he Sentinel-1 GRD high resolution S

he Sentinel-1 GRD high resolution SAR data was pre-processed with Joanneum Research RSG software Version 7.51.0 (www.remotesensing.at, [8]). First, we ingested the images and updated the orbit parameters. Then, we processed the images to gamma naught based on the SRTM (Shuttle Radar Topography Mission) digital elevation model. Next, we applied a multi-looking of 2 by 2 ground range looks to 20m final pixel spacing. Multi-looking can be seen as a kind of mean filter in radar data processing. We registered each image to a master image (first image of the stack) to avoid geometrical inconsistencies. For noise reduction, we employed a Quegan multi-temporal filter with a 3×3 spatial window [50]. Multi-temporal speckle filtered SAR images usually show higher performance in terms of noise reduction and preservation of spatial information than single speckle-filtered SAR images or unfiltered SAR images [51]. The stack of time series data was then analyzed by statistical methods resulting in the following statistics stack: mean, minimum and maximum backscatter, standard deviation, coefficient of variation, trend between first three and last three images. The registered stack of statistics images was then ortho-rectified to 20 m spatial resolution for further use. The main reasons for using this stack as opposed to the individual images are twofold: first, the “mean” image showsmuchlowerspecklethantheindividualimages,whichimprovesclassificationaccuracy. Second, different land cover classes show specific behavior over the year, which are well represented by “standard deviation” and “coefficient of variation” of the stack.