Title: Predicting species distributions from museum and herbarium records using multiresponse models fitted with multivariate adaptive regression splines Running head: Modelling presence-only records with MARS

In press. Predicting species distributions from museum and herbarium records using multiresponse models fitted with multivariate adaptive regression splines. Diversity and Distributions. (A) Abstract 1 Current circumstances-that the majority of species distribution records exist as 2 presence-only data (e.g., from museums and herbaria), and that there is an established need 3 for predictions of species distributions-mean that scientists and conservation managers seek 4 to develop robust methods for using these data. Such methods must, in particular, 5 accommodate the difficulties caused by lack of reliable information about sites where species 6 are absent. Here we test two approaches for overcoming these difficulties, analyzing a range 7 of datasets using the technique of multivariate adaptive regression splines (MARS). MARS is 8 closely related to regression techniques such as generalized additive models (GAMs) that are 9 commonly and successfully used in modeling species distributions, but has particular 10 advantages in its analytical speed, and the ease of transfer of analysis results to other 11 computational environments such as a Geographic Information System. MARS also has the 12 advantage that it can model multiple responses, meaning that it can combine information from 13 a set of species to determine the dominant environmental drivers of variation in species 14 composition. We use data from 226 species from six regions of the world, and demonstrate 15 the use of MARS for distribution modeling using presence-only data. We test whether (i) the 16 type of data used to represent absence or background, and (ii) signal from multiple species, 17 affect predictive performance, by evaluating predictions at completely independent sites 18 where genuine presence-absence data were recorded. Models developed with absences 19 inferred from the total set of presence-only sites for a biological group, and using 20 simultaneous analysis of multiple species to inform the choice of predictor variables, 21 performed better than models in which species were analyzed singly, or in which pseudo-22 absences were drawn randomly from the study area. The methods are fast, relatively simple to 23 understand, and useful for situations where data are limited. (A) Introduction 25 Species, communities and ecosystems are distributed across the earth in interesting and 26 complex patterns. These stimulate scientific research not only for their own sake (Whittaker, 27 1967; Levin, 1992; Graham et al., 2004a) but also because widespread clearing and other 28 disturbances threaten the existence of many species and ecosystems (Margules & Pressey, 29 …