Analysis of Land Cover Through Remote Sensing

Osage County, Kansas

Grassland east-northeast of Osage City, Kansas.
December 5, 2009
All photos by Matt Unruh


Prepared by Matt Unruh

ES 775 Advanced Image Processing
Emporia State University

Fall 2009


Table of Contents

Abstract

Project Area

Methodology

Results

Conclusions

References


Abstract

Analysis of environmental conditions through remote sensing is a common component of most land cover studies. The wide availability of quality data sources for little to no cost allows for individuals or organizations with limited resources to conduct detailed analysis on areas of interest specific to their research objectives. This project examines land cover in Osage County, Kansas, through analysis of Landsat TM data obtained at no cost from EarthExplorer. Results of preliminary analysis were groundtruthed in the vicinity of Osage City, Kansas, to create at current land cover dataset for the project area.



Project Area

Located in east central Kansas directly south of Topeka, Osage County is a rural county with an agricultural-based economy focused around ranching and farming. Lyndon, the county seat, is located in the central part of the county on U.S. Highway 75. Pomona Lake and Melvern Lake are U.S. Army Corp of Engineers reservoirs that provide outdoor recreation facilities for residents of the region as well as flood control for the Marais des Cygnes River Basin. Melvern and Pomona Lakes are located in the southwestern and east-central portions of the county, respectively. These two water bodies are the two dominant hydrologic features within the project area.

This project will examine in detail current land cover conditions in the area of the county around Osage City. This community is located approximately 15 kilometers north of Melvern Lake. Osage City Lake, located directly south of Osage City just to the east of Kansas Highway 170, will be utilized as a reference point for spatial analysis of remote sensing data developed from methodologies employed for the project area. This water body is visible on all outputs associated with the project, so recognition of this feature allows for distances to other features of interest to be determined.



Image 1: Project Area - Osage County, Kansas (highlighted).



Image 2: Project Area - Osage County, Kansas.


Return to Table of Contents



Methodology

The original intent of this project was to attempt to quantify the presence of no-till or reduced tillage agricultural lands within Osage County. These lands are characterized by reduced or no tillage taking place between crop rotations. To determine where these areas are located within the county, Lori Kukyendall, District Manager of the Osage County Conservation District, was contacted to determine location information for no-till crop lands within the county. Information was provided regarding the location of no-till fields for the primary no-till farm family in Osage County. Section/Township/Range information was provided for these fields as well as the approximate location within the section (i.e. northeast 1/4, southwest 1/4, etc.). Once the location of these sections were plotted via ArcMap, the determination was made to focus analysis and subsequent groundtruthing activities in the area of Osage County in the vicinity of these lands.



Image 3: Project Area - No-till Sections.
No-till sections identified as tan-shaded areas.


Once locations with no-till agriculture were identified, vegetation indices to evaluate these lands were sought out. C.S.T. Daughtry and others note utilization of various vegetation indices to evaluate crop residue of agricultural lands in Remote sensing of crop residue cover and soil tillage intensity (Daughtry, 2006). For purposes of this project, normalized difference tillage index (NDTI) and normalized difference senescent vegetation index (NDSVI) were evaluated. Calculations for the NDTI and NDSVI are listed as follows:

NDTI = (TM Band 5 - TM Band 7) / (TM Band 5 + TM Band 7)

NDSVI = (TM Band 5 - TM Band 3) / (TM Band 5 + TM Band 3)


With both indices utilizing Landsat TM data, a current and cloud-free data set needed to be obtained. Data from early October 2009 to early December 2009 was previewed through EarthExplorer, a U.S. Geological Survey (USGS) website in which remote sensing data can be dowloaded for no cost, in an attempt to locate a dataset in which in the project area was mostly to completely cloud free. After previewing data during this time frame it was determined that the Landsat TM dataset collected on October 7, 2009, was the best data set to utilize for analysis of the project area. This data set was completely cloud free, allowing for an unobstructed view of land cover conditions within the project area to take place. One drawback in using the data from October 7 was groundtruthing of preliminary results could not take place until early December. Ideally, data collected as close to the date of groundtruthing as possible would be utilized to evaluate results of analysis activities, but all datasets collected after October 7 showed nearly 100 percent cloud cover over the project area.


Band Color Wavelength (µm) Applications
1
Blue
0.45-0.52
Land-use, soil, and vegetations analysis
2
Green
0.52-0.60
Vegetation Reflectance
3
Red
0.63-0.69
Chlorophyll absorption
4
Near-infrared
0.76-0.90
Land/water analysis and soil/crop contrasts
5
Mid-infrared
1.55-1.75
Vegetation moisture
6
Thermal infrared
10.4-12.5
Geothermal analysis
7
Mid-infrared
2.08-2.35
Geologic rock formation discrimination
Figure 1: Landsat Thematic Mapper Spectral Bands.
Adapted from Jensen (2007, p.205).


After collecting Landsat TM data, remote sensing analysis of the project area was performed utilizing IDRISI Taiga software. The WINDOW function was first utilized to narrow down the dataset (all Landsat TM bands obtained) to roughly Osage County. The southwest corner of Clinton Lake, located in southwestern Douglas County, was selected as the northeast portion of the dataset while an area approximately 12 kilometers southwest of Melvern Lake was selected as the southwest portion. Using these areas as frames for the WINDOW function ensured that all of Osage County would be included in remote sensing analysis. After the dataset was narrowed down to the project area NDTI and NDSVI values were calculated.



Image 4: NDTI - Project Area. Image 5: NDTI - Osage City Lake Area.
Image 6: NDSVI - Project Area. Image 7: NDSVI - Osage City Lake Area.


After evaluation of both NDTI and NDSVI results, there were no noticable differences between areas identified as no-till lands in relation to adjacent agricultural lands. Because of this, tasseled cap transformation (TASSCAP) was performed on the Landsat TM dataset to further evaluate the project area for trends specific to no-till lands. TASSCAP analysis utilizes 6 bands of Landsat TM data using the Gram-Schmidt orthogonalization process to produce three new output bands (Brightness, Greenness, and Moistness). As was the case with the NDTI and NDSVI, TASSCAP analysis did not produce any noticable results that distinguished no-till agricultural lands from adjacent fields. Due to the inability to isolate no-till agricultural lands, the objective of this project changed from evaluation of no-till lands to evaluation of land cover types within the project area.



Image 8: TASSCAP - Brightness Band. Image 9: TASSCAP - Moistness Band. Image 10: TASSCAP - Greenness Band.


To examine areas with similar cell attributes, CLUSTER and ISOCLUSTER analysis were performed on the project dataset. CLUSTER analysis is an unsupervised classification of images utilizing a histogram peak technique, while ISOCLUSTER analysis is an iterative self-organizing unsupervised classification similar to H and K-means procedures. After comparing the results of the CLUSTER and ISOCLUSTER analysis for the area around Osage City, the determination was made to focus on the output from ISOCLUSTER analysis to develop land cover classifications for the entire project area. Although the CLUSTER analysis developed more definitive groups within the project area, the ISOCLUSTER analysis seemed to do a better job establishing distinct groups of one or multiple clusters that appeared to reflect actual field level variation in characteristic types. Sixteen clusters resulted from the ISOCLUSTER analysis, but when grouped together in one visual output it was difficult to determine areas where a high concentration of a particular cluster were present around Osage City. To aid in determining where cluster locations, the RECLASS function was utilized to isolate each cluster.

After viewing each of the clusters individually, a land cover classification was determined for clusters 3 and 8. Cluster 3 appeared to be associated with urban areas and the transportation network of the project area (roads and railroads), so this cluster was classified as developed. For cluster 8, the presence of Melvern, Pomona, Osage City Lake, and other water bodies within the project area made classification of this band straightforward as being water. Once all clusters were isolated, visual analysis was performed in the vicinity of Osage City (using Osage City Lake as a reference point) to determine where large continuous blocks of each remaining cluster were present. Locations where individual cluster types were observed in concentrated areas were identified and visited in person to determine the actual land cover type present at each site.



Image 11: Cluster 8 (water). Image 12: Cluster 3 (developed).


Groundtruthing of clusters in the project area took place on December 5, 2009. One particular cluster (cluster 15) was not present in a continuous tract in the Osage City vicinity. This cluster was prevalent southwest of Clinton Lake, so inspection of this site took place in transit to the Osage City area. All sites were numbered in order of visitation, and one or more pictures were taken at each site. Pictures were downloaded and labeled according to site number for future viewing. In total, 25 sites were visited during groundtruthing activities. In addition to collection of photographs depicting cover conditions, current crop and/or crop residue conditions were noted for each site. This information was later analyzed to determine land cover conditions associated with each of the sixteen clusters.



Image 13: Site 2 - forest. Image 14: Site 22 - harvested milo. Image 15: Site 13 - sunflower/wheat residue.


Figure 2: Groundtruth observation data.


After returning from groundtruthing, land cover conditions observed at each site were compared to primary clusters present in these areas. These clusters were then identified as being either cropland, forest, or grassland/pasture. Taking into account clusters 3 and 8 which were previously classified as developed and water, respectively, land cover classifications were determined for all 16 clusters obtained from the original ISOCLUSTER analysis. No definite crop types seemed to correspond with a particular cluster, so development of a crop cover dataset for the project area did not occur. After labeling these clusters with the appropriate land cover designation, a color palette was developed to help with visual interpretation of land cover classes within the project area. With multiple clusters representing similar land cover classifications visible in this output, the RECLASS function was once again performed to create a final image which highlights each of the five land cover classifications developed through analysis. An appropriate color palette was created to finalize analysis activities.



Figure 3: Groundtruth observation data with identified clusters.


Image 16: Project Area Land Cover (before RECLASS).


Return to Table of Contents



Results

The final land cover dataset developed for the project area, based on information obtained during groundtruthing, seemed to do an adequate job of representing current land cover conditions in and around Osage County. In comparing land cover classifications developed from this analysis with those of the 2001 National Land Cover Dataset (NLCD 2001), most areas seemed to reflect conditions as shown in this widely used land cover dataset. There were some areas in which actual observed conditions did not reflect analyzed land cover classifications, such as site 17 which was depicted as cluster 3 (developed) in spite of cropland evidence being present, but most of the areas identified as a particular land cover type correspond with actual ground conditions. The results of this analysis appear to do a good job of identifying water and forest land within the project area. With the project are being largely grassland/pasture and cropland, the majority of the forested lands in the project area are located within riparian areas.



Image 17: Analyzed Land Cover. Image 18: NLCD 2001.


Image 19: Final Compilation - Project Area Land Cover.

Return to Table of Contents



Conclusions

Overall, this land cover analysis utilizing Landsat TM data did an adequate job of characterizing current conditions within the project area. This particular dataset would be most useful when viewed at the county level, when small scale features are not visible. As land cover conditions are examined at the field level, potential discrepancies in represented land cover types become visible. Further processing of this dataset to correct areas that are misrepresented as well as to clean up areas where individual pixels are present would be warranted.

When compared to the NLCD 2001 dataset, changes in land uses within the project area could account for some of the areas that are represented as cropland or grassland/pasture in one land cover dataset and not the other. In spite of discrepancies being noted in the previously mentioned land cover classes as well as developed land, this analysis appears to have done well in identifying areas of forest and water. Wet weather conditions present in the region during and before this Landsat dataset was first captured could account for some additional areas being classified as water. Being that this dataset was collected October 7, forested areas could appear to cover more area in a scene captured during peak growing season.

Crop type information could potentially be obtained from this type of analysis if datasets were collected at various times throughout the year corresponding with peak greenness of a particular crop type. As an example, the extent of wheat cropland within the project area could potentially be determined by analysis of Landsat TM data obtained during the early growing season (spring). Likewise, grassland/pasture extent could be further explored by performing similar analysis techniques during late winter/early spring. This period corresponds with the controlled burning that occurs annually in this region.



Return to Table of Contents



References


Return to Table of Contents

Last Updated: December 10, 2009