Although water erosion losses are only 2.4 tons per acre per year on average in Kentucky, nearly 19% of cropland in Kentucky loses soil at rates greater than the "tolerable" levels according to 2003 statistics (NRCS, 2007). This is a concern because erosion occurs faster than replacement rates when tolerable levels are exceeded. Erosion could be substantially reduced if grassed waterways were used more frequently in areas with concentrated flow.
Elevation data were obtained with real time kinematic (RTK) global positioning system (GPS) for five fields from a farm in Central Kentucky. Terrain attributes were calculated from these datasets and were used as predictor variables for logistic regression and neural network analyses. The models were trained with observations from field assessments of soil erosion. Leave-one-field-out cross-validation analyses were used to evaluate the quality of predictions generated with the proposed methodologies.
Erosion probability maps generated with this approach are useful for identifying erosion features. Discretized maps of these probability indices also clearly indicated the locations of these features but were much easier to read. The average misclassification statistics of discretized maps (e.g., Figure 1) for non-eroded areas was 11 percent for both logistic regression and neural network analysis. The average misclassification statistic for eroded areas was 18% for logistic regression and 19% for neural network analysis. This general approach could improve the efficiency and accuracy of field site assessments for conservation planning.
Figure 1.
Discretized erosion probability map created with logistic regression overlain by the
boundaries of erosion features for Fields A through E. Dark areas are where the model
predicts a high likelihood of an eroded feature. From Pike et al., 2008 - in submission.
- Mueller, T.G. (PI), T.J. Neiman, H. Cetin, and A.D. Karathanasis. 2005-2008. Precision Land Use, Conservation, and Management: Improving Soil Survey Data with Geospatial Technologies. $62,100.
- Pike, A.C., T.G. Mueller, A. Schörgendorfer, S.A. Shearer, and A.D. Karathanasis. Erosion indicies derived from terrain attributes using Logistic Regression and Neural Networks: An Approach for Precision Conservation. Submitted to the Agronomy Journal on November 24, 2008. In Submission.
- Pike, A.C., T.G. Mueller, S.A. Shearer, T. J. Nieman, and A.D. Karathanasis. 2008. Using Terrain Attributes to identify areas suitable for grassed waterways. In R. Khosla (ed.) Proc. 8th International Conference on Precision Agriculture and other Precision Resources Management. ASA Misc. Publ., ASA, CSSA, and SSSA, Madison, WI. Published on CD.
- Mueller, T.G., H. Cetin, R.A. Fleming, C.R. Dillon, A.D. Karathanasis, and S.A. Shearer. 2005. Erosion probability maps: Calibrating precision agriculture data with soil surveys using logistic regression. Journal of Soil and Water Conservation. 60: 462-468. Abstract.
- Pike, A.C., T.G. Mueller, S.A. Shearer, A.D. Karathanasis, and T.J. Nieman. 2008. Using terrain attributes to identify areas suitable for Grassed Waterways. 2008 Soil and Water Conservation Society Annual Meetings.
