The Quantification of Farm Ponds in Jefferson County, Kansas
By Justin Abel
ES 775: Advanced Image Processing
Most water bodies that are found in Kansas are anthropogenic features. Farm ponds are an example of such man-made water features, which have an immense impact on the biological, chemical, and physical environment surrounding them. It has been estimated that there are between 2.6 and 9 million ponds located within the United States (Buddemeier, 2006). The substantial range in this estimation may be the result of the automated identification of the small impoundments with multispectral imagery as was conducted in this investigation.
The purpose of this project is to compare the efficiency and accuracy of several imagery datasets of differing resolutions in identifying small impoundments, or ponds. Multispectral Landsat imagery with 30 meter resolution was utilized to identify farm ponds in Jefferson County, Kansas and was then re-analyzed after pan-sharpening the dataset to increase the resolution to 15 meters. Both image datasets were classified as a boolean image to quantify the number of ponds. The result was compared to the actual quantity of ponds derived from heads-up digitization utilizing 1 meter resolution NAIP imagery. In this particular application, heads-up digitization may be defined as the identification of farm ponds through remote observation of high resolution imagery and the manual placement of a data point at the center of the pond.
Jefferson County, Kansas was the main study area of this investigation. Jefferson County is located in northeastern Kansas and within the glaciated physiographic region, where the southern boundary is defined by the Kansas River. In 2015, this approximate area recieved 49.39 inches of rainfall, 2.76 inches of which fell during the months of January, February, and March (National Weather Service). This year, 2015, was selected due to the amount of precipitation as to obtain imagery datasets where the water bodies were nearly full. This factor is important as to make the identification of the ponds easier, since they become larger in size with increased water storage. In this application a farm pond may be defined as a small impoundment developed by constructing an embankment across a watercourse for the purpose of storing water for livestock, irrigation, wildlife habitat, recreation, and erosion control (Deal et al., 1997).
The full extent of Jefferson County is shown on the left side as a false-color composite representing BGR bands 5, 6, and 4. The red box located near the center of the image indicates the location of the inset image on the right side. Notice the pixelation of the image, especially around the water bodies represented in black. These images are shown in 15 meter resolution as a result of pansharpening the orignial image dataset.
Jefferson County is shown on the left side in full extent as a false-color composite representing BGR bands 5, 6, and 4. The red box located near the center of the image indicates the location of the inset image on the right side. Notice the pixelation of the image, especially around the water bodies represented in black. These images are shown in 30 meter resolution, representing the original image dataset.
Natural color imagery for the Jefferson County area was acquired from the Kansas Data Access and Support Center (DASC) website (https://www.kansasgis.org/). The National Agriculture Imagery Program (NAIP) dataset was utilized from the Farm Service Agency (FSA), which was acquired in 2015. The NAIP imagery was imported into ArcMap 10.4, along with the Jefferson County boundary, and each pond in the county was visually identified. The imagery was traversed in an east/west orientation at a scale of 1:4,000. With the identification of a pond, a point was then placed in the center of the water body. Only ponds that were recognized as anthropogenic features were included as well as those that are used for irrigation, livestock, and recreation. Small impoundments recognized as lagoons, which are traditionally square-shaped and are surrounded by fencing, or waste storage sites were not included. Vegetation was commonly found on the pond surface and was characterized by a lighter, almost yellow-green color in comparison to the surrounding deciduous, grassland, and agricultural vegetation. Other ponds were found as nearly black bodies while others were similar in color to the soil due to the suspended sediment or turbidity.
Landsat imagery for the Jefferson County area was acquired from the United States Geological Survey (USGS) Global Visualization Viewer website (http://glovis.usgs.gov/). Specifically, a single scene from the Landsat 8 Operational Land Imager was utilized from March 30, 2016, as it appeared to temporally match the date at which the NAIP imagery was aqcuired. This temporal comparison was determined through identification of the agricultural landscape in the NAIP scene and knowledge on local farming practices.The Landsat imagery was extracted from a compressed folder using 7-Zip software. The resulting images were imported in Idrisi Selva from the menu, "File", "Import", "Government/Data Provider Formats", and "Geotiff/tif". The panchromatic image, band 8, did not spatially match the standard bands at the minimum and maximum X/Y coordinates and was therefore projected. This was accomplished at "Reformat" and "Project". The information in the output reference information was deleted and the parameters from one of the standard bands was selected, "copy from existing file". Afterwards, the number of rows and columns was doubled in order to maintain the 15 meter resolution of the panchromatic band. The standard images were then enchanced by selecting "Image Processing", "Enhancement", and "Pansharpen". The local regression transformation was utilized as the transformation type and the input images were the standard images, bands 1 through 7. The panchromatic image used for enhancement was specified as band 8. Since the raw images make up an area much larger than the target scene, a window was created to encompass the Jefferson County area, "Reformat" and then "Window". The rows and columns were the basis for specifying the window extent. The images were then corrected for atmospheric haze using "GIS Analysis", "Mathematical Operators", and "Scalar", so that the minimum values were equal to zero. A false-color composite was generated using the haze-corrected images from bands 5, 6, and 4 by following "Display" and "Composite". A cluster analysis was conducted, "Image Processing", "Hard Classifiers", and "Cluster", and the resulting image was then reclassified, "GIS Analysis", "Database Query", and "Reclass", so that only water was represented.
The .RST file from Idrisi Selva was added directly as a layer into ArcMap 10.4 where the raster dataset was converted into a polygon shapefile. Specific attributes were selected based on the GRIDCODE so that only polygons which represented water were selected. The selected data was then exported as a new shapefile and clipped to the Jefferson County boundary. The polygon shapefile was then edited to remove any features representing the Deleware River or sections of Perry Lake. The remaining polygons were then converted into points. The number of points within the existing shapefile could be found in the attribute table, thus providing a total count of ponds for that dataset.
The result of the 15 meter pansharpening of the Landsat 8 OLI image is shown in comparison to the standard 30 meter image. These images are false-color composites representing BGR bands 5, 6, and 4. The black pixels represent ponds/water bodies while dark green is deciduous forest, light-green/yellow is row-crop agriculture, blue is pasture/grassland, and magenta is urban developement. The left half of the image presents the 30 meter resolution image while the right side shows the pansharpened image at 15 meter resolution. The pansharpened image appears to define the edge of the small impoundments more distinctly than the standard image while also including additional pixel values that are more definitive in smaller bodies.
Results and Discussion:
In the process of analyzing and enhancing the Landsat imagery, a BGR bands 5, 6, and 4 false-color composite was generated in the standard resolution of 30 meters and another in the pansharpened resolution of 15 meters. A cluster analysis of each false-color composite resulted in a booleaen image of each depicting water bodies within Jefferson County. These boolean images then lead to the creation of point shapefiles that provided a numerical representation of pond quantity.
The numerical results are summarized in Table 1 based on the type of imagery, the method at which it was analyzed, the bands utilized, and number of ponds identified. The pansharpened image resulted in the identication of 4,691 ponds, while the 30 meter imagery resulted in 3,695 ponds and the actual is 4,176, as indicated by heads-up digitizing. In reviewing the pond identification points derived from the pansharpened image, it appears that some of the ponds were double counted in some instances. It may also be noted that, due to the increase in spatial resolution, some tributary streams of Perry Lake were falsely classified as ponds. Though, through furthur inspection, it may be recognized that some ponds were not identitified at all. The pond identifaction points derived from the standard 30 meter resolution image indicate suseptibility to false identification of ponds as well along some streams, though not to as a severe extent as the pansharpened image. The number of missed ponds appears much higher than the pansharpened image, most likely due to the coarseness of the resolution. In both imagery datasets, water bodies that do not qualify under the parameters that characterize a pond were classified as ponds, such as city and rural lagoons and waste storage facilities. These observations may be derived from comparison with the point dataset generated from heads-up digitizing, or manual identification.
Table 1. Imagery Analysis
In this application, the automated method of quantifying the number of ponds within an area does not seem effective in regards to accuracy and efficiency. This method still requires manual checking and editing of the dataset to ensure that inaccuracies are not included, such as the false identification of a pond and the omissioin of a pond all together. Although it does not appear efficient, when considering accuracy, heads-up digitizing is the most effective method for identifying ponds or other small water bodies.
||Number of Ponds|
|15 meter (pansharpened)
||Landsat 8 OLI
||5, 6, 4
|30 meter (standard)
||Landsat 8 OLI
||5, 6, 4
The ponds classified under the pansharpened satellite image are presented as a point dataset. The entirety of Jefferson County is shown on the left side and a red rectangle indicates the location of the zoomed-in view on the right side. It may be observed in the close-up view that several of the ponds are double-counted while others are not accounted for at all.
The ponds classified under the NAIP image, heads-up digitization, are presented as a point dataset. The points were generated through a visual investigation and manual placement. It may be observed that the points were placed at the near-center of the ponds and each was successfully identified. The entirety of Jefferson County is shown on the left side and a red rectangle indicates the location of the zoomed-in view on the right side.
The ponds classified under the 30 meter satellite image are presented as a point dataset. The entirety of Jefferson County is shown on the left side and a red rectangle indicates the location of the zoomed-in view on the right side. It may be observed that several of the ponds were completely omitted from the dataset. Also, along the southeast corner of the red box, two datapoints that mark the location of a pond actaully represent a stream.
Buddemeier, Robert W., deNoyelles F.J., Egbert S., Sleezer, R.O., Young, D.P., Zhan, X.Y., Andereck, Z., Houts, M., Mosiman, B., Taylor, P., Vopata, J., Wilson, Renwick, W.H., and Smith, S.V., 2006. Detection and characterization of small water bodies. Kansas Geological Survey Open File Report 2006-9. http://www.kgs.ku.edu/Hydro/Ponds/NASA_Ponds_FinalTechRept.pdf (accessed April 2017).
Deal, Clifton, Edwards, Jerry, Pellman, Neil, Tuttle, Ronald W., and Woodward, Donald, 1997. Ponds-Planning, Design, Construction. Agricultural Handbook 590. https://www.nrcs.usda.gov/Internet/FSE_DOCUMENTS/nrcs144p2_030362.pdf (accessed April 2017).
National Weather Service. NOWData-NOAA Online Weather Data, Topeka, Kansas Area: Monthly summarized data. http://w2.weather.gov/climate/xmacis.php?wfo=top (accessed April, 2017).
Back To Top