Modeling European beech defoliation at a regional scale gradient in Germany from northern lowlands to central uplands using geo-ecological parameters, Sentinel-2 and National Forest Condition Survey data

Since 2018, severe droughts have affected a significant part of central Europe, causing premature leaf senescence in European beech (Fagus sylvatica L.). The correlation between the vitality of Fagus sylvatica L. and various geo-ecological and biological determinants (such as elevation, slope, aspect, tree age, and soil properties) concerning hydrological drought stress is still not well understood, especially when integrating multiple geographical datasets. In addition, the determination of crown condition by remote sensing and geo-ecological parameters is still under development; it would allow the assessment of an area-wide forest health status. Our analysis incorporated annual field data from the German National Forest Condition Survey (Waldzustandserhebung, WZE) as a response variable and employed geo-ecological parameters derived from a digital elevation model, soil properties and vegetation indices from a Sentinel-2 time series to explain and predict the crown defoliation of European beech throughout the drought-impacted period spanning 2016–2022 across the federal states Schleswig-Holstein, Lower Saxony, and Hesse of Germany. In a second step, the results of the modeling were used for mapping of crown defoliation in Hesse, Lower Saxony and Schleswig-Holstein. By employing Gradient Boosting Machines and Random Forest for regression analysis, the study uncovered the relationships between crown defoliation and the used predictors. Training was conducted on 80 % of the dataset, with the remaining 20 % serving as a test set for model validation. Regression findings based on static explanatory variable sets were improved by dynamic explanatory variables such as estimates of soil moisture, vegetation index metrics, and diameter at breast height. Furthermore, we identified key predictors for mapping crown defoliation of Fagus sylvatica L. and recommended using vegetation indices as additional predictors for future studies. The modeling results provided comparably accurate estimates compared to WZE estimates (R² of 0.794 and RMSE of 7.646 %) during testing. Topographic and static soil predictors were significant, with soil moisture being a particularly influential variable for model optimization. Based on the predicted crown defoliation, beech trees with low to moderate crown defoliation predominated in beech distribution areas across the examined federal states, while a small number of beech trees with high defoliation were identified mostly in South Lower Saxony and Hesse. The annual variations in the proportions of beech trees showing increasing and decreasing crown defoliation indicate that the condition of the crown temporarily deteriorated when soil moisture decreased, but beech trees recovered after prolonged periods of drought. Additionally, beech trees in the study region exposed to declining soil moisture may suffer from medium-term declines in vitality. The predicted crown defoliation data can be utilized for future climate-adaptive management practices in European beech forests.