Di He 1, Enli Wang 1
Plant available water holding capacity (PAWC) interacts with climate to determine crop yield and is thus a key factor for predicting spatial yield variations. However, PAWC data at the required spatial resolution are not available because direct soil measurement is expensive and time-consuming. Here, we explore a new approach to inversely estimate PAWC from crop LAI time series using process-based modelling with APSIM together with machine learning. We used the APSIM model to simulate daily LAI of wheat in response to a wide range of PAWC across Australia. Vegetation metrics are derived from simulated LAI time series and were used together with climatic variables to build a machine learning model for predicting PAWC. The model explained 29% to 83% variation of PAWC across ten sites with contrasting climate. This implies a potentially more effective way of PAWC estimation, an alternative to direct soil sampling method.