Machine Learning-Based Harvest Date Detection and Prediction Using SAR Data for the Vojvodina Region (Serbia).

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Accurate crop harvest date information is crucial for agricultural logistics, machinery planning, and yield prediction.
• Machine learning (ML) offers potential for automating harvest date determination and prediction.

Purpose of the Study:

• To determine and predict harvest dates for winter wheat, maize, and soybean using ML techniques.
• To evaluate the effectiveness of C-band synthetic aperture radar (SAR) data from Sentinel-1 for harvest monitoring.

Main Methods:

• Utilized Sentinel-1 SAR data (VH and VV polarizations) from Google Earth Engine for the Vojvodina region (2017-2020).
• Employed clustering techniques (PCA, MDS, t-SNE) for data visualization and Support Vector Machine (SVM) and Multi-layer Perceptron (MLP) for harvest detection.
• Developed regression models using Random Forest (RF) and Long Short-Term Memory (LSTM) networks for harvest date prediction.

Main Results:

• Crop type significantly influenced the separability of harvested and unharvested data.
• SVM achieved high classification accuracies: 79.65% for wheat, 83.41% for maize, and 95.97% for soybean.
• LSTM network outperformed RF for harvest date prediction, yielding an R² of 0.72, MAE of 6.80 days, and RMSE of 9.25 days across all crops.

Conclusions:

• ML applied to Sentinel-1 SAR data provides a viable method for crop harvest date determination and prediction.
• The study demonstrates the utility of remote sensing data for enhancing agricultural management practices.
• LSTM networks show promise for accurate, data-driven harvest date forecasting in agriculture.