Improving Plant-Available Water Estimation Using Model Averaging of National Soil Water Models Brendan P. Malone. CSIRO Agriculture and Food, Black Mountain, ACT, Australia

Brendan Malone^1*Ross Searle¹Siyuan Tian²Thomas Bishop³Yi Yu^1,2

• ¹Commonwealth Scientific and Industrial Research Organisation (CSIRO), Canberra, Australia
• ²Australian National University, Canberra, Australia
• ³University of Sydney, Sydney, Australia

Multiple operational soil water balance (SWB) models provide real-time estimates of soil moisture across Australia, but differences in model structure and output can lead to uncertainty for end users. This study investigates model averaging-specifically the Granger-Ramanathan (GRA) approach-as a strategy to improve in situ and spatial estimates of plant-available water (PAW) in southeastern and southern Australia. Unlike previous studies that apply static weighting schemes, we evaluate a temporally dynamic implementation of GRA and test its effectiveness at both point and landscape scales. Two hypotheses were addressed: (1) that GRA model averaging improves point-scale PAW predictions compared to individual models, and (2) that spatially scaling GRA coefficients yields more accurate PAW maps than equal-weight averaging. Using soil moisture sensor networks across three study regions, GRA consistently outperformed individual models and simple averaging at the probe level, achieving higher agreement with sensor observations (e.g., mean concordance of 0.87 at Boorowa, 0.73 at Muttama, and 0.90 at Eyre Peninsula, compared to 0.29-0.53 for individual models and 0.05-0.60 for equal weighting). Spatial implementations of GRA with temporally varying coefficients showed improved performance over static models, though the addition of environmental covariates did not consistently enhance mapping accuracy and in some cases reduced generalisability. These findings demonstrate the practical utility of dynamically weighted model averaging for point-based soil moisture prediction and highlight key challenges in scaling such methods spatially-particularly when sensor data are sparse or unevenly distributed. This work contributes a novel framework for integrating multiple national-scale soil water models in real-time applications, with implications for both agricultural decision-making and environmental monitoring.