PhD defense Dostalova

Alena Dostalova
Towards the European-Wide Forest Mapping and Classification Using the Sentinel-1 C-Band Synthetic Aperture Radar Data

Forests cover around 38% of the European land surface and are of great economic and ecological importance. Reliable and frequently updated information on forest resources is needed to maintain the functioning of forest ecosystems. Nowadays, terrestrial in-situ observations are complemented by remote sensing techniques that provide area-wide spatial data of many forestry parameters such as forest cover, forest type and composition or biomass. While airborne campaigns can provide high level of details with limited coverage and temporal resolution, the spaceborne remote sensing provides regular acquisitions that help to bridge the gap in the spatial and temporal coverage.In recent years, satellite-based forest maps became available for whole countries, continents, or the entire world. Currently, the majority of these global or continental forest datasets exploit optical data. However, the temporal coverage of optical datasets is limited due to frequent cloud coverage or limited sun illumination of the surface in some areas. For this reason, the synergy of Synthetic Aperture Radar (SAR) and optical sensors or SAR-only based products are also increasingly addressed by research.Launch of the Sentinel-1 constellation in 2014 and 2016 for Sentinel-1 A and B respectively secured a regular global coverage of high-resolution C-Band SAR data. Over the majority of the European continent, 4 to 8 dual polarisation acquisitions are provided every 12 days resulting in very dense time series. Such dense multi-temporal dataset can help to overcome some of the known limitations of C-Band SAR in forestry applications, such as the backscatter saturation at moderate growing stocks or the sensitivity of the signal to the environmental conditions such as moisture or freeze/thaw events.The availability of dense time-series of Sentinel-1 data motivated the development of new forest mapping and classification algorithm that exploits the differences between the temporal signatures of various vegetation types. The suitability of this algorithm for continental-scale forest classification was tested by applying and validating it for the entire European continent. This validation revealed high correspondence with the European-wide Copernicus High Resolution Layers forest datasets (overall accuracies of 86.1% and 73.2% for the forest/non-forest and forest type maps respectively and Pearson correlation coefficient of 0.83 for tree cover density map) as well as with national forest maps (average overall accuracy of 88.2% and 82.7% for forest/non-forest and forest type maps respectively). These results show, that the Sentinel-1 SAR sensors are well suited for the forest mapping and forest type classification over majority of the European Continent. This is especially promising due to the fact, that these maps can be produced with a high degree of automation and that only a single year of Sentinel-1 data is required. Also, further improvements can be achieved in undulated regions by including an additional radiometric terrain flattening step in the SAR data pre-processing.

Wednesday, 04.05.2022, 13:00

 April 19, 2022

 Microwave Remote Sensing