Abstract. 1. Red-listed species are species of great conservation concern, but they are inherently difficult to observe. Therefore, prediction models for hotspots of such species would be very useful for biodiversity monitoring and management.
2. Red-listed species are ecologically heterogeneous, and they may be associated with many different environmental variables that may be more or less correlated. To overcome these challenges, we introduce a new technique to species richness modelling: Partial Least Squares Regression (PLSR), a multivariate regression technique developed specifically for similar statistical problems in econometrics.
3. Partial Least Squares Regression models were developed for hotspots of red-listed beetles associated with oaks. In this system, environmental variables and the richness of red-listed species were often correlated.
4. Without exceptions, the PLSR performed better as prediction models than generalised linear models (GLM) and principal components regression (PCR).
5. Different combinations of variables associated with individual trees were important in continuous forest with a closed canopy and a natural understory vs. man-made wooded park landscapes.
6. Our results suggest that important variables related to hotspots for red-listed species in forests in forests are trees with cavities of intermediate size, as well as dead wood and broadleaved tree species other than oak in the surrounding forest. In parks, large trees with wood mould, but also other oak trees in the surroundings, seem to be important.
7. In conclusion, the PLSR models may be used to improve the prediction of species richness in the presence of many correlated predictors and thus improve biodiversity monitoring and management.