Evaluation of water status of wheat genotypes to aid prediction of yield on sodic soils using UAV-thermal imaging and machine learning

Water stress limits wheat growth and the yield on rain-fed sodic soils. Appropriate selection of traits and novel methods are required to forecast yield and to identify water stress tolerant wheat genotypes on sodic soils. In this study, we proposed a thermal remote sensing and machine learning-based approach to help predict the biomass and grain yields of wheat genotypes grown with variable water stress in sodic soil environments. We employed unmanned aerial vehicle-based thermal imaging to quantify water stress of 18 contrasting wheat genotypes grown on moderately sodic (MS) and highly sodic (HS) soils in north-eastern grains growing regions of Australia and related these to ground-measured plant biomass and grain yields. We evaluated crop water stress indices; standardized canopy temperature index, crop water stress index, stomatal conductance index, vapour pressure deficit, and crop stress index, which were computed from thermal imagery and on-site agro-meteorological parameters close to flowering. We then employed a classification and regression tree (CRT) as a supervised machine learning algorithm to classify crop water stress and predict biomass and grain yields as a function of crop water stress indices. The CRT accurately predicted biomass yield (coefficient of determination (R-2) = 0.86; root mean square error (RMSE) = 41.3 g/m(2) and R-2 = 0.75; RMSE = 47.7 g/m(2) for the MS and HS site) and grain yield (R-2 = 0.78; RMSE = 16.7 g/m(2) and R-2 = 0.69; RMSE = 23.2 g/m(2) for the MS and HS site, respectively). High sodic soil constraints increased crop water stress more than moderately sodic constraints soil that limits wheat yield similar to 40%. Wheat genotypes; Bremer, Gregory, Lancer, Mace, and Mitch were more productive than Gladius, Flanker, Scout, Emu Rock, and Janz in sodic soil environments. The study improves our ability to develop decision-making tools to assist farmers and breeders in securing agricultural productivity on sodic soils.

Automated summary

This study proposed a thermal remote sensing and machine learning-based approach to predict biomass and grain yields of wheat genotypes grown in sodic soil with varying water stress. The findings showed that a classification and regression tree accurately predicted biomass and grain yields, indicating that highly sodic soil constraints have a greater impact on wheat yield compared to moderately sodic soil.