H33E-0869:
Hydrologic-Based Soil Texture Classification
Wednesday, 17 December 2014
Derek Groenendyk, University of Arizona, Tucson, AZ, United States, Ty P.A. Ferré, University of Arizona, Department of Hydrology and Water Resources, Tucson, AZ, United States, Kelly Thorp, USDA/ARS, U.S. ALARC, Maricopa, AZ, United States, Amy Katherine Rice, Colorado School of Mines, Civil and Environmental Engineering, Golden, CO, United States and Wade T Crow, Hydrol and Remote Sensing Lab, Beltsville, MD, United States
Abstract:
Historically, soil texture classifications have been created using particle sizes with the purpose of describing the agricultural function of different soils. Clustering algorithms allow for a wider range of bases for soil texture classifications that can include combinations of observations, behaviors, and soil attributes. Recent studies that have used non-mechanical based clustering methods show promising results (Bormann 2010, Twarakavi 2010). Here we present a methodology that includes procedures for creating and evaluating multiple classifications. The approach uses a k-means clustering algorithm to classify soil textures across the USDA soil texture triangle. These classifications are based on hydrologic responses to the hydrologic processes of infiltration, drainage, and evapotranspiration. Two similarity indices are introduced to quantify differences among cluster-based classifications, including the USDA soil texture classification as well as each classification. To improve our understanding of the new similarity indices, the methods were tested against commonly used indices which are based on distance (Rand, Jaccard, VI) or membership (Hubert’s, Normalized Γ) to compare clusters. Using quantitative measures to compare classifications results in a quick and scalable way to find observations that cluster in similar ways. We propose that this could be a novel approach to identifying measurements that could be characterize hydrologic processes of interest. Intriguingly, some classifications are dissimilar to the USDA soil texture classes. The results are able to capture common but unique hydrologic responses shared by different soil textures that cross multiple USDA soil texture classes. By clustering on a single variable, the resulting clusters have a physical meaning and represent soils that share specific and similar hydrologic responses and behaviors. This provides greater insight into the importance of soil texture heterogeneity and determining relevant soil texture values.
