High‐resolution remote sensing data improves models of species richness

Abstract

Listen

Translate article

AbstractQuestionCan predictors derived from air‐ and space‐borne high‐resolution remote sensing data improve models of species richness commonly built using coarser‐scaled environmental variables?LocationSwitzerland, covering 41 244 km2 of Central Europe.MethodsWe applied linear regressions to model species richness of woody species, herbs, edaphic bryophytes and epiphytic lichens in Swiss forests. We included high‐resolution predictors derived from digital height models and from satellite spectral images. Coarser‐scaled predictors characterizing climatic and topographic conditions were also included, as were soil properties and geology. We applied hierarchical partitioning to regression models to investigate the independent contribution of each predictor set to species richness models.ResultsPredictors derived from high‐resolution remote sensing data substantially improved the species richness models (increase 14–55% of R2). However, coarse‐scaled climatic and topographic predictors still explained a high proportion of the variance in the species richness data in all models, independently of other predictors commonly used. The importance of the remotely sensed variables was strongly dependent on the biogeographic region considered. The species richness models of smaller organisms of the forest floor (herbs and edaphic bryophytes) benefited greatly from adding high‐resolution topographic predictors, indicating the importance of microtopographic heterogeneity for these groups. Both epiphytic lichens and herbs responded strongly to indicators of structural properties of the forest stand.ConclusionsHigh‐resolution remote sensing data is a proxy for micro‐environmental structures and variation in these structures. Our results show that predictors derived from such data can improve species richness models considerably, especially in regions with low climatic and/or topographic variation. High‐resolution remote sensing variables excellently complement coarser‐scaled predictors, as they are available over large areas at low cost.