Modelling of Intra-field Winter Wheat Crop Growth Variability Using In situ Measurements, UAV derived Vegetation Indices, Soil Properties, and Machine Learning Algorithms

Monitoring crop growth conditions during the growing season provides an indication of crop health and informs agricultural management. Since physical and chemical properties of soils tend to be spatially heterogeneous, intra-field soil variability is bound to cause intra-field heterogeneity in crop growth rate. Data fusion of soil properties and derived unmanned aerial vehicle (UAV) Vegetation Indices (Vis) can help to improve model performance accuracy for crop growth assessment. The aim of this study was to investigate and understand the contribution of soil properties and unmanned aerial vehicle UAV data to improve modelling accuracy of intra-field crop growth variability for winter wheat. To achieve this aim, the study used soil data, monthly time-series crop height measurements (cm), and Vis acquired through UAV time-series images. For data analysis, two machine learning methods including optimizable Gaussian process regression (GPR) and optimizable least-squares boosting (LSboost) and bagging (Bag) Ensemble regression (ER) were applied in MATLAB software. Results showed that soil properties, particularly Ca, Mg, K, and Clay were more important than VIs in representing actual crop growth. However, when VIs and soil properties were integrated, the GPR model’s coefficient of determination (R²) improved by 0.01 and 0.03, while the RMSE decreased by 0.25 and 0.78 cm for the two farms, respectively. Overall, GPR (R² = 0.68 to 0.75, RMSE = 15.85 to 18.38 cm) performed slightly better than LSboost-Bag-ER (R² = 0.64 to 0.70 and RMSE = 17.26 to 19.34 cm) for both farms. The findings of this study show that, although intra-field crop growth variability is reasonably predicted by soil properties, UAV data, and Vis separately/independently, the synergistic use of these data sources produced better results than the individual datasets.