Integrating Imaging Spectroscopy-Derived Vegetation Traits into Species Distribution Modelling: A Case Study of the Alpine Marmot (Marmota marmota) in the Swiss National Park

Kurzfassung

Species distribution models (SDMs) are widely used to understand species–environment relationships and support biodiversity conservation and management. SDMs commonly rely on climatic, topographic and other environmental predictors, while detailed information on vegetation characteristics is less frequently incorporated. Imaging spectroscopy can provide detailed spatial information on vegetation structure as well as biophysical and biochemical properties that may improve the modelling of animal habitat. This study investigated whether imaging spectroscopy-derived vegetation traits provide additional information beyond conventional environmental predictors in animal SDMs, using the alpine marmot (Marmota marmota) in the Swiss National Park (SNP) and surrounding areas as a model species. Two predictor configurations at 10 m spatial resolution were compared: one (ENV-only) based on environmental variables only and the other (ENV+VEG) combining environmental and imaging spectroscopy-derived vegetation variables. Correlated predictors were reduced to multivariate gradients using principal component analysis (PCA), and SDMs were fitted using generalized linear models (GLM), generalized additive models (GAM), boosted regression trees (BRT), and random forests (RF). Model performance was evaluated using the area under the curve (AUC) and true skill statistic (TSS) with spatial cross-validation, which was also used to assess variable importance. Final ensemble models based on GAM, BRT, and RF were used to compare spatial patterns of predicted habitat suitability. ENV+VEG models generally achieved higher predictive performance than the corresponding ENV-only models across all modelling algorithms. In the ENV-only models, a topographic gradient combining slope and terrain ruggedness was the most important predictor, accounting for approximately 22% of total variable importance. With the inclusion of vegetation variables, vegetation-related gradients became particularly important. The vegetation development gradient was the most influential predictor, accounting for approximately 47% of total variable importance, while a biochemical trait gradient contributed a further 9%. Overall spatial patterns of predicted habitat suitability remained broadly similar between the two predictor configurations, with differences mainly occurring in localised areas. Overall, the results show that imaging spectroscopy-derived vegetation traits can provide additional ecological information beyond conventional environmental predictors in animal SDMs. Imaging spectroscopy may therefore be a useful addition to conventional environmental data, particularly when vegetation characteristics are relevant to the habitat requirements of the target species. Further studies are needed to assess the potential of this approach for different species and under different ecological settings.