Land classification of Mt. Everest National Park to identify suitable Snow Leopard habitat
FR 5262/3262 Semester Project
Fall 2012
Sujhav Pun (FR 5262)
Martin Gordon (FR 3262)
Introduction:
The endangered snow leopard (Uncia Uncia) disappeared from the Everest region in the 1960s. By the late 1980’s anecdotal reports noted the presence of the transient snow leopard visiting from adjacent Tibet. Since then by understanding that there is an opportunity for conservation, numerous surveys have been done in this region to document their presence and to understand their relationship with their mountain ecosystem. Due to the location and nature of the area, field work has proven to be extremely difficult which has limited our capacity to fully understand the snow leopard’s biology and ecology. Harsh climate and daunting terrain has prevented researchers from getting a good estimate of their population size. Therefore until the present day, knowledge about these leopards have been much lower than the quantity and quality of knowledge that have been gathered about other big cat species which occupy a less challenging terrain. Other aspects being equally important, the availability of strong baseline information about the habitat of these species, in other words, a good spatial knowledge about this mountainous terrain is needed that will provide a firm stepping stone for further studies. Therefore, we understand that there is a tremendous opportunity to use Remote Sensing data to understand the terrain complexity of the Mt. Everest region and derive information that will be suitable for understanding these mountains in regards to quality and quantity of habitable area that is available for this population of snow leopards. The likely application of RS technology in this context will be to first classify the RS data/image of this area to all possible classes and sub-classes of land cover and land use types. This will assist us in determining the availability of habitable land, qualify them and also assess it quality by comparing them against known factors that affects the dispersal/movement of the species. We plan to carry out an unsupervised classification method to ensure that we account for all possible feature types, followed by a supervised classification to verify our classification and present accurate to users seeking information.
Study Area:
Elevation ranges from 2,810 m to 8,848 m at the summit of Mt. Everest. Based on surveys, snow leopard signs have been found between the elevations of 3,500 m to 5,500 m suggesting their altitudinal home range (Aryal, Raubenheimer, Web). Their average density estimates to 1-2 snow leopards per 100 km2 in Bhutan to 4-5 per 100 in 100 km2 in Nepal. Tourism in the Everest area has flourished since the ascent of Mt. Everest in 1953 increasing the visitor/tourist population which now exceeds the local population by six folds. As a succession to this ‘tourist boom’, the Everest region became Mt. Everest National Park in the year 1976 (Lovari,,Boesi, Minde , 2009). Since then, snow leopards seem to have made a comeback to the world’s highest national park.
Figure 1 – Main Valley Map
Preferred Leopard Habitat:
Snow Leopards have been greatly difficult to study due to their low numbers, low densities and large geographical ranges all of which have made them a very elusive species. However, through the use of sign mark data collected by researchers in recent years a better picture has been painted of what habitat preferences they tend to prefer in the Mt. Everest National Park. A study done by Wolf and Ale between 2004-2006 indicated a preference by the cats within this park for areas of low elevation, close proximity to rivers, less steep areas, areas away from high human traffic and areas with less snowfall (Wolf and Ale).
Methods
In order to classify the Mt. Everest region, a suitable and most recent Landsat-5 image was acquired for the region that had 30 m spatial resolution with 7 radiometric spectral bands. The software ERDAS Imagine 11 was used to analyze the satellite data that was obtained from the USGS website (http://landsatlook.usgs.gov/). The image we downloaded had less than 20 percent cloud cover. Landsat 5 carries a thematic mapper with seven spectral bands that provides finer spatial resolution, improved geometric accuracy, greater radiometric detail and more detailed spectral information in more precisely defined spectral regions than its previous versions.
We used two thematic data layers downloaded from Mountain GeoPortal website (http://geoportal.icimod.org/Downloads/Download.aspx?ID=2). The first was a shape file of a country wide national park boundaries including Mt. Everest National Park and the second was a shape file constituting ecological data of the entire country of Nepal. These two thematic layers were then re-projected to WSGS 85 UTM-45N to make it align accurately with the Landsat image of the park district. These thematic layers were overlaid onto the Landsat 5 image and clipped using AcrMap 10 to show the union of those layers only within the park boundaries. The image was brought into ERDAS Imagine where the Landsat data and ecological thematic data sets were merged into one single image. The park boundary layer was used to clip the Landsat image so that data processing was only done within the park boundaries. Following the processing of these raw data, we performed an unsupervised classification and two supervised classifications – the first showing all possible land cover types within the park boundary and the second showing different vegetation types that are available in the park.
Data Source:
http://landsatlook.usgs.gov/
www.landsat.gsfc.nasa.gov
www.earthnow.usgs.gov
Nepal datasets: http://geoportal.icimod.org/Downloads/Download.aspx?ID=2
Main valley Map: http://vacationtrekking.com/images/map/big/everest_base_camp.gif
Results
Unsupervised Classification –
We began running this classification with 20 maximum classes and 15 iterations to have an idea about different natural classes present in the image. Comparing the results with the Landsat image, we realized that 20 classes were much more than what we needed since the result didn’t match with what we observed in the image. Thus, we repeated the process while narrowing down the maximum classes assigned. We felt an optimum result was obtained when we narrowed it down to six classes since the result watched well with our interpretation of the image. Since the classification was unable to distinguish between vegetative cover types the classified image was overlaid with ecological data shape files. Orange regions of the classified image signify vegetative cover while the black lines delineate the boundaries of ecological zones within the Mt. Everest region. Together these two function to serve as indicators to give estimations of the dominant cover type within the outlined polygons. A review of literature describing Snow Leopard’s prey species preferred cover types indicates where the cats may be encountered during hunting endeavors. The Snow Leopard also does denning in rock crevices. While the identification of rock caves was not possible using the data we analyzed, we were able to identify barren rock and/or soil cover where leopards could plausibly be found to be denning colored in red-brown.
Figure 2: Unsupervised Classification of Mt. Everest National Park
Supervised Classifications:
The first supervised classification was performed to classify the park by its different cover features. This classification was done based on our interpretation of the image. We set different band combinations in the image in an attempt to make the image visible to us in all possible scenarios so we could separate different types of features. In particular, false color image of R/G/B – 4/3/2 was of importance. Based on the knowledge derived from the band combinations, we specified minimum of six training sites as an area of interest for each class. The pixels within training sites were used to generate signatures. After running the classification, we got a desired output image with 8 different classes which were consistent with our manual interpretation of the Landsat image. We ran the classifier with maximum likelihood parametric rule after merging the signatures.
Figure 3- Supervised Classification of Mt. Everest National Park showing land features
In figure,
Green - Vegetation
Cyan – Icy surface
Brown – Bare rock surface
Yellow – Glacier lake
Red – Lakes
Magenta – Rivers
White – Snow covered area
Blue – Glacier moraines
The second supervised classification was performed to classify the park by different vegetation types. Using the ecological map, we were able to get information of what possible cover types are present in the park and by overlaying the ecological shape onto the park image we specified training sites for the classifier. This classification was done based on information the in ecological data. Similarly, we specified minimum of six training sites for each forest type. After running the classification, we got a desired output image with 6 different classes of forest types. Maximum likelihood was chosen as the parametric rule.
Figure 4 – Supervised classification image showing different cover types
In figure,
White – Nival zone
Yellow – Juniper scrub
Magenta – Sub-alpine Juniper forest
Purple – Upper-temperate Blue Pine forest
Green – Fir-birch Rhododendron forest
Tan – Sub-alpine Pasture
DEM Data:
We acquired a STRM Digital elevation data for the region ranging from 2500 m to 7500 m. This data was initially manipulated to get a 200 m increment elevation map to see how elevation is distributed throughout the whole district. Literatures suggest that snow leopard signs and marking have been found within the elevation of 3500 m to 5500m. We merged the DEM map so that we got one single elevation class of 3500 m to 5500 m and compared that with our image. The comparison suggested that most of the park area (more that 75 percent) fell under the elevation range suggesting that in terms of altitude home range; most of the park area is suitable for snow leopards habitat. This range mostly coincided with vegetation cover and bare rock surfaces.
Accuracy Assessment:
Classification is followed by its accuracy assessment. We made an attempt to run an accuracy assessment report and obtain an error matric for our supervised classifications. However, due to technical problems we face during the process and the lack of time to go around it, we were unable to perform an accuracy assessment. We were able to get to a point where we could create random points from the accuracy assessment dialogue box in our classified image after linking the 2D view containing the classified scene. However, when trying to open our reference image (Landsat image) from the accuracy window, we got an error message saying “The image doesn’t contain thematic layer”. This kept us from proceeding further with our error analysis.
Conclusion:
Using Landsat image, we were able to classify the image by cover types and vegetation types. We were also able to learn the skills of doing unsupervised classification and how you can repeat the classification to get an ideal image given you have knowledge of the area; in this case it was image interpretation by employing different band combinations. We realized how important it is that all the data needs to be properly projected so that they align with each other. In this project, we had a set of ‘raw data’ - image file, shape files and DEM file; using these data we were able to meet our project goals. Several tasks had to be done in ArcMaps and bring the outputs to ERDAS Imagine where we could do further analysis/operations. If we had an opportunity to do further tasks in the project, we would do an accuracy assessment report and do a classification involving elevation data, vegetation types and land features and incorporate all three in one image that will be informative to researchers and conservation planners. Another possibility is to study climate change in the area using satellite images since it has been an area of growing concern in the mountainous areas worldwide.
Literature Cited:
Aryal, Achyut, David Raubenheimer, Suman Subedi, and Bijaya Kattel. "Spatial Habitat Overlap and Habitat Preference of Himalayan Musk Deer (Moschus Chrysogaster) in Sagarmatha (Mt. Everest) National Park, Nepal." Current Research Journal of Biological Sciences, n.d. Web.
Lovari, S., Boesi, R., Minder, I., Mucci, N., Randi, E., Dematteis, A., and Ale, S. B. (2009).
Restoring a keystone predator may endanger a prey species in a human-altered ecosystem:
the return of the snow leopard to Sagarmatha National Park. Animal Conservation 12: 559-570.
Wolf, Mosheh, and Som Ale. "SIGNS AT THE TOP: HABITAT FEATURES INFLUENCING
SNOW LEOPARD UNCIA UNCIA ACTIVITY IN SAGARMATHA NATIONAL PARK,
NEPAL."Journal of fMammalogy,n.d.Web.<http://www.asmjournals.org/doi/pdf/10.1644/08-M
AMM-A-002R1.1>.