Comparison of the performance of species response models in several landscape units in the province of Noord-Holland, The Netherlands

In The Netherlands, empirical statistical response models are used in the Dutch nature conservation policy to predict vegetation response to management options. The development of empirical statistical models is time consuming and costly, due to the need for databases that extensively describe the biotic and environmental variation in the research area. For this reason it is important to optimize the sampling procedure and thus reduce costs. Our research aimed at finding out to what extent species response models will improve when an environmental gradient is sampled outside the range optimal for a species, which may be especially relevant where species presence is related to geographical landscape units. We tested for 46 species and two landscape units in our research area, a dune and a polder area in the province of Noord-Holland, if it would be worthwhile to extend sampling to the other landscape unit. Herefore we compared the predictive values of response models based on either a dune or polder subset with those of response models based on a combined dune/polder set. In general the predictive values of models based on the combined set were not significantly higher than those of models based on the subsets. Thus, where time and financial resources are limited, we recommend focusing sampling attention especially on the regions of interest. Extrapolation of subset response models to the other landscape unit was also tested. Considering the extreme poor predictions, we advise against the extrapolation of response models to regions with different environmental conditions. (C) 1998 Elsevier Science B.V. All rights reserved.