DYNAMIC NEAREST-NEIGHBOR METHOD FOR ESTIMATING SOIL WATER PARAMETERS

Abstract

For simulation of soil water dynamics, the lower limit (LL) of plant water availability and the drained upper limit(DUL) are required but are generally not available. Methods that estimate these properties using soil survey data are in demand.The objective of this study was to develop a self-calibrating, non-parametric method to estimate DUL and LL. Usingthe K-nearest-neighbor (NN) method and a large dataset of field-measured soil-water retention parameters, textures, organiccarbon, and bulk density, we developed and demonstrated its use to estimate DUL and LL. R2 was 79% and 73% for DULand LL, respectively. Estimated root mean squared error (RMSE) values were the lowest for potential extractable soil water(PESW) (0.024 cm3 cm-3) and about the same for DUL (0.035 cm3 cm-3) and LL (0.037 cm3 cm-3). Overall, LL and DULvalues of coarser soils (i.e., sand) were more accurately estimated than for fine-textured (i.e., clay loam) soils using acombination of clay, sand, and organic carbon. The NN method does not require prior assumptions as to the relationshipsamong soil water parameters and texture variables, and it is valid for a wide range of textures with no apparent limitationslike many existing methods. Based on this study, the NN method provides an efficient tool for estimating soil water retentionparameters with a high degree of accuracy through a database of clay, sand, and organic matter.