Identification of irrigation events using Bayesian statistics-based change detection and soil moisture measurements

A comprehensive knowledge of irrigation information is crucial for agricultural water management. However, current investigations have mainly focused on extracting spatial extent of irrigated farmlands and quantifying irrigation amounts, lacking an understanding of irrigation timing at the field scale. In this study, a novel approach for detecting irrigation events from soil moisture (SM) time-series was proposed. To this end, in-situ SM measurements with different depths (10 cm, 25 cm, and 50 cm) were primarily decomposed into seasonal, trend, and residual components using the Bayesian Estimator of Abrupt change, Seasonal change, and Trend (BEAST) model over a period of seven years from 2014 to 2020. The rationale for the determination of a specific irrigation timing relies on the observed rising abrupt change of SM time-series in its trend component when precipitation is unavailable. Specifically, the BEAST model was primarily optimized over two irrigated farmlands in the University of Nebraska Agricultural Research and Development Center near Mead, Nebraska, US. were subsequently used to identify irrigation. Results indicate that the decomposed SM time-series by the BEAST model correlate well with in-situ SM measurements with an average coefficient of determination of 0.98 and 0.97 over farmlands with continuous maize and maize-soybean rotation, respectively. Furthermore, it was found that SM measurements with a depth of 10 cm are optimal for detecting irrigation timing over the study area. When compared with local irrigation records, the accuracy of detected irrigation timing over farmlands with continuous maize and maize soybean rotation can reach 84 % and 89 %, respectively, revealing promising prospects for deriving irrigation timing with SM measurements. These results provide a reference for detecting irrigation timing using satellite-derived SM data.