Evaluating effects of rescaling and weighting data on habitat suitability modeling

Abstract

Abundance index (AI), used to establish the suitability index (SI), provides critical information in habitat suitability index (HSI) modeling. The distributions of AIs derived from fisheries-independent surveys tend to be right skewed because of heterogenous distributions of fishes. The existence of large AI values and failure to consider it might result in underestimation of HSI values for most sampling areas. We compared the performance of HSI models based on original AIs (without any transformation) versus rescaled AIs (i.e., log-transformed AIs) using American lobster (Homarus americanus) along the coast of Gulf of Maine as an example. Impacts of weighting environmental variables on HSI modeling based on boosted regression tree (BRT) were also evaluated. Both cross-validation and predicted habitat suitability maps suggested that the weighted HSI model based on log-scaled AI data tended to yield a more reliable prediction of optimal habitats for American lobster. The unweighted HSI model based on the original AI data, however, tended to underestimate optimal habitats and overestimate suboptimal habitats. We recommend using log-transformed AIs and determining the weights of different environmental variables based on the BRT method in HSI modeling, especially when AI data are highly skewed.