Early-wilted forest following the Central European 2018 extreme drought
Cite this dataset
Brun, Philipp et al. (2020). Early-wilted forest following the Central European 2018 extreme drought [Dataset]. Dryad. https://doi.org/10.5061/dryad.d51c5b019
During the summer of 2018, Central Europe experienced the most extreme drought and heat wave on record, leading to widespread early leaf-shedding and die-offs in forest trees. We quantified such early-wilting responses by associating Sentinel-2 time-series statistics of the Normalized Difference Vegetation Index with visually classified orthophotos, using a random forest classifier. The predictions of our classifier achieved a high accuracy of 0.90 ±0.014 and estimated the area of affected forest at 21’500 ±2800 km2. Early wilting was especially prevalent in eastern and central Germany and in the Czech Republic and it was related to high temperatures and low precipitation at large-scales, and small to medium-sized trees, steep slopes, and shallow soils at fine-scales. The present dataset includes spatial predictons of 2018 early-wilting presence/absence for entire Central Europe (c. 800'000 km2) at 10×10 m resolution. It may be used for high-resolution studies of early-wilting patterns, to study how factors like physiology or species identity relate to early-wilting patterns, and/or as testbed for alternative approaches quantifying water stress in forests.
The data consists of spatial presence/absence predictions of a random forest classifier. As dependent variable, we used >304'000 visually interpreted pixels associated with 1022 polygons of early-wilting presence or absence. Visual interpretation was largely based on Google Earth aerial imagery with <1 m spatial resoltuion. As predictors, we used ten time-series statistics of the 2018 Normalized Difference Vegetation Index (NDVI) as sensed by the Sentinel-2 mission. These statistics included NDVI spring minimum, the fall minimum, the spring maximum, the fall maximum, the summer mean, the magnitude of change at the summer change point, the timing of the summer change point, the temporal summer trend, the significance of the temporal summer trend, and the mean absolute error during summer. We defined the forests as areas with a tree cover density of >80%, as detailed by the Tree Cover Density Map 2015 from Copernicus.
Contains: 87 raster layers in GeoTIFF format corresponding to individual tiles of the Sentiel-2 tiling system.
Values: 0 = early-witling absence, 1 = early-wilting presence, NA = non-forested area
Resolution: 10x10 m
Projection: ETRS89-extended / LAEA Europe (EPSG:3035)