Answers are grounded in this paper and cite their sources.
In this work, we present a novel dynamical framework for crop monitoring that integrates SAR and optical satellite imagery. The proposed methodology employs two-dimensional state transitions to explicitly model the expected temporal evolution of crops. By combining such crop modeling with multi-sensor observations, including dual-polarimetric (VV and VH) radar images gathered by Sentinel-1 and optical imagery from Sentinel-2, the developed framework produces robust and consistent estimations using the well-established Kalman Filter. Specifically, the approach is applied to retrieve the biomass values of sugarcane crops located in the Cauca Valley (Colombia), over a full growing season (from summer 2024 to summer 2025). An intensive field campaign was carried out, so that ground-truth measurements in a large number of points, located in multiple fields, were available to develop and validate the proposed integration algorithm. Results show that biomass is accurately estimated especially when all available observations, both SAR and optical, are employed. In this case, a global root mean square error of 0.103 kg, a mean absolute error of 0.064 kg and an accumulated biomass difference of 13.8 kg·day are obtained after evaluating the model multiple times on randomly-selected points. The analysis also highlights a key limitation of optical imagery in the study area, as a large number of images are affected by cloud cover, resulting in long gaps within the time series and reduced estimation accuracy. Furthermore, the best performance of the proposed integrated model is achieved when all available SAR products, including dual-polarization (VV and VH) coherence and intensity products, are exploited, although very good results are also obtained using single-polarization coherence and intensity data (either at VV or VH polarizations). Comparison with alternative approaches, including partially integrated methods using only one type of information (either SAR or NDVI), and methods based solely on machine learning, indicates that the proposed methodology consistently offers superior accuracy in biomass estimation.
No takes yet. Share an insight, caveat, or question.
Mestre-Quereda et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: