Seasonality and directionality effects on radar backscatter are key to identify mountain forest types with Sentinel-1 data

Systematic Sentinel-1 acquisitions provide an unprecedented stream of SAR data which allows to describe forest temporal dynamics in detail, a powerful tool for phenological studies and forest type classification. Several studies have explored the temporal variation of backscatter intensity in this context, but none considered that scattering directionality of canopies may vary. Said directionality is related to target-sensor geometry (incidence angle), forest height, and optical depth, associated with leaf dynamics. This study explicitly models backscatter dependance on incidence angle by fitting a regression model for each Sentinel-1 image and forest type. Residuals are accumulated across the time series and used to classify pixels into the most likely forest type using the smallest accumulated residual. This modelling and classification strategy has been applied over a North-South transect across the Carpathian Mountains, including forests with different physiognomies, from deciduous broadleaf forest, to mixed broadleaf-needleleaf and pure perennial needleleaf forests. These forests were classified with increasing detail, assessing the results against in-situ forest stand data and satellite-based land cover classification products (Copernicus Forest type layer). The accuracy of our classification was K > 0.8, OA > 90% when separating broadleaf from needleleaf forest types. The accuracy decreased (K > 0.6, OA > 79%) when also separating mixed forest types. Our results suggest that incorporating directional effects into classification models can improve SAR-based forest classification of temperate forest over mountainous terrain. Furthermore, models fitted between backscatter and incidence angle provide an estimate of n, a parameter related to optical depth that has been shown to vary with leaf dynamics. n could be used to improve image normalization in studies aiming at the estimation of biomass, or to aid the estimation of fast-changing parameters such as leaf area index or leaf moisture content.