Applicability of temporal stability analysis in predicting field mean of soil moisture in multiple soil depths and different seasons in an irrigated vineyard

摘要

Accurately estimating field mean of soil water content (SWC) is important for irrigation management in an agricultural field. Temporal stability (TS) analysis has been widely applied to identify representative locations (RLs) for predicting spatial mean of SWC in natural landscapes. The objectives of the study were to examine TS patterns of SWC at various soil depths in different years, and to evaluate the performance of TS analysis in predicting field mean in an irrigated vineyard. Soil water content was measured regularly at 135 locations in the vineyard at soil depths of 0-20, 20-40, and 40-60 cm in 2012-2014. Temporal stability indices of mean relative difference (MRD), standard deviation of relative difference (SDRD), the comprehensive indicator (CI), and the mean absolute bias error (MABE) were obtained for each location at different depths in different seasons. The range of MRD and standard deviation of MRD (SDMRD) increased with soil depth for the three years. Interseason correlations of the indices at the same depth were stronger than those between different depths of the same year. The RLs identified by CI and MABE were different between different soil depths and between the three seasons. Temporal stability analysis outperformed the random sampling method greatly, and on average one single RL could achieve RMSE about 2% and the maximum absolute error (MAE) less than 4% at a given sampling day during the validation period when the same year data were used to identify the RL. Estimation error could be reduced by 40% with the number of RLs increased from one to four. For most cases, four RLs identified by 2012 data were good enough to obtain RMSE and MAE less than 3% for the following two years. Soil texture information was found unable to identify RLs that could reliably estimate field mean. Temporal stability analysis is a valuable tool in identifying RLs to estimate field mean in an irrigated agricultural field, while intensive SWC sampling during half an irrigation cycle is needed to identify the RLs.