Land cover mapping and change detection around Diffa-SAFE project
Prerequisites These scripts run best through RStudio in SEPAL, although it is possible to run the scripts on a Linux system.
Dependencies include: GDAL
Use SEPAL to download your single year mosaics (sentinel 1 and 2). Set your custom parameters in the first script and follow the order to run the other scripts, line by line.
Methodology/Approach: The following scripts serve to:
-
create a vegetation (shurbs + tree cover) mask year 2013 to be used/applied over the changes. based on Sentinel images combined list of indexes and normalized bands (for both wet and dry seasons) by computing mean of a weighted 3 by 3 pixels window
-
create a change detection map as result of the integration of: reclassified BFAST magnitude (thresholds have been compared: from statistics (i.e. mediam values on loss training polygons) vs visual asssessment (pixels values across validated loss) and based on patch size as follow > 0.5 ha loss (for class 4 and 5 of the magnitude of change output), > 0.5 class 3, degradation < 0.5 ha (3 -5) degradation. supervised change detection using different input layers of change and the standard deviation of a weighted window of 3 by 3 pixels. (e.g. Burnt ration, NDVI difference, IMAD for wet and dry season, BFAST raw data and ). his with a RF result of change input layers using: bfast (raw data) + IMAD raw data (wet and dry) + NDII- Normalized Difference Infrared Index (wet and dry) + NBR (Normalized BUrn Ratio) and ther difference indexes in NDVI (〖[(NDVI"humide" -NDVI"sèche" )/ (NDVI"humide" +NDVI"sèche" )]〗^2013- 〖[(NDVI"humide" -NDVI"sèche" )/ (NDVI"humide" +NDVI"sèche" )]〗^2018
-
Land cover 2018 (similar approach use in point 1.) based on Sentinel 2 and 1