Enhancing winter wheat growth indicator prediction with multi-task learning and multi-source data

Timely and accurate prediction of wheat growth indicators is crucial for yield enhancement and extreme weather impact mitigation. Research on efficient monitoring of growth indicators using multi-task learning combined with multi-source information remains limited. Furthermore, the growth stage-specific prediction should be emphasized to reveal the effect of growth stages on the indicators. This study aims to predict winter wheat growth indicators at different growth stages using machine learning, deep learning, and multi-task learning based on multi-source and multi-temporal features, such as from spectral, moisture, and meteorological data, to evaluate and improve the accuracy of prediction. Field-collected growth indicators including leaf area index (LAI), chlorophyll (CHL), plant nitrogen accumulation (PNA), plant dry matter (PDM), plant nitrogen content (PNC), nitrogen nutrition index (NNI), and the above features were analyzed alongside feature selection based on Pearson correlation coefficients (PCCfs). Models were developed using Random Forest (RF), Long Short-Term Memory (LSTM), and Multi-Task Learning (MTL), with consideration given to the contribution of features to indicators. The results demonstrated that RF model outperformed LSTM, with average R2 values ranging from 0.54 to 0.92 versus 0.08-0.88, respectively. The MTL enhanced model speed and accuracy, particularly with large datasets or deep learning applications. Each indicator exhibited optimal performance at specific growth stages, such as LAI during the jointing and PDM during the flowering. Vegetation Index (VI) emerged as the most significant features for growth indicators, followed by the canopy equivalent water thickness (CEWT) and meteorological features. This study presents a novel approach to winter wheat growth indicator prediction, significantly enhancing prediction accuracy and contributing to the achievement of precise field management.