Spatio-temporal patterns of soil water storage under dryland agriculture at the watershed scale

Abstract

Summary Spatio-temporal patterns of soil water are major determinants of crop yield potential in dryland agriculture and can serve as the basis for delineating precision management zones. Soil water patterns can vary significantly due to differences in seasonal precipitation, soil properties and topographic features. In this study we used empirical orthogonal function (EOF) analysis to characterize the spatial variability of soil water at the Washington State University Cook Agronomy Farm (CAF) near Pullman, WA. During the period 1999–2006, the CAF was divided into three roughly equal blocks (A, B, and C), and soil water at 0.3 m intervals to a depth of 1.5 m measured gravimetrically at approximately one third of the 369 geo-referenced points on the 37-ha watershed. These data were combined with terrain attributes, soil bulk density and apparent soil conductivity (ECa). The first EOF generated from the three blocks explained 73–76% of the soil water variability. Field patterns of soil water based on EOF interpolation varied between wet and dry conditions during spring and fall seasons. Under wet conditions, elevation and wetness index were the dominant factors regulating the spatial patterns of soil water. As soil dries out during summer and fall, soil properties (ECa and bulk density) become more important in explaining the spatial patterns of soil water. The EOFs generated from block B, which represents average topographic and soil properties, provided better estimates of soil water over the entire watershed with larger Nash–Sutcliffe Coefficient of Efficiency (NSCE) values, especially when the first two EOFs were retained. Including more than the first two EOFs did not significantly increase the NSCE of soil water estimate. The EOF interpolation method to estimate soil water variability worked slightly better during spring than during fall, with average NSCE values of 0.23 and 0.20, respectively. The predictable patterns of stored soil water in the spring could serve as the basis for delineating precision management zones as yield potential is largely driven by water availability. The EOF-based method has the advantage of estimating the soil water variability based on soil water data from several measurement times, whereas in regression methods only soil water measurement at a single time are used. The EOF-based method can also be used to estimate soil water at any time other than measurement times, assuming the average soil water of the watershed is known at that time.