The Sydney Soil Water-Energy Balance (SWEB) model generally tracked observed soil-moisture patterns, with correlation coefficients — a measure of how closely two sets of values move together — of 0.70 to 0.85 across most regions. The national distribution peaked around 0.75, although its lowest values fell below 0.55.
The model's errors were lower in western regions than in southeast Australia. Root-mean-square error (RMSE), a measure of the gap between an estimate and an observation, was 0.05 to 0.07 in western regions and 0.09 to 0.11 in southeast Australia. Most national RMSE values were concentrated between 0.05 and 0.08.
How SWEB makes its estimates
SWEB couples a surface energy-balance component for estimating actual evapotranspiration, or crop water use, with a soil-water-balance component that represents the soil as vertical layers and tracks water entering, moving through and being stored in the profile.
The surface component combines Landsat 8/9 land-surface temperature with SILO meteorological inputs. The thermal observations are at 100 metres and are resampled to 30 metres, so the finer output grid does not necessarily represent independent 30-metre thermal observations.
A national check against monitoring data
Researchers calibrated SWEB using differential evolution to estimate its parameters, minimising the root-mean-square error between observed and simulated surface soil moisture at 50 mm depth over a calibration period. The optimisation used a population size of 10 and up to 30 iterations, and required at least 30 valid observation points for reliable parameter estimation.
Validation compared the model with independent observations from soil-moisture monitoring networks spanning key Australian grain-growing regions and climate zones. The networks covered varied soil types, depths, spatial supports and management contexts, while only stations meeting predefined data-completeness and quality criteria were included.
The paper does not report how many stations, observations or sites contributed to the evaluation, leaving the size of the validation evidence base unclear.
A proposed farm use, not a tested outcome
SWEB is intended to produce daily estimates of actual evapotranspiration and root-zone soil moisture. The paper proposes combining near-real-time plant-available-water nowcasts — estimates of the soil-water supply available to crops — with a water-use-efficiency equation to estimate water-limited yield potential and inform nitrogen-fertiliser top-up recommendations.
Those applications are proposals, not reported tests of yield forecasts or management outcomes. The evidence presented here concerns model performance against soil-moisture observations, not whether the proposed recommendations improve farm results.
There is also a calibration caveat: the model's parameters were fitted to surface soil-moisture observations at 50 mm, while the model is intended to estimate root-zone soil moisture. The supplied analysis says the transfer from the surface calibration signal to root-zone estimates is not fully quantified.
The authors' conclusion
The authors conclude that validation against multiple independent soil-moisture networks supports SWEB's national applicability across diverse climates, soils and management systems. The research was supported by growers through trial cooperation and by the Grains Research and Development Corporation.
The work is identified as a preprint, arXiv version 1 dated 20 Aug 2026.
Paper data and sources
Original title: SoilWaterNow: Soil water nowcasting for mapping plant available water (PAW) across paddocks for improved on-farm decision-making
Authors: Yi Yu, Mikaela J. Tilse, Patrick Filippi, Thomas F. A. Bishop
Journal/Repository: GRDC Grains Research Update 2026, Goondiwindi, Australia, 3-4 March, 149-157
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text