waldmonitoring.ch: NDVI difference rasters for annual forest change in Switzerland (Sentinel 2 based): 2016 - 2023

NDVI difference rasters for annual forest change in Switzerland (Sentinel 2 based): 2016 - 2023

Date format: GeoTIFFData type: Int16 - Sixteen bit signed integer*Spatial Resolution: 10 x 10 mSpatial Extent: Switzerland and Liechtenstein, masked with swisstopo swissTLM3D Forest Mask (2021) Coordinate Reference System: EPSG:2056 - CH1903+ / LV95, Swiss. Obl. Mercator*: NDVI Difference Values (-1 to 1) are multiplied by 10'000 to allow using Integer 16 bit vs. Float 32 bit while maintaining a precision of 5 digits. The values have to be interpreted accordingly: -10'000 means an NDVI difference of -1, +10'000 an NDVI difference of +0.

The NDVI difference rasters for annual forest change in Switzerland are created by using Sentinel 2 based NDVI composites (Normalized Difference Vegetation Index). The code for the generating method can be found in the waldmonitoring-repository, the method itself is also described and translated in further detail in the corresponding waldmonitoring-wiki: For the automatic detection of areas of change, the differences between two years were examined using the NDVI. In order to automatically filter out cloudy images, the maximum NDVI value of all available images of the summer months (June - August) was used for each pixel (10 x 10 m). During this time, practically all the vegetation is green. This results in almost cloud-free, annual raster images with the maximum NDVI ("NDVI maximum composite"). The difference between two years is formed from these composites. The difference values accordingly reflect the strength of the change. 

As an example for interpretation, values of -0.1 or smaller (closer to -1.0) indicate strong forest changes (e.g. clearing), whereas positive values indicate vegetation regeneration or re-greening of previously unvegetated areas. Using a threshold value (we suggest -0.06 for forest applications), areas with considerable negative change can be separated out and be vectorized (converted to polygons) to create a dataset that can be queried.