Landscape Classification of Kansas Cross Timbers
ES 771 Remote Sensing

Curtis McCaslin
Nathan Holoubek

Abstract

The cross timbers region of Kansas represents a unique landscape. Predominantly post and blackjack oak, it was formerly an extensive savanna/forest mosiac. This variation in the landscape is beneficial to many bird species during breeding season, a topic being investigated by current research at ESU. Our goal was to estimate, at a large scale, the amount of various habitat types that are available on a regional scale. To do this we utilized Landsat 5 imagery, captured in August 2012. A variety of vegetation analyses were attempted before deciding to use TASSCAP. We obtained brightness measurements for the landscape and used those to map areas of dense forest, lightly forested woodland, savanna/prairie, and bare ground/urban areas. A 6,000 acre ranch of known cover type was used to calibrate estimated cover types which were applied to a large scale. The resulting maps show distribution and amount of land area in each class. This can be used at a fairly local scale (within the Cross Timbers ecosystem) to provide a rough estimate of the available habitats.


Eastern Elk, Chataqua, Western Wilson and Montgomery counties, KS and portions of northern Oklahoma. Elk City Lake and Copen Lake are labeled and the ranch that served as a ground truth point is indicated by the star. Predominant land uses are cattle ranching and row crop agriculture. Drainages can be seen clearly outlined by darker woody vegetation. In this August 2012 image the region is in a 2 year drought, with summer temperatures commonly over 100 degrees F. This caused much of the upland vegetation to go dormant and increased contrast between timber (green and active) and prairie (dry and dormant)


This image is focused on two cattle ranches in eastern Elk county. Here the difference between open dry prairie and drainages and forest is very clear. Some agricultural fields border the eastern edge of the image. Bedrock is near the surface on the ranch and outcrops are visible surrounding the drainage basins. This area is of a known cover type and was used to identify the categories created using vegetation analyses.


An NDVI image of the ranches. The few water bodies that exist in the area show up clearly as dark areas. Most of the prairie has a low NDVI value, reflecting the lack of active growth. Established trees include drought resistant post and blackjack oak in the uplands and osage orange in the wetter drainage basins. We first used this NDVI to create a classification, but found it was unable to differentiate easily between bare ground and prairie.


A reclass based on the NDVI shown above. Notice poor accuracy in classifying agricultural fields on eastern edge. Total land area in each cover type is:

Water: 17 ha
Dense canopy forest: 1140 ha
Woodland: 971 ha
Shrub: 1903 ha
Grassland: 4163 ha
Bare ground/Urban: 499 ha


TASSCAP analysis was run, with the greenness shown above. Again, trees show up with highest values and water bodies the lowest. Prairie had values near zero, indicating little greeness due to grasses being dormant from drought. Bare ground and water are red in this image.


This moisture result from TASSCAP shows drainages and forested areas as the only land cover containing much moisture. The August image was useful in telling the difference between wooded cover types and prairie, however it had trouble in all areas separating bare ground from dormant prairie grasses.


This TASSCAP brightness image is what we eventually decided to use in our cover type analyses. It differentiated prairie from bare ground the best and was able to determine several types of woody vegetative cover (see image below).


This final classification was determined by customizing groups from the TASSCAP brighness image above. Ranges of values were determined to correspond to dense canopy forest, lightly forested woodlands, open prairie and savanna, and bare ground. Some bare ground was classified as prairie, however this image separated the two best of all analyses attempted. Shrubland was not included here because it was not significant enough on the landscape to appear. Land area per cover type is:

Water: 26 ha
Dense canopy forest: 1,789 ha
Woodland: 1,759 ha
Grassland: 4,023 ha
Bare ground/Urban: 1,097 ha


The NDVI analysis at the large landscape scale. Waterbodies are clearly delineated, with Elk City Lake, KS in the northwestern part of the image and Copan Lake, OK below it. Agricultural fields can be seen to dominate the landscape in the north central part of the image, while prairie dominates the southern and western parts. Major drainages are highlighted by the active vegetation surrounding them.


The NDVI reclass image at a large scale shows water bodies very well, however it over estimates amount of forest and shrubland, as well as poorly differentiating bare ground and prairie. Total land area per cover type is:

Water: 6,691 ha
Dense canopy forest: 210,677 ha
Woodland: 134,183 ha
Shrub: 162,958 ha
Grassland: 142,120 ha
Bare ground/Urban: 70,179 ha


Final land cover map based on TASSCAP brightness composite. Water bodies are clearly delineated, forests and drainages are accurately shown, and bare ground/urban areas are clearly seen. This classification was the most accurate in separating prairie from bare ground. Total hectares in each class are:

Water: 9,276 ha
Dense canopy forest: 105,062 ha
Woodland: 68,728 ha
Shrub: 94,216 ha
Grassland: 350,729 ha
Bare ground/Urban: 98,797 ha

Sources of information


United States Geological Survey, Global Visualization Viewer, World Wide Web homepage http://glovis.usgs.gov. Accessed November 7, 2012.

Aber, JS. 2010. ES 771 Lab 4: Vegetation Analysis. World Wide Web homepage http://academic.emporia.edu/aberjame/remote/syllabus.htm. Accessed September 19, 2012.