Crop Yield Prediction Using Multi-Temporal Hyperspectral Data and GeoAI Deep Learning Algorithm

Accurate and timely crop yield prediction is essential for effective agricultural management and global food security. This study assesses the effectiveness of hyperspectral imagery combined with deep learning model for crop yield prediction in agricultural fields of interest. Distinct vegetation indices are derived to reflect key physiological and structural crop traits by using hyperspectral imageries from early crop growth insight for detecting stress and predicting potential yield trends to peak growth information for reliable estimates of final crop yield, along with ground truth yield data. In addition, independent ancillary datasets, such as Digital Elevation Models (DEMs), critical soil parameters, and cropping treatments, are incorporated to capture topographic and edaphic influences on crop growth. The Deep learning algorithms such as Multilayer Perceptron (MLP) are employed, and model performance evaluated using Mean Absolute Error (MAE) and coefficient of determination (R²) values. The critical role of ShortWave InfraRed (SWIR) and Visible and Near-InfraRed (VNIR) based indices are investigated with respect to the yield estimations. The proposed methodology is applied at the field-plot scale as shown in the figure, using long-term experimental data from a temperate agricultural research site in the Midwestern United States. The analysis focuses on agricultural plots within the Main Cropping System Experiment (MCSE), comprising different cropping treatments (T₁-T₄) such as:

• conventional (T₁),
• no-till (T₂),
• reduced-input (T₃), and
• biologically based practices (T₄).