Accurate forecasting of cotton canopy temperature ( T c ) can effectively predict water stress for cotton crops, informing timely irrigation decisions. The BIOTIC (Biologically-Identified Optimal Temperature Interactive Console) approach monitors plant stress by measuring the duration in hours (stress-hours) that T c exceeds an optimal temperature of 28 °C. Accumulation of stress-hours is directly related to plant water stress, influences yield, and can be used to guide irrigation planning. A successful forecast of T c must utilise weather information to predict T c , particularly periods when especially T c > 28 °C, and stress-hours accurately. Current T c models lack advanced time series techniques and robust evaluation, particularly for forecasting high T c and stress-hours. Time-series models (lightGBM and auto-ARIMA) are used with existing T c forecasting models (periodiCT and periodiCTs) to forecast T c , T c > 28 °C and stress-hours across three field-year combinations across NSW, Australia. LightGBM improves T c > 28 °C predictions by 13% and stress-hour predictions by 18% over the current best method and by 30% over the current industry approach. Models had only slightly reduced forecasting ability for T c > 28 °C and stress-hours when using just air temperature over additional biologically-relevant information (solar radiation, relative humidity and wind speed), providing accurate forecasts with minimal data. Additionally, reducing T c transmission frequency from 0.25 hours to one hour reduces cost without compromising model accuracy. This study establishes an effective forecasting methodology for cotton T c , enabling precise BIOTIC-based management of cotton crops, and shows the power of data-driven models to improve crop water management and canopy monitoring. • Improved forecasts of canopy temperature ( T c ) for accurate detection of water stress. • Using best time-series models on 4–7 day horizons for 3 cotton irrigation treatments. • Evaluate on T c > 28 °C and stress-hours to link performance to plant water stress. • Machine learning outperforms current best T c model on water stress metrics by 15%. • Accurate forecasts for water-stress conditions achieved using only air temperature.
Rogers et al. (Tue,) studied this question.