Building realistic assemblages with a Joint Species Dis-

12 1. Species distribution models (SDMs) can be used to predict how individual species—and 13 whole assemblages of species—will respond to a changing environment. Until now, these 14 models have either assumed (1) that species’ occurrence probabilities are uncorrelated, 15 or (2) that species respond linearly to preselected environmental variables. These two 16 assumptions currently prevent ecologists from modeling assemblages with realistic co17 occurrence and species richness properties. 18 2. This paper introduces a stochastic feedforward neural network, called mistnet, which makes 19 neither assumption. Thus, unlike most SDMs, mistnet can account for non-independent 20 co-occurrence patterns driven by unobserved environmental heterogeneity. And unlike 21 recently proposed Joint SDMs, mistnet can also learn nonlinear functions relating species’ 22 occurrence probabilities to environmental predictors. 23 3. Mistnet makes more accurate predictions about the North American bird communities 24 found along Breeding Bird Survey transects than several alternative methods tested. In 25 particular, typical assemblages held out of sample for validation were nearly 50,000 times 26 more likely under the mistnet model than under independent combinations of single-species 27 models. 28 4. Apart from improved accuracy, mistnet shows two other important benefits for ecological 29 research and management. First: by analyzing co-occurrence data, mistnet can identify 30 unmeasured—and perhaps unanticipated—environmental variables that drive species 31 turnover. For example, mistnet identified a strong grassland/forest gradient, even though 32 only temperature and precipitation were given as model inputs. Second: mistnet is able 33 2 . CC-BY 4.0 International license peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/003947 doi: bioRxiv preprint first posted online Apr. 9, 2014;