Predictive mapping of alpine grasslands in Switzerland: Species versus community approach

TLDR

The study developed separate logistic regression models to predict the distribution and large‑scale spatial patterns of dominant graminoid species and communities in alpine grasslands. The models used four bioclimatic variables—degree‑days of growing season, July moisture index, March solar radiation, and continentality—derived from monthly climate data and interpolated with a 50 m digital elevation model, along with geology and slope as proxies for nutrient availability and soil water, and were calibrated with field surveys and literature data and validated on an independent test set from three climatic zones. The models achieved similar agreement for species and communities, but community predictions had higher κ‑values (0.539 vs. 0.201), and land‑use information emerged as a key factor that could further improve both models, while the climatic variables already explained most of the observed patterns.

Abstract

Abstract. Separate logistic regression models were developed to predict the distribution and large‐scale spatial patterns of dominant graminoid species and communities in alpine grasslands. The models are driven by four bioclimatic parameters: degree‐days of growing season (basis 0 °C), a moisture index for July, potential direct solar radiation for March, and a continentality index. Geology and slope angle were used as a surrogate for nutrient availability and soil water capacity. The bioclimatic parameters were derived from monthly mean temperature, precipitation, cloudiness and potential direct solar radiation. The environmental parameters were interpolated using a digital elevation model with a resolution of 50 m. The vegetation data for model calibration originate from field surveys and literature. An independent test data set with samples from three different climatic zones was used to test the model. The degree of coincidence between simulated and observed patterns was similar for species and communities, but the κ‐values for communities were generally higher (κ= 0.539) than for species (mean individual κ= 0.201). Information on land use was detected as a major factor that could significantly improve both the species and the community model. Nevertheless, the climatic factors used to drive the model explained a major part of the observed patterns.