Multi-Stage Corn Yield Prediction Using High-Resolution UAV Multispectral Data and Machine Learning Models

Timely and cost-effective crop yield prediction is vital in management decision-making. This study evaluates the efficacy of Unmanned Aerial Vehicle (UAV)-based Vegetation Indices (VIs) coupled with Machine Learning (ML) models for corn (Zea mays) at vegetative (V6) reproductive (R5) growth stages using a limited number training samples farm scale. Four agronomic treatments, namely Austrian Winter Peas (AWP) (Pisum sativum L.) cover crop, biochar, gypsum, fallow sixteen replications were applied during non-growing season to assess their impact on following yield. Thirty different variables (i.e., four spectral bands: green, red, red edge, near-infrared twenty-six VIs) derived from UAV multispectral data collected V6 R5 utility prediction. Five ML algorithms including Linear Regression (LR), k-Nearest Neighbor (KNN), Random Forest (RF), Support Vector (SVR), Deep Neural Network (DNN) evaluated One-year experimental results treatments indicated negligible overall Red canopy chlorophyll content index, edge absorption ratio green normalized difference vegetation band, index among most suitable predicting The SVR predicted Coefficient Determination (R2) Root Mean Square Error (RMSE) 0.84 0.69 Mg/ha 0.83 1.05 stage, respectively. KNN achieved higher accuracy AWP (R2 = RMSE 0.64 1.13 R5) gypsum treatment 0.61 1.49 0.80 1.35 R5). DNN biochar 0.71 1.08 0.74 1.27 For combined (AWP, fallow) treatment, produced accurate an R2 0.36 1.48 0.41 1.43 R5. Overall, treatment-specific was more than treatment. Yield accurately other regardless model used. outperformed Yields similar both stages. Thus, this demonstrated that VIs can be used multi-stage scale, even data.