Predictive vegetation mapping can be defined as predicting the geographic distribution of the vegetation composition across a landscape from mapped environmental variables. Comput erized predictive vegetation mapping is made possible by the availability of digital maps of topography and other environmental variables such as soils, geology and climate variables, and geographic information system software for manipulating these data. Especially important to predictive vegetation mapping are interpolated climatic variables related to physiological tolerances, and topographic variables, derived from digital elevation grids, related to site energy and moisture balance. Predictive vegetation mapping is founded in ecological niche theory and gradient analysis, and driven by the need to map vegetation patterns over large areas for resource conservation planning, and to predict the effects of environmental change on vegetation distributions. Predictive vegetation mapping has advanced over the past two decades especially in conjunction with the development of remote sensing-based vegetation mapping and digital geographic information analysis. A number of statistical and, more recently, machine-learning methods have been used to develop and implement predictive vegetation models.

Aim  To demonstrate that multi‐modelling methods have effectively been used to combine static species distribution models (SDM), predicting the geographical pattern of suitable habitat, with dynamic landscape and population models to forecast the impacts of environmental change on species’ status, an important goal of conservation biogeography. Methods  Three approaches were considered: (1) incorporating models of species migration to understand the ability of a species to occupy suitable habitat in new locations; (2) linking models of landscape disturbance and succession to models of habitat suitability; and (3) fully linking models of habitat suitability, habitat dynamics and spatially explicit population dynamics. Results  Linking species–environment relationships, landscape dynamics and population dynamics in a multi‐modelling framework allows the combined impacts of climate change (affecting species distribution and vital rates) and land cover dynamics (land use change, altered disturbance regimes) on species to be predicted. This approach is only feasible if the life history parameters and habitat requirements of the species are well understood. Main conclusions  Forecasts of the impacts of global change on species may be improved by considering multiple causes. A range of methods are available to address the interactions of changing habitat suitability, habitat dynamics and population response that vary in their complexity, realism and data requirements.

A large array of species distribution model (SDM) approaches has been developed for explaining and predicting the occurrences of individual species or species assemblages. Given the wealth of existing models, it is unclear which models perform best for interpolation or extrapolation of existing data sets, particularly when one is concerned with species assemblages. We compared the predictive performance of 33 variants of 15 widely applied and recently emerged SDMs in the context of multispecies data, including both joint SDMs that model multiple species together, and stacked SDMs that model each species individually combining the predictions afterward. We offer a comprehensive evaluation of these SDM approaches by examining their performance in predicting withheld empirical validation data of different sizes representing five different taxonomic groups, and for prediction tasks related to both interpolation and extrapolation. We measure predictive performance by 12 measures of accuracy, discrimination power, calibration, and precision of predictions, for the biological levels of species occurrence, species richness, and community composition. Our results show large variation among the models in their predictive performance, especially for communities comprising many species that are rare. The results do not reveal any major trade‐offs among measures of model performance; the same models performed generally well in terms of accuracy, discrimination, and calibration, and for the biological levels of individual species, species richness, and community composition. In contrast, the models that gave the most precise predictions were not well calibrated, suggesting that poorly performing models can make overconfident predictions. However, none of the models performed well for all prediction tasks. As a general strategy, we therefore propose that researchers fit a small set of models showing complementary performance, and then apply a cross‐validation procedure involving separate data to establish which of these models performs best for the goal of the study.

Recent studies suggest that species distribution models (SDMs) based on fine-scale climate data may provide markedly different estimates of climate-change impacts than coarse-scale models. However, these studies disagree in their conclusions of how scale influences projected species distributions. In rugged terrain, coarse-scale climate grids may not capture topographically controlled climate variation at the scale that constitutes microhabitat or refugia for some species. Although finer scale data are therefore considered to better reflect climatic conditions experienced by species, there have been few formal analyses of how modeled distributions differ with scale. We modeled distributions for 52 plant species endemic to the California Floristic Province of different life forms and range sizes under recent and future climate across a 2000-fold range of spatial scales (0.008-16 km 2 ). We produced unique current and future climate datasets by separately downscaling 4 km climate models to three finer resolutions based on 800, 270, and 90 m digital elevation models and deriving bioclimatic predictors from them. As climate-data resolution became coarser, SDMs predicted larger habitat area with diminishing spatial congruence between fine-and coarse-scale predictions. These trends were most pronounced at the coarsest resolutions and depended on climate scenario and species' range size. On average, SDMs projected onto 4 km climate data predicted 42% more stable habitat (the amount of spatial overlap between predicted current and future climatically suitable habitat) compared with 800 m data. We found only modest agreement between areas predicted to be stable by 90 m models generalized to 4 km grids compared with areas classified as stable based on 4 km models, suggesting that some climate refugia captured at finer scales may be missed using coarser scale data. These differences in projected locations of habitat change may have more serious implications than net habitat area when predictive maps form the basis of conservation decision making.

Generalized additive, generalized linear, and classification tree models were developed to predict the distribution of 20 species of chaparral and coastal sage shrubs within the southwest ecoregion of California. Mapped explanatory variables included bioclimatic attributes related to primary environmental regimes: averages of annual precipitation, minimum temperature of the coldest month, maximum temperature of the warmest month, and topographically-distributed potential solar insolation of the wettest quarter (winter) and of the growing season (spring). Also tested for significance were slope angle (related to soil depth) and the geographic coordinates of each observation. Models were parameterized and evaluated based on species presence/absence data from 906 plots surveyed on National Forest lands. Although all variables were significant in at least one of the species' models, those models based only on the bioclimatic variables predicted species presence with 3-26 % error. While error would undoubtedly be greater if the models were evaluated using independent data, results indicate that these models are useful for predictive mapping-for interpolating species distribution data within the ecoregion. All three methods produced models with similar accuracy for a given species; GAMs were useful for exploring the shape of the response functions, GLMs allowed those response functions to be parameterized and their significance tested, and classification trees, while sometimes difficult to interpret, yielded the lowest prediction errors (lower by 3-5 %).