Modelling Changes in Basal Area and Broadleaf Fraction in Swiss Forests Using Machine Learning
Predicting local forest responses to climate change at the Swiss national scale is a challenge that has not been fully addressed with validated, data-driven models. This thesis develops a machine learning model for Swiss forests, using an Extreme Gradient Boosting (XGBoost) algorithm trained on data from the Swiss National Forest Inventory and CH2018 climate scenarios to forecast changes in basal area and broadleaf fraction. The model is compared against a linear Lasso baseline using temporally independent data. The XGBoost model performed slightly better, achieving R2 of 0.659 for predicting basal area and 0.906 for broadleaf fraction. Further the model is used to make projections to 2099 indicate that basal area will decrease at lower elevations while increasing at higher elevations, and the broadleaf fraction is projected to increase at lower elevations. The results of the projections are compared with ones coming from ForClim, a processed based model.
By: Dea Rieder
Work: Bachelor thesis, 2025
Supervisor: Dr. Giacomo Vaccario (ETH Zurich)
