Quantification of forest extent in Germany by combining multi-temporal stacks of Sentinel-1 and Sentinel-2 images

Abstract

Information regarding the extents of forests and forest biomass is crucial for the quantification of the terrestrial carbon budget. While field surveys are time consuming and expensive, remote sensing techniques offer an efficient and fast alternative. The Copernicus programme provides large amounts of Synthetic Aperture Radar (SAR) and multi-spectral data that can be used for this purpose. This study presents two methods for forest cover classification, one using a multitemporal dataset of SAR images from one orbit, and the other combining SAR images acquired in both ascending and descending orbits and almost cloud-free (<10%) multi-spectral images. The SAR-LC classification system, a rule-based decision tree which is designed to classify land cover types using radar backscatter is used to extract forest cover extent from the 2016 dataset. For the second method, Sentinel-1 images from 2017 in both ascending and descending orbits are combined with 10 m resampled almost cloud free Sentinel-2 images to form one multi-temporal dataset with 88 bands. This is then segmented into objects before forest extent is classified using a rule-based classification. The SAR-LC thresholds were optimised to include the ReNDVI values from the Sentinel-2 images for this purpose. While the final objective of this study is to produce forest cover maps for the whole of Germany, this paper will only focus on the forests around the region of Frankfurt. The challenges, limitations and accuracy of each method is reported and discussed.