A0528
Title: A changepoint approach to modelling soil moisture dynamics
Authors: Mengyi Gong - Lancaster University (United Kingdom) [presenting]
Rebecca Killick - Lancaster University (United Kingdom)
Christopher Nemeth - Lancaster University (United Kingdom)
Abstract: Soil moisture is an important measure of soil health that scientists model via soil drydown curves. The typical modelling process requires manually identifying the drying process and fitting exponential decay models to them. This can be time-consuming and the result is a static overview of the drydown property. Motivated by the spike-train problem in neuroscience, a novel changepoint-based approach is proposed to automatically identify structural changes in the soil drying process. Changes caused by sudden rises in soil moisture content over a long time series are captured and the parameters characterising the drying processes are estimated simultaneously. Segment-specific parameters are used to capture potential temporal variations in the drying process. An algorithm based on the penalised exact linear time (PELT) method was developed to identify the changepoints. Applying the algorithm to simulated and real data show the good performance of the method. To improve flexibility, the method is extended such that different types of models can be used to describe different segments. This allows the segmentation of the time periods when no drydown happens due to low temperature or saturation, in addition to the typical drying periods. An approach based on the Bayesian changepoint detection method and the particle Metropolis-Hastings is being investigated.
