Wetlands Action Network, 3–18–1 #105, Higashi-Toyoda, Hino-shi, Tokyo 191, Japan‡Department of Geology and Geography, and Eberly College of Arts and Sciences, West Virginia University,P.O. Box 6286, Morgantown, WV 26505, U.S.A.§Geography Department, Kansas State University, Manhattan, KS 66506, U.S.A.**Northern Fisheries Centre, Queensland Department of Primary Industries, P.O. Box 5396,Cairns, Qld 4870, Australia

Abstract:

National Aeronautics and Space Administration (NASA) photographs taken by astronauts fromlow Earth orbit can provide information relevant to conservation biology. This data source is now more ac-cessible because of improvements in digitizing technology, Internet file transfer, and availability of imageprocessing software. We present three examples of conservation-related projects that benefited from the use ofastronaut photographs. First, NASA scientists requested that astronauts photograph the area of the controver-sial Isahaya Bay reclamation project in Japan. Japanese researchers used photographs from before and afterthe reclamation as a tool for communication with the public about the effects of tidal-flat loss. The newly ac-quired images and the availability of high-resolution digital images from NASA archives provided timely pub-lic information on the observed changes. Second, we digitally classified and analyzed a Space Shuttle photo-graph of Chobe National Park in Botswana to identify the locations of woodlands affected by elephants. Fieldvalidation later confirmed that areas identified on the image showed evidence of elephant damage. Third, weused a summary map from intensive field surveys of seagrasses in Shoalwater Bay, Australia, as referencedata for a supervised classification of a digitized photograph taken from orbit. The classification distin-guished seagrasses, sediments, and mangroves with accuracy approximating that in studies using other satel-lite remote-sensing data. Astronaut photographs are in the public domain, and the database of nearly400,000 photographs from the late 1960s to the present is available at a single searchable location on the In-ternet (http://eol.jsc.nasa.gov/sseop). These photographs can be used by conservation biologists as a source ofgeneral information about the landscape and for quantitative mapping.

Usos en la Conservación de Fotografías Tomadas por Astronautas: Pérdida de Zonas de Mareas ( Japón), Efectosde Elefantes en la Vegetación (Botswana), y Monitoreo de Pastos Marinos y Manglares (Australia)

Resumen:

Las fotografías de la Administración Nacional de Aeronáutica y el Espacio (NASA) tomadas porastronautas en vuelos de órbita terrestre a baja altura pueden aportar información relevante para la bi-

††

email [email protected] submitted April 5, 2000; revised manuscript accepted January 31, 2001.

Conservation BiologyVolume 15, No. 4, August 2001

Robinson et al. Conservation Data from Astronaut Photographs

877

ología de la conservación. Esta fuente de datos es ahora más accesible debido a los avances en la tecnologíade digitalización, la transferencia de archivos por Internet y la disponibilidad de software para el procesa-miento de imágenes. Presentamos tres ejemplos de proyectos relacionados a la conservación que se han ben-eficiado del uso de fotografías tomadas por astronautas. (1) Científicos de la NASA requirieron que los astro-nautas fotografiaran el área del controvertido proyecto de reclamación de la controversial Bahía Isahaya enJapón. Los investigadores japoneses usaron fotografías anteriores y posteriores al proyecto de reclamacióncomo una herramienta de comunicación con el público sobre los efectos de la pérdida de praderas de mar-eas. Las imágenes recientes y la disponibilidad de imágenes digitales de alta resolución de los archivos de laNASA aportaron información pública oportuna sobre los cambios observados. (2) Clasificamos digitalmentey analizamos una fotografía tomada desde el cohete espacial del Parque Nacional Chobe en Botswana paraidentificar las localidades de bosques arbolados afectadas por elefantes. La validación de campo confirmómás tarde que las áreas identificadas en las imágenes mostraron evidencias del daño ocasionado por los ele-fantes. (3) Utilizamos un mapa sinóptico de los reconocimientos de pastos marinos en la Bahía Shoalwater,Australia como datos de referencia para una clasificación supervisada de una fotografía digitalizada to-mada en órbita. La clasificación distinguió pastos marinos, sedimentos y manglares con precisión aprox-imándose a estudios que usan otro tipo de datos de percepción remota. Las fotografías tomadas por astro-nautas son de dominio público y la base de datos de casi 400,000 fotografías de finales de 1960s hasta lafecha esta disponible en una sola ubicación de exploración de Internet (http://eol.jsc.nasa.gov/sseop). Estasfotografías pueden ser usadas por biólogos de la conservación como una fuente de información general so-

bre el paisaje y para mapeo cuantitativo.

Introduction

In a guiding document on conservation research for de-velopment agencies, the U.S. National Research Council(1992:5) noted that “Additional research and technicaldevelopment are needed to advance the utility of re-motely sensed data for ecosystem monitoring in devel-oping countries.” Since that time, technical expertiseand availability of remote sensing and geographic infor-mation systems (GIS) as tools for monitoring and con-serving biodiversity have spread widely throughout theworld.

Remote-sensing data can provide multiple observa-tions of the same area over time, are at a suitable spatialresolution, and can be linked with ground observations.The most common remote-sensing data sources in useare from automated satellites—Landsat, AVHRR (ad-vanced very high resolution radiometer), IRS (India Re-mote Sensing), and SPOT (Satellite pour l’Observationde la Terre). If a research question requires purchasingand analyzing multiple images, data acquisition can beprohibitively expensive because data are distributed atcost of acquisition (all users share the cost of operatingthe satellite and distributing the data, [Paulsson 1992],although the cost structure has been reduced signifi-cantly for U.S. Landsat-7 images). Data fully available inthe public domain, such as AVHRR, are at coarse resolu-tion (scale 1 pixel

�

1 km

2

) and are unsuitable for stud-ies of smaller areas.

A lesser-known and underused source of remote-sens-ing data are photographs of Earth taken by astronauts in

orbit (Wobber 1969; Helfert & Lulla 1989; Lulla et al.1996; Nedeltchev 1999). The photographs are avail-able in the public domain at the cost of reproduction.The nearly 400,000 images taken to date are catalogedin a database that can be searched via the Internet (http://eol.jsc.nasa.gov/sseop) with links to low-resolutionbrowse images. Astronaut photographs differ from auto-mated satellite data in several ways. First the data set be-gins with the early Mercury missions in the 1960s, farearlier than the automated satellites, and continues tothe present. Second, the data are photographic ratherthan collected by multispectral scanner. Photographsare usually taken with film cameras ranging from 35- to241-mm formats (a 70-mm format camera with 55

�

55 mm image size has been used for the majority of pho-tographs). Most photographs are taken with color positivefilm, but black and white, spectral bandwidth filters, andfalse color infrared have also been used. Before imagesare processed by computer, photographs must be digi-tized to create three bands, red, green, and blue. Third,the photographs are taken by astronauts out the win-dows of spacecraft and with several different lenses, sothey are much more variable in angle and scale than theautomated data. If an image is to be incorporated into aGIS, it must be georeferenced by the user. Finally, unlikecommonly used satellite data, photographs are taken at avariety of solar angles. Variable illumination can accenttopographic, geologic, vegetative, or cultural features,but it also makes quantitative comparisons of reflec-tance at a given location more challenging. Variations indate, time, look angle, and illumination also make the

878

Conservation Data from Astronaut Photographs Robinson et al.


creation of an image mosaic difficult, except when donewith photographs taken seconds apart.

We present three applications to illustrate the types ofinformation that can be obtained from astronaut photo-graphs and how that information can contribute to con-servation projects. By introducing this tool to conserva-tion biologists and providing examples of the ways inwhich astronaut photographs have contributed to largerprojects, we hope to aid other ongoing research activi-ties. First, as an example of visual documentation oflarge environmental modifications, we compared photo-graphs of Isahaya Bay, Japan, before and after the dikingof a 3000-ha tidal mudflat. Second, we used astronautphotographs to detect vegetation stress caused by denseelephant populations in Botswana. Finally, through su-pervised classification of a digitized photograph of Shoal-water Bay, Australia, we created a map of seagrass andmangrove habitats.

Documenting Wetland Loss in Isahaya Bay, Japan

To test the potential for rapid environmental assessmentthrough the use of astronaut photographs, we identifiedand tracked a large wetland area that was (1) known tobe of importance to migratory waterbirds (e.g., qualifiedto be listed on the Ramsar List of Wetlands of Interna-tional Importance, [Frazier 1999], or in one of the re-gional Shorebird Reserve Networks) and (2) subject toan immediate catastrophic threat. Through newspaperarticles and press releases from the Japan Wetlands Ac-tion Network ( JAWAN), we identified two such areas inJapan: Isahaya Bay (lat 32.9

�

N, long 130.2

�

E) in the Ari-ake Sea (Ariakekai) near Nagasaki and Fujimae tidal flat( lat 35.0

�

N, long 136.9

�

E) near Nagoya City.Plans to reclaim these areas of tidal mudflats have

been the focus of environmental activism by Japaneseand international nongovernmental conservation organi-zations, including JAWAN and World Wide Fund for Na-ture, that are opposed to destruction of the few remain-ing tidal flats in Japan. Based on use by migratorywaterbirds, activists consider these tidal flats importantlinks in the chain of wetlands used by migratory water-birds in the East Asian–Australasian Flyway (flyways inthe region are described by Anonymous 1996). Anotherconcern was the use of Isahaya Bay by species of conser-vation concern such as endangered Saunder’s Gulls(

Larus saundersi

), which winter in the area, and a localmudskipper (

Boleophthalmus pectinirostris

), which isclassified as vulnerable. At Isahaya Bay on 14 April 1997,293 steel slabs were used to complete a dike that cut offapproximately 3000 ha of tidal flats from the rest of theAriake Sea. In contrast, the proposal to reclaim the areaof Fujimae tidal flat was still under consideration by theAichi Prefecture (Suzuki 1998).

As part of real-time mission operations, we sent a re-

quest for photographs of these sites to the STS-90 SpaceShuttle crew in April 1998 (cf. Reilly et al. 1998). Wealso found photographs taken prior to the diking of Isa-haya Bay by searching the database of astronaut photo-graphs of Earth. We digitized photographs from second-generation film (copied from a master made from theoriginal) at 2400 pixels/in (10.5

�

m/pixel) with a flat-bed scanner. We used Photoshop (version 4.0, AdobeSystems Inc., Mountain View, California) to rotate, re-scale, and align the two images for optimal visual com-parison.

Two comparative photographs for Isahaya Bay areshown in Fig. 1. In the 1989 photograph (left), the shal-low area (arrow) appears similar to the deeper bay, withsubtle tan colors in the water, suggesting the presenceof either sediments or inundated mudflats. The dike canbe seen in the April 1998 photograph (right) as an un-naturally sharp boundary (arrow) between the blue ofthe bay and the extremely uniform light-colored regionthat has been separated from it. The former tidal flat ap-pears almost white because the mud flat has been iso-lated from tidal action and salts have been exposed byextended evaporation.

This visual pattern of sharp boundaries and high re-flectance would appear similar in remote-sensing imagesfrom other satellites if bands were selected and dis-played to approximate red, green, and blue colors. Al-though different for each image, the spatial resolution ofthe two images used is superior to that yielded by theLandsat thematic mapper (TM; 30 m/pixel). The 1989image has a digital spatial resolution of 9–13 m/pixel(calculated based on a 326-km altitude, 250-mm lens,55-mm original image size, and 50 line-pairs/mm film re-solving power), and the 1998 image has a digital spatialresolution of 17–25 m/pixel (calculated based on a 246-kmaltitude, 100-mm lens, 55-mm original image size, and 50line-pairs/mm film resolving power).

Astronauts were able to respond to information sentdaily from the ground and to take the requested photo-graph. The coverage by previous photographs was ex-tensive enough to allow us to find a comparative photo-graph showing the original state of Isahaya Bay, eventhough this had not been a specific target for previousmissions.

This application is an example of how astronaut pho-tographs can be a valuable source of public informationrelevant to conservation issues. The photographs in Fig.1 and a photograph of the Fujimae tidal flat area (elec-tronic still camera image S90e5239, not shown here)were used at public hearings to illustrate the magnitudeof tidal-flat loss in Japan (A. Tsuji, personal communica-tion) and have since been used in articles about tidal-flatprotection in Japan (Kashiwagi 1998). The example ofthe visible change in Isahaya Bay was presented at thetime local governments were evaluating plans to reclaimother tidal flats. For example, on 8 February 1999, after



879

receiving a large amount of public comment, the Mayorof Nagoya officially declared that the city would aban-don a proposed landfill project at the Fujimae Tidal Flatsand apply to have the area listed in the Ramsar Conven-tion on Wetlands (A. Tsuji, personal communication).

Identifying Areas of Elephant-DamagedVegetation, Botswana

As a side project accompanying construction of detailedland-use and land-cover maps for Botswana (Coleman etal. 1996), we sought a method for estimating locationswhere African Elephants (

Loxodonta africana

) weremodifying vegetation at scales relevant to mapping andpark management. Our area of interest was Chobe Na-tional Park, Botswana (lat 18.5

�

S, long 24.5

�

E). The parkhas robust and increasing elephant populations (50,000–80,000; Nellis et al. 1990; Ben-Shahar 1993; Herremans1995) that damage Zambezi teak (

Balklaiea plurijuga

)and associated forest species. The negative effects of ex-panding elephant populations on woodland vegetationare of concern (e.g., Ben-Shahar 1993, 1996

a

, 1996

b

;Herremans 1995), particularly to local land managersand national conservation planners seeking a sustainablebalance between wildlife populations and natural vege-tation.

After a search of the database of astronaut photogra-phy of Earth, we selected one photograph to demon-strate the use of this imagery in the detection of vegeta-tion stress (NASA photograph STS008–33–993; Fig. 2,left). The photograph shows part of the study area southof the Chobe River just west of its confluence with theZambezi River. It was taken on 5 September 1983 duringlocal spring and just before the onset of the rainy sea-son. We took this image into the field in 1989 for pre-liminary assessment, conducted image analysis in 1989

(Nellis et al. 1990), and then conducted qualitative fieldvalidation during three trips from 1990 to 1994.

A portion of the photograph was video digitized to512

�

512 pixels (30 pixels

�

20 km, or 667 m/pixel)with three-color bands (red, green, blue). This spatialresolution is far less than the photographic or maximumdigital resolution of approximately 24–35 m/pixel (cal-culated based on a 353-km altitude, 100-mm lens, 55-mmoriginal image size, and 50 line-pairs/mm film resolvingpower). Our subsequent analyses used an Earth Re-sources Data Analysis System (ERDAS) image-analysissystem. We identified six landscape types with an unsu-pervised cluster analysis and the multiple-pass isodataroutine (Fig. 2, right). Based on additional informationon soil and topographic position (Coleman et al. 1996),these vegetation units represented six classes: river, ri-parian–tall grass, elephant-damaged area, marginal grass–woodland, medium-density teak woodland, and high-density teak woodland. The classes follow sequentiallyfrom bare soil near the Chobe River (lightest areas inFig. 2), to increasingly dense vegetation, to dense wood-lands at higher elevations and greater distances from theriver. Lighter areas near the water corresponded to theareas most affected by elephants.

We extracted additional information about the spatialvariability in contrast in different parts of the imagethrough textural analysis (defined by Russ [1995] andused similarly to examine vegetation heterogeneity byBriggs & Nellis [1991]). We first used a density-slicingtechnique to examine the gross pattern of variation inthe different bands. The red band generated the greatestdegree of variation across the image, so we used it forsubsequent analyses. Empirical selection of the red bandfor texture analysis is supported by spectral analysis forother remote-sensing studies of vegetation in Botswana.Ringrose et al. (1990, 1999) determined that increases inreflectance indicating increased amounts of exposed soilwere best measured using the red band (Landsat MSS2

Figure 1. Reclamation of the Isa-haya Bay tidal flat off of Ariakekai, Japan, as seen from the Space Shut-tle. Both images are enlarged de-tails of extensive photographs. The image on the right is much more enlarged than the image on the left. Images have been rotated and scaled to facilitate comparison. Ar-rows indicate the area within Isa-haya Bay before and after the dike was constructed. National Aero-nautics and Space Administration photographs: left, STS034–78–048, 20 October 1989; right, STS090–739–079, 27 April 1998.

880



or TM3, roughly corresponding to the red band in thisdigitized photograph) and near infrared (not recordedby the film used for this photograph). Functionally, thered band corresponds to the region of chlorophyll absor-bance.

We measured texture by sequentially examining therange of values in the red band within a 3

�

3 pixel win-dow. We first determined means and standard devia-tions for red-band textural values for each of the majorlandscape units of the Chobe District that had beenidentified by the cluster analysis. The higher the texturalvalue, the greater the degree of contrast in the land-scape unit. We then compared the contrast in variousparts of the image with these means to identify patchi-ness. The mean textural values were similar throughoutthe photograph (Nellis et al. 1990). The standard devia-tion was greater, however, for areas we believed to beaffected by elephants than for other woodland vegeta-tion classes (elephant area SD

�

24, high-density wood-land SD

�

7, marginal grass–woodland SD

�

3). The dif-ference in standard deviation makes functional sensebased on the behavior of elephants. Elephants concen-trate in areas based on their accessibility to water. In theareas of concentration, vegetation damage is highly vari-able and irregular in pattern (Ben-Shahar 1993). Thisvariable vegetation damage corresponds to the irregularpatchiness identified from the astronaut photograph.

A team of field scientists who were field-validating anumber of GIS products (Coleman et al. 1996) took boththe original astronaut photograph and the classified im-age to Botswana during several field campaigns (1990–1994). They qualitatively verified the vegetation classesand the location and condition of areas identified as po-tentially affected by elephants. Elephant effects weregreater near permanent water sources (see also Ben-Sha-har 1993; Verlinden & Gavor 1998) near the Chobe river(Fig. 2, location a). Locations (b) and (c) in Fig. 2 indi-cate areas away from the water that are affected by ele-phants. Qualitative verification results indicate that wecould detect major vegetation effects of elephants in thishabitat using the astronaut photograph.

A useful extension of the analyses presented would beto add the texture measure to the three original bands(red, green, and blue) as a fourth band and then reclus-ter the data. Such an approach should improve the abil-ity to identify elephant-damaged areas and distinguishthem from similar but undamaged vegetation. Further-more, a quantitative accuracy assessment is needed tofully establish the use of texture analysis with astronautphotographs. Unfortunately, project constraints preventedadditional analyses or quantitative field verification, butwe believe that our qualitative verification is sufficientto justify the use of this analysis technique for studies ofother large-scale vegetation disturbances.

Remote Sensing of Seagrass Beds,Shoalwater Bay, Australia

Submerged aquatic vegetation can serve as an indicatorfor coastal environmental monitoring (Dennison et al.1993). Seagrasses are important ecological indicators be-cause they support coastal fisheries, stabilize sedimentsto maintain water clarity (which links them to coral reefhealth), and support endangered species such as sea tur-tles and dugongs (Hatcher et al. 1989; Lee Long et al.1996b). Landsat data and aerial photographs have beenused to differentiate seagrass meadows from sandy bot-tom areas and to monitor changes in cover types (e.g.,Ferguson et al. 1993; Luczkovich et al. 1993). We testedwhether astronaut photographs could serve similar pur-poses.

Our objective was to see if astronaut photographscould be used for identification of seagrass and man-grove habitats. We studied seagrass meadows in Shoal-water Bay, in tropical Queensland, Australia (lat 22.5

�

S,long 150.3

�

E). We chose this site because a seagrass maphad been compiled from detailed field surveys and an as-tronaut photograph suitable for classification was avail-able (250-mm lens, near vertical look angle, and sharpfocus; STS51D–45–63, April 1985). The photograph wasdigitized from second-generation film at 2400 pixels/in

Figure 2. Left: the Chobe River area, Northern Botswana, a por-tion of the Chobe District study area, and part of a more extensive photograph (National Aeronautics and Space Administration photo-graph STS008–33–993, September 1983). Right: Unsupervised image classification. White areas were predicted to have higher levels of el-ephant damage. Letters mark areas where vegetation damage by ele-phants was verified in the field.



881

(10.6

�

m/pixel) with a flatbed scanner. Georeferencingwas performed with ERDAS Imagine software (version8.3). We registered the digitized image to a base map(Lee Long et al. 1996b) by developing a first-order poly-nomial model with a minimum of 20 tie points identifi-able in the image and on the map. We first chose threeground-control points uniformly arrayed around the bay.Additional ground-control points were selected so as tobe uniformly distributed in and around the bay, withroot mean square error (RMS error) and total RMS errordetermined incrementally as each point was added (ERDAS1997; examples provided by McRay et al. 2000; Robin-son et al. 2000). Final total control-point error was4.1643 pixels. In the original digitized image, each pixelrepresented approximately 17

�

17 m on the ground(calculated based on a 417-km altitude, 250-mm lens,55-mm original image size, and digitized at 2400 pixels/in).The scale of the resampled image was reduced to matchthe map; after registration, the image conformed to mappixel sizes of approximately 143

�

143 m.The georeferenced image was then exported for classi-

fication with MultiSpec (PC–based) freeware (Landgrebe &Biehl 1997). We identified six habitat classes based onreferences to a map of seagrass habitats constructedfrom diver-based field surveys (Lee Long et al. 1996b,1997): seagrass, land, water, clouds, sediment, and man-grove. We selected 115 training fields, encompassing26479 pixels, and 43 test fields, encompassing 6528 pix-els, and performed maximum-likelihood classification.Class performance was estimated by the resubstitution

method and given in terms of percent accuracy for eachclass, overall percent accuracy, and an overall (esti-mate of kappa, the proportional reduction in error gen-erated by a classification process compared with the er-ror from a completely random classification; Congalton1991).

The georeferenced photograph of Shoalwater Bay isshown in Fig. 3. A relatively simple supervised classifica-tion correctly identified the majority of pixels (Table 1).Clouds and water could be distinguished and eliminatedfrom consideration with

�

93% accuracy based on testfields. When the entire image was analyzed, mangroveareas in the photograph were commonly misclassified asuplands (Table 1). We repeated the analysis after elimi-nating upland areas in the reference map (a shorelinemask), and identification improved from 59.8% to 93.5%accuracy based on test fields.

The greatest challenge in the image classification wasdistinguishing shallow, turbid (sediment-heavy) watersfrom seagrass beds. When the entire image was classi-fied, seagrasses were identified with relatively high accu-racy (84.3% based on test fields) but were also fre-quently misclassified as sediments (12.7% based on testfields; Table 1). But sediments were not well identified(11.3% accuracy based on test fields) and were oftenmisclassified as seagrasses (85.1% based on test fields). Alikely explanation for the difficulty in identifying shal-low areas that did not contain seagrass meadows rests inthe way we combined all seagrass species into a singlecategory. More sparsely growing seagrasses are unlikely

K̂

Figure 3. Left: Shoalwater Bay, Queensland (National Aeronautics and Space Administration photograph STS51D–45–63, 17 April 1985; color values shown have been standard-deviation stretched for display). Right: overlay of the summary map of seagrass meadows (rendered in grayscale, so black lines represent seagrass meadows and medium-gray shading represents mangroves; Lee Long et al. 1996b) and the classification results for the complete image (in color). In this analysis, species-specific information on seagrass composition (indicated by different black bar pat-terns) was not considered.

882



to be distinguishable from sediments in an image takenfrom orbit (see images of Shoalwater Bay seagrass mead-ows in Lee Long et al. [1997]). In the analysis with theshoreline mask, seagrass classification accuracy declinedand sediment classification accuracy improved to 61.7%based on test fields (Table 1).

The next stage of the seagrass analysis would be to re-peat the study using the full-resolution seagrass GIS de-veloped from diver-based field transects. In the reanaly-sis, predictions about which seagrass communities formmeadows that would be identifiable from astronaut pho-tographs could be tested. The GIS–based analysis wouldhave the further advantage of better preserving the reso-lution present in the original image before georeferenc-ing. Using an image acquired closer to the date of thefield surveys would also likely improve classification ac-curacy.

Discussion

Astronaut photographs have the potential to contributeto a range of conservation applications, from qualitativeimage interpretation to quantitative remote-sensing anal-ysis. They provide a visual context for large-scale envi-ronmental changes in a way that is easily communicatedto the public. They are also suited for quantitative identi-fication of vegetation cover in terrestrial and shallow-water systems.

The Isahaya Bay example demonstrates the capacity ofastronaut photographs to provide information on envi-ronmental change to the public. A better-informed pub-lic can have wide-ranging effects on conservation. It alsoillustrates the value of historical imagery and long-termdata collections such as astronaut photographs. With ap-proximately 375,000 images taken to date, older photo-graphs serve as valuable references on the state of the

environment over the last 30 years. In many cases, a suit-able comparative photograph can be identified, eventhough the area was not a specific research target in thepast. Information on all the photographs from Gemini,Apollo, Skylab, the Space Shuttle, Shuttle-

Mir

, and Inter-national Space Station missions are maintained in a sin-gle database that can be searched online with links tobrowse images and high-resolution digital images whenavailable (http://eol.jsc.nasa.gov/sseop).

The effects of human populations on the landscapeare the most visible population effects from orbit, butthe example of elephants in Botswana shows that cer-tain effects of other animal species can also be identi-fied. The particular image we employed (Fig. 2, left) wastaken with a 100-mm lens and represents moderate reso-lution compared to other NASA astronaut photographs.Further applications using astronaut photographs to ex-amine large areas of vegetation disturbance would beuseful. Patterns of clear-cut forestry in the tropical ortemperate regions (e.g., Amazonia, the Pacific North-west United States, and Siberia) and regeneration follow-ing hurricanes or fire would all be suitable studies. Im-age enhancement and analysis techniques used withother types of remotely sensed images can also be ap-plied to digitized astronaut photographs.

The most detailed example presented here, identifica-tion of seagrasses and mangroves, illustrates the poten-tial for astronaut photographs to be used for land-useclassification in much the same way other remote-sens-ing imagery is used. Our results compare favorably toseagrass classification studies based on Landsat thematicmapper (TM) data at other locations (Luczkovich et al.1993; Ferguson & Korfmacher 1997). Excellent resultshave been obtained for land-use classification of coastalhabitat types including mangroves (comparing LandsatTM and astronaut photographs; Webb et al. 2000). Im-provements in the optics of astronaut photography and

Table 1. Supervised classification

a

performance for digitized astronaut photograph of Shoalwater Bay, Queensland, Australia, 17 April 1985.

Class

Entire image classified Upland excluded

b

training accuracy% (

n

pixels)test accuracy% (

n

pixels)misclassification

c

% (class)test accuracy% (

n

pixels)misclassification

c

% (class)

Clouds 97.2 (886) 94.6 (129) 93.5 (77)Water 83.2 (16242) 93.1 (3509) 92.6 (3509)Seagrass 80.3 (1433) 84.3 (345) 12.7 (sediment) 43.8 (345) 38.3 (sediment)Sediment 26.7 (664) 11.3 (248) 85.1 (seagrass) 61.7 (248) 28.2 (seagrass)Land 77.4 (5626) 76.9 (1657) 22.0 (mangrove)Mangrove 51.6 (1628) 59.8 (640) 34.5 (land) 93.5 (613)Total 78.9 82.2 87.6

(%) 66.0 78.7 variance 0.0046 0.0048

a

Supervised classification is an analysis in which each pixel is allocated to one of a number of classes based on the values for the pixel in eachband or channel (here, red, green, and blue). The analysis is supervised because the class signatures are extracted from a training set of areasof known type.

b

See details in text. Number of pixels differ for some categories because of test fields that were within the area eliminated by the shoreline mask.

c

Primary misclassification categories are presented for cases at

�

10%.

K̂K̂



883

in image processing are expected to help reduce the de-gree of misclassification and increase the overall accu-racy of ecosystem mapping and monitoring. This studywas performed at a location where seagrasses are mostlydense and visible at low tide. In most locations, seagrasshabitat is of low density or submerged in turbid waterand would not be reliably detected by any remote sens-ing alone. Ground surveys will always be necessary tomonitor such habitats (e.g., Coles et al. 1996).

A major benefit of the astronaut photographs is thatthey are economical enough to use for visual interpreta-tion only, which is expensive with other remotelysensed imagery. Thus, scientists can use the images toget a general idea of a potential study area prior to fieldreconnaissance. Moreover, the images can be enhancedand visually interpreted with easy-to-use software suchas Adobe Photoshop. This feature makes astronaut pho-tographs a highly useful source of qualitative data, and itis available without researchers having to invest in ex-pensive and complicated image-analysis software. Thus,we strongly recommend that projects requiring any typeof geographical representation of a potential survey areashould make use of the database of astronaut photogra-phy of Earth, regardless of the image-analysis capabilitiesof the individual scientists.

For quantitative applications, the availability of Inter-net-based search engines and file transfer of digitized im-ages makes the imagery more accessible than ever. ThePC-based geographic information systems and image-analysis programs make it possible for a number of sci-entists to apply these images to research questions inways that would have been impossible just a few yearsago. There have not yet been enough studies to delin-eate the research questions that can most easily be ad-dressed using data from astronaut photographs. In gen-eral, we believe that applications requiring pixels of8–100 m in width (low contrast) are likely to be mostsuited to quantitative analyses.

Astronaut photographs provide a continuing, noncom-mercial, and public-domain record of environmentalchanges over the last 30 years. As a data source, they canbe combined through GIS with other soil or habitatmaps and even other remote-sensing data. Astronautphotographs can provide timely and important data forpublic information and can guide field surveys and habi-tat classification, all of which are crucial for conserva-tion biology and applied ecology.

Acknowledgments

Information and updates on the Fujimae Tidal Flats wereprovided by A. Tsuji. C.E.B. and M.D.N. were supportedduring their research by the U.S. Agency for Interna-tional Development. We would like to thank R. Allen forgeoreferencing the Shoalwater Bay photograph and E.

Sterling, E. L. Webb, and two anonymous reviewers forcomments on the manuscript.

Literature Cited

Anonymous. 1996. Asia-Pacific migratory waterbird conservation strat-egy: 1996–2000. Wetlands International—Asia Pacific, Kuala Lum-pur, publication 117; and International Waterfowl and WetlandsResearch Bureau, Japan Committee, Tokyo. Available at http://www.biodiversity.environment.gov.au/bg/environm/wetlands/mwp/apstrat.htm (accessed 8 August 2000).

Ben-Shahar, R. 1993. Patterns of elephant damage to vegetation innorthern Botswana. Biological Conservation

65:

249–256.Ben-Shahar, R. 1996

a.

Do elephants over-utilize mopane woodlands innorthern Botswana? Journal of Tropical Ecology

12:

505–515.Ben-Shahar, R. 1996

b.

Woodland dynamics under the influence of ele-phants and fire in northern Botswana. Vegetation

123:

153–163.Briggs, J. M., and M. D. Nellis. 1991. Seasonal variation of heterogene-

ity in the tallgrass prairie: a quantitative measure using remotesensing. Photogrammetric Engineering and Remote Sensing

57:

407–411.Coleman, T. L., M. D. Nellis, C. E. Bussing, M. Nkambwe, and S. Rin-

grose. 1996. Integrating remote sensing and GIS for natural re-source management in Botswana. Final report of the U.S. Agencyfor International Development Grant PCE–5053–G–00–3044–00.Alabama A & M University, Normal.

Coles, R. G., W. J. Lee Long, and L. J. McKenzie. 1996. A standardfor seagrass resource mapping and monitoring in Australia. Australianmarine data collection and management guidelines workshop, 1995.Commonwealth Scientific, Industrial, and Research OrganizationMarine Laboratories, Hobart, Tasmania. Available from http://www.environment.gov.au/marine/manuals_reports/standards/general/coles.html (accessed 20 July 1999).

Congalton, R. 1991. A review of assessing the accuracy of classificationof remotely sensed data. Remote Sensing of Environment

37:

35–46.Dennison, W. C., R. J. Orth, and K. A. Moore. 1993. Assessing water

quality with submersed aquatic vegetation: habitat requirements asbarometers of Chesapeake Bay health. BioScience

43

:86–94.Earth Resources Data Analysis System (ERDAS). 1997. Imagine field

guide. Atlanta.Ferguson, R. L., and K. Korfmacher. 1997. Remote sensing and GIS

analysis of seagrass meadows in North Carolina, USA. Aquatic Bot-any

58:

241–258.Ferguson, R. L., L. L. Wood, and D. B. Graham. 1993. Monitoring spa-

tial change in seagrass habitat with aerial photography. Photogram-metric Engineering and Remote Sensing

59:

1033–1038.Frazier, S., editor. 1999. Directory of wetlands of international im-

portance (CD-ROM). Wetlands International, Wageningen, The Neth-erlands. Available at http://www.wetlands.agro.nl/ramsar_database/ramsar.html (accessed 8 August 2000).

Hatcher, B. G., R. E. Johannes, and A. I. Robertson. 1989. Review of re-search relevant to the conservation of shallow tropical marine eco-systems. Oceanography and Marine Biology Annual Review

27:

337–414.Helfert, M. R., and K. P. Lulla, editors. 1989. Human-directed observa-

tions of the earth from space. Geocarto International

4:

2–79.Herremans, M. 1995. Effects of woodland modification by African ele-

phant

Loxodonta africana

on bird diversity in northern Botswana.Ecography

18:

440–454.Kashiwagi, M. 1998. Yuuryo-wa Umi-o Koete: Shuttle-ga mita Nihon-no

Higata (Concern spreads over the ocean: Japanese tidal flat as seenfrom Space Shuttles). SCIaS (Science Asahi Magazine)

17July:

80–81.Landgrebe, D., and L. Biehl. 1997. An introduction to MultiSpec Ver-

sion 7.97. Purdue Research Foundation, West Lafayette, Indiana.Available from http://dynamo.ecn.purdue.edu/

�

biehl/MultiSpec/documentation.html (accessed 26 May 1998).

884



Lee Long, W. J., L. J. McKenzie, M. A. Rasheed, and R. G. Coles. 1996

a

.Monitoring seagrasses in tropical ports and harbours. Pages 345–350 in J. Kuo, R. C. Phillips, D. I. Walker, and H. Kirkman, editors.Seagrass biology: proceedings of an international workshop, 1996.Faculty of Sciences, The University of Western Australia, Perth.

Lee Long, W. J., L. J. McKenzie, and R. G. Coles. 1996

b

. Seagrass studies.In Shoalwater Bay update: discovering the forgotten wetlands. ReefResearch

6(3)

. http://www.gbrmpa.gov.au/corp_site/info_services/publications/reef_research/issue3_96/3wot.html

.

Lee Long, W. J., L. J. McKenzie, and R. G. Coles. 1997. Seagrass com-munities in the Shoalwater Bay region, Queensland: Spring (Sep-tember) 1995 and Autumn (April) 1996. Information SeriesQI96042. Queensland Department of Primary Industries, Brisbane,Australia.

Luczkovich, J. J., T. W. Wagner, J. L. Michalek, and R. W. Stoffle. 1993.Discrimination of coral reefs, seagrass meadows, and sand bottomtypes from space: a Dominican Republic case study. Photogram-metric Engineering and Remote Sensing

59:

385–389.Lulla, K., C. Evans, D. Amsbury, J. Wilkinson, K. Willis, J. Caruana, C.

O’Neill, S. Runco, D. McLaughlin, M. Gaunce, M. F. McKay, and M.Trenchard. 1996. The Space Shuttle Earth observations photogra-phy database: an underutilized resource for global environmentalsciences. Environmental Geosciences

3:

40–44.McRay, B., J. Scott, and J. A. Robinson. 2000. Technical applications of

astronaut orbital photography: how to rectify multiple imagedatasets using GIS. Earth Observations and Imaging: National Aero-nautics and Space Administration, Washington, D.C. Available fromhttp://eol.jsc.nasa.gov/newsletter/gis.

National Research Council Board on Science and Technology for Inter-national Development. 1992. Conserving biodiversity: a researchagenda for development agencies. National Academy Press, Wash-ington, D.C.

Nedeltchev, N. M. 1999. Space Shuttle photography: a powerful toolof remote sensing investigation of human effects on the Earth. In-ternational Journal of Remote Sensing

20:

183–188.

Nellis, M. D., K. Lulla, J. M. Briggs, and C. E. Bussing. 1990. Interfacinggeographic information systems and Space Shuttle photography formonitoring elephant impact in Botswana. Papers and Proceedingsof Applied Geography Conferences

13:

10–15.Paulsson, B. 1992. Urban applications of satellite remote sensing and

GIS analysis. The World Bank, Washington, D. C.Reilly, J. F., W. R. Muehlberger, P. W. Dickerson, H. H. Schmitt, and

J. W. Young. 1998. Looking back at Earth. Geotimes

11:

14–20.Ringrose, S., W. Matheson, F. Tempest, and T. Boyle. 1990. The devel-

opment and causes of range degradation features in southeastBotswana using multi-temporal Landsat MSS imagery. Photogram-metric Engineering and Remote Sensing

56:

1253–1262.Ringrose, S., S. Musisi-Nkambwe, T. Coleman, D. Nellis, and C. Bus-

sing. 1999. Use of Landsat thematic mapper data to assess seasonalrangeland changes in the southeast Kalahari, Botswana. Environ-mental Management

23:

125–138.Robinson, J. A., B. McRay, and K. P. Lulla. 2000. Twenty-eight years of

urban growth in North America quantified by analysis of photo-graphs from Apollo, Skylab and NASA-

Mir

. Pages 25–41, 269–270in K. P. Lulla, L. V. Dessinov, editors. Dynamic Earth environments:new observations from Shuttle-

Mir

missions. Wiley, New York.Russ, J. C. 1995. The image processing handbook, 2nd edition. CRC

Press, Boca Raton, Florida.Suzuki, M. 1998. 21st century committee questionnaire reveals “worst

100 wasteful public works projects.” Japan Environment Monitor

97-98:

8–15.Verlinden, A., and I. K. N. Gavor. 1998. Satellite tracking of elephants

in northern Botswana. African Journal of Ecology

36:

105–116.Webb, E. L., Ma. A. Evangelista, and J. A. Robinson. 2000. Digital land-

use classification using Space-Shuttle-acquired orbital photographs:a quantitative comparison with Landsat TM imagery of a coastal en-vironment, Chanthaburi, Thailand. Photogrammetric Engineeringand Remote Sensing

66:

1439–1449.Wobber, F. J. 1969. Orbital photographs applied to the environment.

Photogrammetric Engineering

35:

852–864.

conservation applications of astronaut photographs of ......national aeronautics and space...

Documents