Species distribution modeling with Gaussian processes: A case study with the youngest stages of sea spawning whitefish (Coregonus lavaretus L. s.l.) larvae

We propose to use Gaussian processes for species distribution modeling. The objectives are to predict the spatial occurrence pattern and to identify the environmental variables that describe the occurrence. Our model combines a non-linear predictor with a spatial random effect and we show how to treat both of them under the Gaussian process framework. We propose also a pragmatic procedure to detect the most relevant environmental variables in prediction using average predictive comparisons. We conduct a fully Bayesian inference and evaluate our model in an interpolation and extrapolation tasks. The results show that the goodness of the model depends on the covariance function of the Gaussian process. The proposed methods are applied for a case study with the youngest stages of sea spawning whitefish (Coregonus lavaretus L. s.l.) larvae in the Gulf of Bothnia. Results on the case study show that in large scale the most important variables describing the potential larval areas are bottom type, prolonged ice period in spring, ecological status of coastal areas, distance to large shallow sand areas and water depth.

► We propose a Gaussian process classifier for a species distribution modeling.
► We use average predictive comparisons to highlight the most important predictors.
► The methods work reliably and provide theoretically justified uncertainty estimates.
► We apply the model to a case study with the youngest stages of sea spawning whitefish.
► Results highlight several factors that describe the potential larval areas.