Learning-Based Tracking of Crop Biophysical Variables and Key Dates Estimation From Fusion of SAR and Optical Data

Monitoring crop development is of crucial importance to ensure sustainable management practices while promoting efficient land use. The ability satellite remote sensing data cover large areas offers a robust tool aid this task. In paper, we propose filtering framework which uses Gaussian Process-based dynamic and observation models, an unscented Kalman filter (GP-UKF) fusion multitemporal SENTINEL-1 SENTINEL-2 monitor biophysical variables. This method complements state the art frameworks given its learn models uncertainties from exploit imagery temporal dimension. enables be transferable other types, variables, locations. We test methodology track asparagus below ground carbohydrates, season age forecast key dates. amount carbohydrates stored in plant's root system highly associated with yield establish healthy canopy. Validation truth showed that use more than one orbit data, provided best tracking performances reliable way for handling missing sensor. Under configuration, achieves Mean Absolute Error (MAE) 1.802 brix degrees (surrogate carbohydrates). Similarly, it can retrieve harvest date, MAE 6 days. Remotely may contribute towards reducing destructive sampling required measurement field.