2026/01/01 by Andries Potgieter, Jason Brider, Davoud Ashourloo +8 · 1 voice
Agricultural and Biological Sciences · Environmental Science · #Climate change impacts on agriculture #Light effects on plants #Remote Sensing in Agriculture
paper · doi:10.1093/insilicoplants/diag018
openalex publication_date 2026/01/01 · openalex created_date 2026/07/05 · openalex updated_date 2026/07/30
Abstract Australian dryland grain production systems are under increasing pressure from climate variability, extremes, and market uncertainty, driving the need for earlier and more spatially explicit estimates of crop condition, grain yield, and production. This paper presents an overview of CropVision, a crop forecasting system developed by The University of Queensland, Australia. It is presented here as an integrated framework for monitoring and predicting crop performance across regional, farm, and field scales. CropVision is an integrated framework that links seasonal climate forecasts, biophysical crop modelling, and Earth observation (EO) to predict crop performance from region to field for wheat and sorghum in Australia. In this study, crop performance refers to spatial and temporal variation in yield, crop area, crop stress, and hyperspectral-derived proxies for solar-induced fluorescence (SIF) used to characterize crop condition and production across scales. CropVision is designed to improve both temporal lead time and spatial resolution. Seasonal forecast ensembles were calibrated and downscaled to drive crop simulations, providing useful production outlooks up to 4 months before sowing, although forecast lead time and accuracy varied across regions and between crop types. High-resolution EO data at peak canopy were used to scale crop estimates from field to farm to regional levels. Crop indices were derived from Sentinel-2 multispectral data linked to canopy structure, chlorophyll, and crop stress. As a preliminary analysis, we evaluated field-scale wheat yield estimation using SIF proxies derived from EnMAP hyperspectral imagery across 25 fields over two contrasting winter seasons. At field scale, the AI–EO–biophysical wheat yield model achieved robust performance across 375 fields over five winter seasons, with R2 = 0.76 and RMSE = 0.46 t/ha. At farm scale, application of the modelling approach in 2024 produced a predicted mean wheat yield of 3.40 t/ha compared with an observed mean of 2.91 t/ha, equivalent to an absolute error of 0.49 t/ha and a +16% deviation. At regional scale, wheat production estimates generated by the approach for Moree Plains district in the 2020 census year differed by 12.5% from Australian Bureau of Statistics estimates, illustrating the robustness of the framework in integrating crop area and yield to derive production. In the hyperspectral analysis, the best-performing fluorescence-sensitive proxy explained up to 87% of observed field-scale wheat yield variability across contrasting Western Australian environments. Together, these results show that integrating seasonal climate foresight, crop biophysics, and EO can improve crop performance forecasting in Australian wheat and sorghum systems by linking pre-sowing regional outlooks with within-season field-scale estimates.