Collinearity refers to the non independence of predictor variables, usually in a regression‐type analysis. It is a common feature of any descriptive ecological data set and can be a problem for parameter estimation because it inflates the variance of regression parameters and hence potentially leads to the wrong identification of relevant predictors in a statistical model. Collinearity is a severe problem when a model is trained on data from one region or time, and predicted to another with a different or unknown structure of collinearity. To demonstrate the reach of the problem of collinearity in ecology, we show how relationships among predictors differ between biomes, change over spatial scales and through time. Across disciplines, different approaches to addressing collinearity problems have been developed, ranging from clustering of predictors, threshold‐based pre‐selection, through latent variable methods, to shrinkage and regularisation. Using simulated data with five predictor‐response relationships of increasing complexity and eight levels of collinearity we compared ways to address collinearity with standard multiple regression and machine‐learning approaches. We assessed the performance of each approach by testing its impact on prediction to new data. In the extreme, we tested whether the methods were able to identify the true underlying relationship in a training dataset with strong collinearity by evaluating its performance on a test dataset without any collinearity. We found that methods specifically designed for collinearity, such as latent variable methods and tree based models, did not outperform the traditional GLM and threshold‐based pre‐selection. Our results highlight the value of GLM in combination with penalised methods (particularly ridge) and threshold‐based pre‐selection when omitted variables are considered in the final interpretation. However, all approaches tested yielded degraded predictions under change in collinearity structure and the ‘folk lore’‐thresholds of correlation coefficients between predictor variables of |r| >0.7 was an appropriate indicator for when collinearity begins to severely distort model estimation and subsequent prediction. The use of ecological understanding of the system in pre‐analysis variable selection and the choice of the least sensitive statistical approaches reduce the problems of collinearity, but cannot ultimately solve them.

BIOMOD is a computer platform for ensemble forecasting of species distributions, enabling the treatment of a range of methodological uncertainties in models and the examination of species-environment relationships. BIOMOD includes the ability to model species distributions with several techniques, test models with a wide range of approaches, project species distributions into different environmental conditions (e.g. climate or land use change scenarios) and dispersal functions. It allows assessing species temporal turnover, plot species response curves, and test the strength of species interactions with predictor variables. BIOMOD is implemented in R and is a freeware, open source, package.Species distribution models (SDM, Guisan and Thuiller 2005) are being used in nearly all branches of life and environmental sciences. A quick search in ISI Web of Science (18/02/08) using ''species distribution models'' OR ''niche models'' OR ''habitat models'' OR ''bioclimatic models'' highlights 21 973 papers, 74% of which published in the past 10 yr, in fields as varied as environmental sciences (53% of the records), zoology (15%), marine and freshwater biology (15%), life sciences and biomedicine (9%), biodiversity and conservation (8%), evolutionary biology (8%), fisheries (6%), forestry (6%), oceanography (5%), genetics and heredity (5%), amongst others. Advancement of knowledge in these fields is now intertwined with technical innovation in species distribution modelling and dependent on the existence of suitable software for fitting models and examining results. One difficulty with the use of species distribution models is that the number of techniques available is large and is increasing steadily, making it difficult for ''non-aficionados'' to select the most appropriate methodology for their needs (Elith et al. 2006, Heikkinen et al. 2006. Recent analyses have also demonstrated that discrepancies between different techniques can be very large, making the choice of the appropriate model even more difficult. This is particularly true when models are used to project distributions of species into independent situations, which is the example of projections of species distributions under future climate change scenarios (Thuiller 2004, Pearson et al. 2006. A possible solution to account for this inter-model variability is to fit ensembles of forecasts by simulating across more than one set of initial conditions, model classes, model parameters, and boundary conditions (for a review see Araújo and New 2007) and analyse the resulting range of uncertainties with bounding box, consensus and probabilistic methodologies rather than lining up with a single modelling outcome New 2007, Thuiller 2007). BIOMOD offers such a platform for ensemble forecasting (Fig. 1) using freeware and open-source R software (R Development Core Team 2008). It overcomes some of the limitations of existing software (e.g. being able to fit and compare different models) and incorporates several features for testing models (e.g. k-fold cross validation) ...

Many species are projected to become vulnerable to twenty-first-century climate changes, with consequent effects on the tree of life. If losses were not randomly distributed across the tree of life, climate change could lead to a disproportionate loss of evolutionary history. Here we estimate the consequences of climate change on the phylogenetic diversities of plant, bird and mammal assemblages across Europe. Using a consensus across ensembles of forecasts for 2020, 2050 and 2080 and high-resolution phylogenetic trees, we show that species vulnerability to climate change clusters weakly across phylogenies. Such phylogenetic signal in species vulnerabilities does not lead to higher loss of evolutionary history than expected with a model of random extinctions. This is because vulnerable species have neither fewer nor closer relatives than the remaining clades. Reductions in phylogenetic diversity will be greater in southern Europe, and gains are expected in regions of high latitude or altitude. However, losses will not be offset by gains and the tree of life faces a trend towards homogenization across the continent.