Speaker
Description
Reliable estimates of above-ground biomass in mountain grasslands are crucial for assessing production losses and supporting index-based insurance schemes, particularly as climate change causes more frequent drought events increasingly threatening alpine agriculture. However, biomass retrieval from SAR imagery remains challenging in mountainous regions because complex terrain introduces severe geometric distortions and radiometric variability into radar observations. These effects complicate the interpretation of radar signals and can reduce the reliability of retrieval methods.
Over permanent meadows of South Tyrol, in Italy, we previously investigated machine learning (Göhner, 2024) and deep learning (Singh et al., 2025) model architectures combining Sentinel-1 observations and ancillary variables to fill gaps in Sentinel-2 Leaf Area Index (LAI) time series, using LAI as a proxy for permanent meadow biomass (Castelli et al., 2023). However, their predictions were predominantly driven by meteorological and temporal features rather than by the EO signal itself. This limited the models’ capacity to exploit the information conveyed by Sentinel-1 backscatter and motivated a shift towards the direct estimation of above-ground biomass. Achieving this objective first requires a better understanding of how radar backscatter responds to grassland biophysical properties and how this response is influenced by terrain, vegetation development, and agricultural practices such as mowing events.
Accordingly, this study moves towards a physically informed analysis of Sentinel-1 observations, starting from a sensitivity analysis of amplitude-based SAR features to in situ measurements of vegetation water content, above-ground biomass, sward height, and LAI, as well as to agricultural practices, using observations collected over two years across eight permanent meadows in South Tyrol (Castelli et al., 2026). Particular attention is given to the role of acquisition geometry, as the interaction between the radar signal and the complex alpine terrain produces substantial spatial variability in the local viewing geometry. We therefore analyse the Projected Local Incidence Angle to determine whether this topographic variability can provide a proxy for multi-angular observations and support the identification of dominant scattering mechanisms and their seasonal evolution. This sensitivity analysis provides the physical boundary conditions for a subsequent biomass retrieval using inversion frameworks such as the Water Cloud Model, thereby supporting more robust and interpretable above-ground biomass estimation in managed alpine grasslands.
Preliminary results confirmed the feasibility of exploiting the topographic complexity of South Tyrol to construct a multi-angular time series from Sentinel-1 SAR observations over permanent grasslands. Specifically, more than 85% of permanent meadow pixels in the study area exhibited a PLIA variation range exceeding 8°, meeting the threshold established in the literature from Hahn et al. (2017) for reliable angular characterization. Furthermore, the temporal window required to compose this multi-angular time series ranged between 5 and 11 days during 2023–2024 (operating with Sentinel-1A only), when in situ data are available. Further analyses will evaluate the feasibility of leveraging this temporal density for dynamic time-series smoothing, aiming to isolate seasonal phenological trends from rapid environmental noise without compromising the detection of abrupt biomass changes caused by mowing events.
References
Göhner, C. (2024). Modelling LAI of Alpine Grassland from Sentinel-1 SAR and Sentinel-2 Data using Spatial Gap-filling in North-Eastern Italy. Zenodo. https://doi.org/10.5281/zenodo.17865651
Singh A., et al. (2025). Transformer-based Convolutional Regressor to include SAR Backscatter Signals in Monitoring Alpine Grasslands. IEEE Geoscience and Remote Sensing Letters. Doi: 10.1109/LGRS.2025.3645675
Castelli et al. (2023). Insuring Alpine Grasslands against Drought-Related Yield Losses Using Sentinel-2 Satellite Data. Remote Sensing. Doi: 10.3390/rs15143542
Castelli M. et al. (2026). Two years of in situ measurements of biophysical variables of mountain meadows in northeastern Italy. Scientific Data. https://doi.org/10.1038/s41597-026-08074-6
Hahn S. et al. (2017), Dynamic Characterization of the Incidence Angle Dependence of Backscatter Using Metop ASCAT. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. Doi: 10.1109/jstars.2016.2628523.