A Novel Fusion of Sentinel-1 and Sentinel-2 with Climate Data for Crop Phenology Estimation using Machine Learning
🤖 AI Summary
To address the need for precise crop phenological monitoring, this study proposes an end-to-end machine learning framework integrating time-series Sentinel-1 (SAR), Sentinel-2 (optical) remote sensing, and high-resolution climate data. Methodologically, it innovatively couples radar backscatter, red-edge vegetation indices, and dynamic temperature/precipitation features, employing a spatiotemporally adaptive weighted fusion mechanism to mitigate cloud/rain interference and regional phenological heterogeneity. Multi-source temporal alignment, sliding-window phenophase encoding, and joint preprocessing are introduced, with ensemble modeling via XGBoost and Random Forest. Validated across three major European agricultural regions, the framework achieves mean absolute errors ≤5.2 days for emergence, flowering, and maturity estimation—37% lower than single-source baselines—and attains an F1-score of 0.89. It enables sub-meter-scale field-level phenological mapping and supports crop model calibration.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Estimating crop phenology stages using Sentinel-1, Sentinel-2, and climate data
Improving phenology prediction accuracy with Machine Learning at 20m scale
Enhancing crop model calibration via fused remote sensing and weather data
Innovation
Methods, ideas, or system contributions that make the work stand out.
Fusion of Sentinel-1 and Sentinel-2 data
Machine Learning LightGBM model
High-resolution climate data integration
🔎 Similar Papers
No similar papers found.
