Remote Sensing for Rapid Ecological Assessment in Mountain Environments: Landscape analysis of the Makalu Barun National Park and Conservation Area, Nepal
Robert Zomer*, Susan L. Ustin, and Jack D. Ives
*Corresponding Author
Robert Zomer
Center for Spatial
Technologies and Remote Sensing (CSTARS)
Dept. of Land, Air and Water Resources
University of California
Davis, CA 95616 USA
rjzomer@ucdavis.edu
The design and management of national parks and other protected areas requires a broad base of both geographic and geo-ecological information. This paper evaluates the effectiveness of satellite remote sensing for photogrammetric stereo-mapping and DEM extraction, within the context of a cost effective methodology for rapid ecological assessment in remote mountainous terrain. As a case study, a landscape analysis of the Makalu Barun National Park and Conservation Area of east Nepal was examined. The study area is a highly articulated and rugged mountain landscape, with extreme topographic relief and an elevational gradient spanning more than 8300m. A DEM extracted from stereo SPOT imagery resulted in an median disagreement of 58m when compared to a DEM generated from a conventionally digitized GIS dataset of topographic contours (Scale = 1:250,000). Visual comparison of the two DEM's showed substantial agreement at the landscape scale, while larger scale comparison of 100m contours revealed some localized differences. The SPOT extracted DEM provided equal or better orthorectification of SPOT imagery then the conventional DEM. Derivative landscape analysis outputs, such as hydrological modelling, drainage networks and watershed boundaries, compared well with results based upon the conventional dataset. Intermediate map products useful for field research and mapping included production and orthorectified satellite base map image. Additionally, a fused mulitsensor high resolution image of the study area, combining Landsat TM and SPOT imagery at the higher 10 m resolution, was orthorectified to produce a false-color satellite image map highlighting the spectral discrimination between land cover classes. The utility of this methodology within the context of both rapid ecological assessment and detailed landuse / land cover mapping is discussed.
Tables and Figures:
The design and management of national parks and other protected areas requires that policy makers and managers have access to a broad base of physiographic, geomorphic, and ecological information. Access to this type of data has become increasingly available in the developed countries, often on-line and at no cost. However, for many parks in resource-poor developing countries, reliable information on physical attributes, extent and type of vegetation, wildlife habitat, and/or land use patterns is lacking (Yonzon et al. 1991). Due to either physical remoteness, lack of funds and personnel for research, and/or political considerations, spatially referenced data sets for many protected areas are either unreliable or of a scale inadequate for the purposes of resource management or ecological research. In many cases, even basic geographic data, such as large-scale topographic maps, correct place names, or accurate feature locations may be unreliable or even unavailable. This can be especially true in remote, mountainous areas, such as the eastern Himalaya, where difficulty of access and rugged terrain hinder mapping efforts and ground surveys.
Landscape analysis based upon satellite remote sensing can provide an efficient and cost-effective method for acquiring up to date and accurate landscape level information ( Millette et al. 1995) for use by resource planners, researchers, and conservationists. Satellite remote sensing has been used to map and classify landuse and vegetation cover classes in various remote, inaccessible areas and mountainous regions (Frank 1988, Lal et al. 1991), including Nepal (Blamont and Méring 1987, Millette et al. 1995). Areas of high topographic relief, however, present unique problems due to highly variable illumination angles and reflection geometry (Leprieur et al. 1988). Spectral response patterns in mountainous areas are strongly dominated by topographic effects (Franklin 1991). Vegetation mapping, a prime function of environmental remote sensing and often an important management priority, can also prove difficult in mountainous regions due to a lack of unique spectral reflectance / absorbance patterns across variations in cover types (Frank 1988), and the merging of plant communities along altitudinal gradients. Inclusion of topographic data has been shown to improve prediction accuracy significantly (Talbot and Markon 1986, Frank 1988, Senoo et al. 1990). Digital terrain data, primarily elevation and aspect, has been used in combination with spectral data to classify vegetation in mountain environments (Frank and Thorn 1985, Frank 1988, Leprieur et al. 1988, Skidmore 1989, Senoo et al. 1990, Giles and Franklin 1994)). Equally important, the digital elevation model (DEM) provides an essential base layer for further analysis, as well as a common framework for the geo-referencing and orthorectification of satellite imagery, aerial photos, and other available datasets. Derivation of landscape features and physiographic attributes, e.g. slope, aspect or curvature profile maps, based upon the DEM, is a primary component of GIS-based landscape analysis (Franklin 1989).
For many remote regions within the developing world adequate topographic data may not be available. Lack of this basic information hinders the ability to apply GIS spatial analysis to environmental management, and can be a significant gap in the required knowledge base for selection and design of new nature reserves. The digitizing of high relief mountain topographic data for inclusion into a georeferenced GIS database, particularly production of an adequately scaled digital elevation model (DEM), can be time consuming and expensive, and is often beyond the means of small or remote national parks, or other lesser known protected areas.
As part of a larger research effort to delineate a cost effective methodology for rapid ecological assessment in remote or rugged terrain (Zomer et al. 1998a, b), this paper evaluates the effectiveness of satellite remote sensing for photogrammetric stereo-mapping and DEM extraction within complex mountainous terrain with high topographic relief and an extreme elevation gradient. Intermediate map products, i.e. those designed to be useful to facilitate on-going fieldwork, including a DEM based topographic contour map, and derivative landscape analysis outputs including hydrological modelling of drainage networks and watershed boundaries were also produced. The extracted DEM and intermediate map products are evaluated through comparison with results obtained based upon a more conventionally digitized dataset (Scale = 1:250,000), available from the International Center for Mountain Research and Development (ICIMOD 1996), in Kathmandu. As and example of a useful intermediate map product, a fused multisensor high resolution satellite base map was produced as an field aid for vegetation and feature mapping. An outline of the utility of this methodology within the context of both rapid ecological assessment and detailed landuse / land cover mapping is discussed.
2. Study Site:
The Makalu Barun National Park and Conservation Area (MBNPCA) of eastern Nepal (approx. 86°40' - 87°30' E, 27°30 - 28°00 N) encompasses a wide diversity of habitats and bioclimatic regions, among which are several of the last remaining pristine inter-Himalayan valleys (Byers 1996). This relatively new national park (est. 1992) is a major addition to the formidable conservation efforts by the Government of Nepal to preserve biodiversity and threatened wilderness regions within the Himalaya (MBCP 1992). The park and its surrounding conservation buffer zone are a unique, biologically rich area encompassing over 2,330 sq. km, with habitats and ecotypes ranging from tropical monsoonal rain forests to alpine meadows and snowy peaks, some of which are among the highest in the world. Notable for the purposes of this study, an extremely steep elevational gradient stretches from the low "tropical zone" elevations at the southeast corner of the study area (approx. 350m asl) to the summit of Mt. Makalu, fifth highest mountain in the world at 8463 m. As a result, remote sensing images of the study area encompass the extreme of possible altitudinal change and local topographic relief, i.e. a steep gradient of more than 8300m of elevation change over less than 40 km horizontal distance. In addition to the expected effects of aspect on spectral response and radiance levels, cloud belts (which tend to hide the temperate zone cloud forest in most images), and differential atmospheric effects associated with the variation of the atmosphere by altitude, complicate remote sensing analysis within these steep mountains. Deeply cut and steep river valleys, and the deep shadows associated with them, present additional difficulties to image interpretation. At this time the park is remote and inaccessible by road, and has relatively few foreign visitors.
3. Materials and Methods:
3.1 Field sampling:
Six expeditions to the MBNPCA, between 1992 and 1997, conducted a generalized survey of forest ecology within the study area (Carpenter and Zomer 1996), and detailed vegetation studies on 256 forest quadrats (Zomer et al. 1998a). Nearly 200 of these quadrats, along with approximately 300 additional ground control points (GCP’s), were georeferenced using several GPS receivers, including Garmin GPS 45, GPS 75 and GPS 100. Auxiliary antenna extensions on 10 m extendable poles were used under tree canopy or within steep ravines, to improve signal strength and reduce positional error (RMS error approx. 100 m).
3.2 ICIMOD 1:250k GIS Data: Initial Processing
Several digitized GIS datasets for the study area were subset and clipped from the ICIMOD 1:250k GIS Database (ICIMOD 1996). These included a topographic "Contour" coverage of the MBNPCA, based upon maps published in 1988 by the Topographic Survey Branch of the Dept. of Survey, H.M.G. Nepal. From this topographic contour (i.e. vector) coverage, a raster-based DEM was generated using the TopoGrid function found within ArcInfo / GRID (UNIX Version 7.2; ESRI Inc., Redlands, CA). Spatial resolution for the resulting DEM was set to be 10 m grid cells. Other datasets relevant to this study were "Spot Heights", "Rivers", and the Protected Areas "Boundaries" coverage. All of the ICIMOD datasets are in UTM Projection (Everest Spheroid), and required reprojection to be converted into the more appropriate UTM Zone 45 coordinates used for this study.
3.3 Satellite Remote Sensing Datasets:
Two "Systeme Probatoire de l’Observation de la Terre", or SPOT (SPOT Image Corp.), panchromatic images were acquired of the study area, for the purposes of stereoscopic, software-based digital photogrammetry and DEM extraction. Spatial resolution for each of these single band images is a 10m grid cell size. The first SPOT image was taken on Dec 25, 1993, with an incidence angle of Left 14.8°. The second SPOT image was taken on Jan. 21, 1994, with an incidence angle of Right 20.6°. The base to height ratio for this stereopair was calculated to be 0.64, an acceptable ratio for satellite stereo-mapping. Both images were obtained as orbit-oriented, level 1A data, as required for subsequent DEM extraction and orthorectification.
One Landsat TM (EOSAT, Inc., Lanham, MD) multispectral (seven band) image (taken Sept. 22, 1996) was obtained which was relatively cloudless and included the entirety of the study area. This dataset was acquired as orbit-oriented, level 1A data, with only systematic geodetic processing, as required for subsequent orthorectification. The image was pre-processed using a nearest-neighbor resampling algorithm to a 25m grid cell size.
All remote sensing data was georeferenced to the ICIMOD 1:250k GIS Database. GCP’s (n = 113) were chosen using the "Rivers" and "Contour" coverages as reference, and based upon the ability to identify these points within the imagery. Landsat TM spectral bands 4, 5 and 7 (83.0 mm, 1.65 mm, 2.22 mm respectively) were enhanced with high pass filtering and contrast stretching, and displayed as an red-greeen-blue (RGB) image, to facilitate identification of GCP’s.
3.4 DEM Extraction
All satellite remote sensing based DEM’s were extracted using the Satellite Image Orthorectification Package (SDEM - Satellite DEM Automatic Extraction) found within the Easi/Pace software. The extraction of a DEM from a satisfactory stereopair is a multistep process (PCI Easi/Pace Manual 1998). First a mathematical model is generated from the satellite orbit data associated with each scene, and the GCP’s. This model is then used to create an epipolar projection for one of a pair of stereo SPOT images. This epipolar projection removes the y-parallax between left and right stereo images, however errors associated with GCP collection can affect these results. A automatic image matching technique is then used to extract the elevation for matching pixels within the respective datasets, and produce a DEM for the area of overlap between the images. A suite of DEM editing tools is available within the EASI/PACE package, including interpolation, filtering and smoothing functions, which allow for manual correction of obvious errors and interpolation of failed areas. For the purposes of this study, areas within the study zone where extraction failed were interpolated using the ICIMOD 1:250k topographic contour GIS dataset.
3.5 Orthorectification
All remote sensing datasets used were eventually orthorectified, based upon the stereopair extracted DEM, using the Satellite Image Orthorectification Package within the Easi/Pace software (PCI 1998). In order to facilitate a comparison of resulting accuracies and precision of registration, the first of the SPOT stereopair images (Dec. 25, 1993) was orthorectified twice. Once, using the DEM derived from the ICIMOD 1:250k GIS database, and then again using the stereopair-extracted DEM as the source of the terrain information. Digital orthorectification georeferences remote sensing data using a model of the terrain to account for parallax effects associated with incidence angle and the various aspect /slope angle combinations in the landscape. Orthorectification is particularly important in mountainous terrain, in order to achieve an acceptable level of geometric accuracy and to remove the spatial distortion associated with high relief.
3.6 Comparative Landscape Analysis
A series of derived map products and landscape analyses were produced based upon each of the DEM’s. These results were evaluated both in terms of their comparative accuracy, and the useful of the respective DEM’s for the purposes of that analysis. Three 10 x 10 km test sites were selected within the study area for large scale analysis of the results. The test sites were located within sectors of the DEM which had a minimal number of unclassified pixels (i.e. failed values), in order to minimize introduced errors. The three sites were chosen to be at low, mid, and high elevation ranges, and to be representative of topographic conditions as typically found within the park.
Contour lines were generated, at the 100m interval, for both DEM datasets. A stream coverage, mapping the major rivers and streams, was derived from each of the DEM datasets based upon a GIS cell-based hydrological modeling of stream flow (ESRI 1994). Watershed boundaries were then delineated in a further modeling step, based upon the results of the hydrological analysis. Three-dimensional terrain visualization (Graf et al. 1994), based on each of the DEM datasets, was used to visually explore the respective DEM’s, at both the landscape level, and within the test-sites. The GIS processing of these various analyses was performed using ArcInfo/GRID, and ArcView with Spatial Analyst, on a UNIX operating system. Terrain visualization was preformed using ArcView (ESRI Inc.) with 3D Analyst on UNIX, and Dimple on a Macintosh.
3.7 Production of High Spectral and Spatial Resolution Orthorectified Base Map
Orthorectified SPOT and TM were combined, or fused (Franklin and Blodgett 1993), using the FUSE module in PCI’s EASI/Pace software (PCI 1998). This method converts a Landsat TM RGB image to a hue-satuaration-intensity image (HSI), then replaces the intensity channel with the SPOT image, and reconverts back into a RGB image resampled at the higher spatial resolution of the SPOT imagery (10m). The resulting output merges the higher spatial resolution of the SPOT image with the spectral resolution (i.e. color discrimination) of the Landsat TM RGB image.
4. Results
The DEM extraction procedure allows for estimations of error at several stages within the extraction process. A measure of horizontal displacement is calculated by the SMODEL program for the selected GCP’s (n = 99) before the actual extraction process based upon the RMS residual error, estimated at 44 m, for the actual versus estimated x, y coordinates of these GCP’s. Residual error in this case refers to the mean difference between the conventionally produced and the stereo-imagery produced DEM. Upon the completion of the extraction algorithm, the difference in actual elevation of the GCP’S with that estimated in the DEM was calculated as a RMS residual error of 141 m. Sixty-seven of the GCP’s fell within the extracted overlap between the stereopair, and were compared with pixels which were successfully classified by the extraction matching procedure. Approximately 35% of all pixels within the extracted area were not classified (i.e. failed), as a result of various confounding circumstances, namely clouds and snow at higher altitudes that differed between scene acquisition dates. In order to facilitate subsequent analysis, these pixels were manually interpolated based upon the ICIMOD 1:250k GIS topographic contour dataset.
A comparison of estimated elevation as determined by the extracted DEM (Fig 2a) and the 1:250k topographic map-derived DEM (Fig 2b), was measured directly by calculating the difference between them. Using basic map algebra, the respective DEM’s were subtracted and their absolute value calculated, to determine the residual error. Results (Table 1) are given for: 1.) all originally classified pixels within the extracted area (before DEM editing); 2.) all pixels within the extracted area after DEM editing and interpolation; 3.) each of the three test sites (after DEM editing). For the entirety of the extracted area, initially classified pixels had an average difference of 127 m. This improved substantially after DEM editing to 106 m, however standard deviation (std. dev.) remained high at 140 m. Nonetheless, fifty percent of all pixels within this final DEM had a difference of less than 58 m., and almost 64% were less than 100 m difference. The low elevation test site showed the best results, with the lowest average difference (93 m) and the lowest std. dev. (91 m). The highest error values (average difference 148, std. dev. 170) were associated with the higher elevations and greater range in elevations found at the upper elevation test site.
Extraction of topographic contours from each DEM at the 100 m interval level provided visual comparison of the two DEM's. Results from the three test sites were similar and are presented for only the low elevation site (Fig 3 a & b). In general the contours show close correspondence between the two DEMs. The SPOT extracted DEM contours provide more topographic details and have a less stylized appearance than the 1:250K contours. Localized inaccuracies, however, exerted an exaggerated effect on derivation of contour lines, leading to several dense concentrations of closely spaced contours describing sharp peaks and holes in the terrain, most probably associated with errors in the DEM extraction.
The utility of the extracted DEM as a base reference for orthorectification of the SPOT panchromatic satellite imagery in high relief terrain was evaluated by comparing results of orthorectification, based upon each of the respective DEM’s (Figure 4 a & b). Only the Dec. 1993 SPOT panchromatic image was used for orthorectification. The ICIMOD 1:250k Rivers coverage was overlaid as a landmark reference feature by which to gauge accuracy and precision of registration of the orthorectified images. Results are presented for the middle elevation test site (Fig. 5). A more conventionally geocorrected , i.e. not orthorectified, image (Fig 5a) showed substantial feature distortion and spatial dislocation. This image was geocorrected based upon a 1° order polynomial model. Registration with the rivers coverage is poor, and appears markedly better for linear features going in a northwest direction. Registration was greatly improved in both orthorectified images. Both images displayed highly accurate registration of the river coverage with the image. The extracted DEM provided equal, or better, registration, comparing favorably with both overall results, and test site results. The other test sites displayed similar good results. The upper elevation site, however, is primarily glaciated, i.e. without rivers for reference, so that topographic contours were used to evaluate this site.
Hydrologic modelling of drainage networks and comparative registration of watershed boundaries displayed close congruence between the two DEM's. (Fig 6 and Fig. 7). However, drainage networks (Fig 6) reveal some effects of discontinuities in the extracted DEM, most probably associated with DEM editing and interpolation. These discontinuities along several of the larger rivers had an effect on watershed (basin) delineation, which resulted in some incongruencies at the watershed sub-division level (Fig 7).
Three dimensional terrain visualization was used to assess the nature of the landscape portrayal by the respective DEM’s, both for the landscape as a whole, and for each of the three test sites. Results are presented for the lower elevation test site (Fig 8). The 1:250k based DEM displayed a smoother, more stylized landscape, with realistic features and a great deal of coherence. The extracted DEM revealed a rougher, more articulated surface, however many slopes appear steep and cliff-like. This might not be unrealistic within the high topographic relief of the study area, but is likely exaggerated due to the variability associated with the extracted data. Overall, while it is easily evident that both are scenes representative of the same landscape, the curvature profile and textural roughness of the surface appear very different when examined at the larger scale of the test sites. This is less a factor at the landscape level (Fig 9), but still evident within hillshade and three-dimensional terrain visualization. These differences could also be a scale resolution effect between the data sets. Production of a high resolution base map image (i.e. the orthorectified and then fused Landsat TM and SPOT images) demonstrated that intermediate map products useful to researchers, managers, and policy makers could be produced rapidly (Fig 10). Based solely upon the orthorectified remote sensing data, various land features were easily and quickly mapped, e.g. lakes, glaciers, rivers, and bare rock which have distinct spectral characteristics that allow them to be easily extracted from the multispectral imagery. Additionally, land use and land cover change was analyzed based upon a similarly processed (i.e. orthorectified) historical remote sensing dataset (Landsat MSS; acquisition date 1972). The results of that investigation are reported separately in Zomer et al. (1998b).
5. Discussion
Accuracies attained in this study did not approach the precise and highly accurate results normally associated with the automated DEM extraction and orthoimage generation process, when based upon stereo SPOT imagery. For example, under nearly ideal circumstances, a large-scale mapping project in an arid region of northeastern Jordan (Al-Rousan 1997) reported sub-pixel accuracies of ± 3 to 4 m, over an elevation range of 1500. However, an assessment of the suitability of SPOT stereo data for relief analysis and morphometric studies as applied to slope instability and erosion problems in the Nepalese Himalaya (Grabmaier et al. 1988) produced error estimates for extracted elevation values that are comparable to the results presented here (i.e. 79% of all pixels within ±50 m). Even though these accuracies did not approximate the 10 to 20 m mark deemed as desirable, Grabmaier et al. concluded that the SPOT stereo measurements were more accurate than could be expected of the available topographic map of the area. They also remarked that, in the absence of ground truth data, there is no objective measure by which to assess the absolute accuracy of either the extracted DEM or the topographic map. By comparison, the Eastern Nepal Mapping Project (HMG Nepal Topographic Survey Branch 1996) required many years of effort to produce a series of highly accurate maps, and was funded through a bilateral agreement between the governments of Nepal and Finland. In the absence of such large-scale mapping efforts, a cost effective methodology of known precision and accuracy is a critically needed component for land management, GIS and rapid ecological assessment methodologies.
Although each of the SPOT stereopair scenes are rated as high quality, several site-specific issues confounded the extraction process in our study. Clouds and their shadows confound stereoscopic matching, so that even limited cloud cover can create difficulties with finding well-distributed control points and significantly interfere with complete mapping of selected study sites. This was particularly evident along temperate zone ridges within our study area, where cloud belts form regularly as a result of orographic topoclimatic conditions. Whereas mapping of the resultant cloud forests may be of particular interest, they are mostly obscured by clouds in all available images (dating back to 1972) of the study area. More significantly, difficulties were also caused by the 28 days interval between the two acquisition dates for the stereopair scenes. Notable differences with respect to snow cover were evident at elevations between 3000 to 6000 m, and this resulted in numerous failed tie points for these areas within the automatic DEM extraction. These difficulties were encountered particularly in the ecologically significant upper valleys, and near the various snowlines. The sensitivity of the extraction process to these difficulties highlights the need for careful selection of imagery. Seasonal constraints, including atmospheric haze, monsoonal cloud cover, and snowfall limit optimal acquisition dates in the eastern Himalaya (Grabmeir et al. 1988). Late autumn is most likely the optimal season for remote sensing data acquisition over the study area for the purposes of photogrammetric mapping.
Several issues confronted in attempting to produce the most accurate possible DEM with available resources are likely to be generic to low-cost spatial analysis efforts either in remote mountainous terrain, or within the lesser developed countries. Among the most difficult (within remote areas) of the processes required for DEM extraction is the collection of accurate GCP’s. Since groundtruthing in this study was based upon the identification of recognizable points in the ICIMOD 1:250k topographic contour and river coverages, GCP’s are roughly estimated to have a RMSE value of at least ±100 m in the x and y direction. Elevation, because it was estimated for GCP’s using the contour layer as a reference, possibly exceeds this value. The lack of geographic data and the error associated with geo-referencing of the imagery to small-scale datasets (e.g. 1:250k), limits the attainable geometric accuracy by introducing error associated with the selection of GCP’s, and significantly degrades the quality of the resulting extracted DEM. Differentially-corrected global positioning system (GPS) collection of high precision GCP’s would substantially improve the results of the DEM extraction. Similarly, availability of ancillary data such as spot heights with aerial photos for improved feature recognition could significantly enhance results.
The extracted DEM proved extremely useful as a terrain base for subsequent orthorectification of the SPOT stereopair image. The orthorectified images, compared with the SPOT image which was geocorrected based upon a more common first degree polynomial model, removed the panoramic distortion associated with the obligatory off-nadir angle of the SPOT stereopair images and the displacement due to terrain. The increased importance of orthorectification within mountainous terrain is well known, and is here demonstrated by a significantly improved geometric registration of the river line features. Whereas some degree of error is likely to be propagated from the DEM through to the orthorectification process, DEM accuracy proved sufficient to remove the major sources of distortion and displacement associated with the terrain surface. The ability to use an SPOT stereopair extracted DEM for othrorectification allows a complete remote sensing based solution to the production of a satellite ortho-map image.
Hydrologic modeling and watershed delineation was easily facilitated by the extracted DEM. Automated stream and watershed delineation demonstrated that, although DEM errors were propagated, results were sufficiently accurate that only minor data editing was necessary to produce an accurate streams delineation. Although river networks were sporadically discontinuous (due to errors in the DEM), resulting in sub-division of the watershed, these errors are readily identified. Gaps and inaccurate stream links are easily corrected to conform with projected river features by overlaying the results onto a orthorectified SPOT, or Landsat TM, image. A coherent watershed delineation is attained in a next step after only minor editing of the automated stream delineation.
Three-dimensional terrain visualization was able to significantly enhance cognitive feature recognition and provides an easily accessible intuitive understanding of the landscape (Graf et al., 1995). Terrain visualization is a powerful error detection tool when working with topographic data, as inconsistencies in the data become readily apparent to the eye when presented in the isometric or three dimensional view. Further, by draping landuse, vegetation or other data over the visualized surface, spatial distribution of land-cover types in relation to three-dimensional space can be interactively explored.
6. Conclusion
Overall, the results of this study
shows that remote sensing-based DEM extraction can provide essential geographic
information and a basis for analysis of remote, rugged, or mountainous
areas, especially in the absence of other data sources. Advantages ascribed
to this methodology include: cost-effectiveness, timeliness, and equal
access to high spatial resolution data. This methodology can quickly provide
a basis for building an accurate and powerful GIS land management database.
Digitized map products having the accuracies demonstrated in this study
would be of great value within most of the under-mapped developing countries
and remote mountainous areas of the world. For the purposes of rapid ecological
assessment, preliminary survey, or detailed field mapping, intermediate
map products are quickly produced, enhancing the value of fieldwork (and
GPS units) through the quick availability of georeferenced maps and images
for use in the field. By quickly providing a common framework for georeferencing,
early in a project cycle, the satellite ortho-image base map and extracted
DEM can ensure congruence in subsequent data collection activities, and
among diverse field workers and researchers. As demonstrated here, in the
absence of other available or acceptable data, satellite ortho-image maps
and landscape analysis based upon stereopair DEM extraction can provide
cost effective GIS data within remote, mountainous terrain, even under
conditions of extreme topographic relief. With ongoing improvement in sensor
technology and ground pixel resolution (Rao et al. 1996), DEM extraction
from stereo imagery will likely become an increasingly useful tool.
Acknowledgments
The authors would like to acknowledge the EOSAT Corp., SPOT Image Corp., and the Mountain Institute for assistance and partial funding for this project. Particular thanks go to Dr. Gabriel Campbell, Dr. Alton Byers and Robert Davis of the Mountain Institute; Dr. Phillip Cheng of PCI, Inc.; Dr. Nanda Joshi, Narayan Poudel and Ramesh Shrestha of the Makalu Barun Conservation Project; Pramod Pradhan and Basanta Shrestha of ICIMOD - MENRIS; and Dr. T.B. Shrestha of IUCN - Kathmandu. Fieldwork for this project was conducted in collaboration with Dr. Chris Carpenter and the Wildlands Studies Program, San Francisco State University, College of Extended Learning, and with the indispensable help of Laxmi Dewan and the staff at Nilgiri Trekking. Image processing and data analysis were done at the Center for Spatial Technologies and Remote Sensing (CSTARS), University of California, Davis.
References
Al-Rousan, N., Cheng, P., Petrie,
G., Toutin, Th., and Valdan Zoe, M.J. 1997. Automated DEM Extraction and
Orthoimage Generation from SPOT Level 1B Imagery. Photogrammetic Engineering
and Remote Sensing, 63 (8): 965-974.
Blamont, D. and C. Méring,
1987: Use of remote sensing for vegetation and landuse mapping in mountainous
areas: The case of central Nepal. Adv. Space Res. 7(3): 41-46.
Byers, A. 1996. Historical and Contemporary Human Disturbance in the Upper Barun Valley, Makalu-Barun National Park and Conservation Area, East Nepal. Mountain Research and Development, 16 (3): 1996.
Carpenter , C. and Zomer, R. 1996. Forest Ecology of the Makalu-Barun National Park and Conservation Area, Nepal. Mountain Research and Development. 16 (2): 1996.
ESRI, 1994. Cell-based Modeling with GRID. Environmental Systems Research Institute, Inc., Redlands, CA, 92373.
Frank, T.D. and C.E. Thorn, 1985: Stratifying alpine tundra for geomorphic studies using digitized aerial imagery. Arctic and Alpine Research, 17: 179-188.
Frank, T.D. 1988: Mapping Dominant Vegetation Communities in the Colorado Rocky Mountain Front Range with Landsat Thematic Mapper and Digital Terrain Data. Photogrammetic Engineering and Remote Sensing, 54(12): 1727-1734.
Franklin , S.E. and Blodgett, C.F. 1993. An example of satellite multisensor data fusion. Computers and Geosciences, 19 (4): 577-583.
Franklin, S.E. 1989. Ancillary data input to satellite remote sensing of complex terrain phenomena. Computers and Geosciences, 15 (5): 799-808.
Franklin, S.E. 1991: Image transformation in mountainous terrain and the relationship to surface patterns. Computers and Geosciences, 17 (8): 1137-1149.
Giles, P.T., Chapman, M.A. and Franklin, S.E. 1994. Incorportation of a digital elevation model derived from steroescopic satellite imagery in automated terrain analysis. Computers & Geosciences, 20 (4): 441-460.
Grabmaier, K., Tuladhar, A.M., and Th Verstappen, H. 1988. Stereo mapping with SPOT. ITC Journal, 2: 149-154.
Graf, K. Ch., Suter, M., Meier, E., Meuret, P. and Nüesch, D. 1994. Perspective terrain visualization - A fusion of remote sensing , GIS, and computer graphics. Computers and Graphics, 18 (6): 795-802.
ICIMOD . 1996. GIS Database of Key Indicators of Sustaninable Mountain Development in Nepal. International Center for Integrated Mountain Development (ICIMOD), Kathmandu.
Lal, J.B., Gulati, A.K. and M.S. Bist, 1991: Satellite mapping of alpine pastures in the Himalayas. Int. J Remote Sensing. 12 (3): 435-443.
Leprieur, C.E., Durand, J.M., and J.L. Peyron, 1988: Influence of topography on forest reflectance using Landsat Thematic Mapper and Digital Terrain Data, Photogrammetic Engineering and Remote Sensing, 54(4): 491-496.
MBCP Task Force. 1990 The Makalu-Barun Conservation Project Managment Plan. Dept. of National Parks and Wildlife Conservation, HMG Nepal and Woodlands Mountain Institute, Franklin W. Virgina, USA.
Millette T.L., Tuladhar A.R, Kasperson R.E, Turner B.L. 1995. The Use And Limits Of Remote Sensing For Analysing Environmental And Social Change In The Himalayan Middle Mountains Of Nepal. Global Environmental Change-Human And Policy Dimensions, 5 (4): 367-380.
PCI, 1998. Easi/Pace Software Manual, Version 6.2. PCI Inc., Ont. Canada, L4B 1M5.
Rao, T. Ch. M., Rao, K.V., Kumar, A.R., Rao, D.P., and Deekshatula, B.L., 1996. Digital terrain Model (DTM) from Indian Remote Sensing IRS) Satellite Data from the Overlap Area of Two Adjacent Paths Using Digital Photogrammetric Techiniques. Photogrammetic Engineering and Remote Sensing, 62 (6): 727-731.
Senoo, T., Kobayashi, F., Tanaka, S., and T. Sugimura.. 1990: Improvement of forest type classification by SPOT HRV with 20 m mesh DTM. Int. J. Remote Sensing 11(6): 1011-1022.
Skidmore, A.K., 1989. An expert system classifies eucalypt forest types using thematic mapper data and a digital terain model. Photogrammetic Engineering and Remote Sensing, 55 (10): 1449-1464.
Talbot, S.S. and Markon, C.J. 1986: Vegetation mapping of Nowitna National Wildlife Refuge, Alaska using Landsat MSS Digital Data. Photogrammetic Engineering and Remote Sensing, 52 (6): 791-799.
Yonzon, P., Jones, R., and J. Fox, 1991: Geographic Information Systems for Assessing Habitat and Estimating Population of Red Pandas in Langtang National Park, Nepal. Ambio. 20(7): 285-288.
Zomer, R. , Carpenter C. ,Ustin, S., and J. Ives. 1998a. Tropical Riparian Corridors of the Middle Hills Of Eastern Nepal: Structure, composition and dynamics of low elevation forest within the Makalu-Barun Conservation Area. Submitted to Plant Ecology.
Zomer, R. and Ustin, S. 1998b.
Landuse and landcover change along tropical riparian corridors within the
Makalu Barun Conservation Area. In press: Mountain Research and Development.
Average and median error (difference), and their standard deviation (std. dev.), assessing the correspondence between a DEM extracted from SPOT stereopair imagery and a DEM generated from digitized topographic contours at the 1:250k scale. Results of image subtraction are reported for (1) all pixelswithin the study area for which an elevation was estimated in the unedited DEM extraction; (2) all pixels within the study area after post-extraction editing of the extracted DEM; (3) all pixels within each of the test sites, after post-extraction editing.
| DEM |
|
|
|
|||
| MBNPCA: all initially classified pixels |
|
|
|
|||
| MBNPCA: all pixels - post-processed |
|
|
|
|||
| Test Site 1: all pixels |
|
|
|
|||
| Test Site 2: all pixels |
|
|
|
|||
| Test Site 3: all pixels |
|
|
|
|||
