Combining climatic and soil properties better predicts covers of Brazilian biomes

• Autor(a) principal: Arruda, Daniel M.
• Data de Publicação: 2017
• Outros Autores: Fernandes-Filho, Elpídio I., Solar, Ricardo R. C., Schaefer, Carlos E. G. R.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: https://doi.org/10.1007/s00114-017-1456-6; http://www.locus.ufv.br/handle/123456789/18385
• topic: Biogeography; Biome model; Ecological niche model; Map comparison; Plant functional types; Random forest

Several techniques have been used to model the area covered by biomes or species. However, most models allow little freedom of choice of response variables and are conditioned to the use of climate predictors. This major restriction of the models has generated distributions of low accuracy or inconsistent with the actual cover. Our objective was to characterize the environmental space of the most representative biomes of Brazil and predict their cover, using climate and soil-related predictors. As sample units, we used 500 cells of 100 km2 for ten biomes, derived from the official vegetation map of Brazil (IBGE 2004). With a total of 38 (climatic and soil-related) predictors, an a priori model was run with the random forest classifier. Each biome was calibrated with 75% of the samples. The final model was based on four climate and six soil-related predictors, the most important variables for the a priori model, without collinearity. The model reached a kappa value of 0.82, generating a highly consistent prediction with the actual cover of the country. We showed here that the richness of biomes should not be underestimated, and that in spite of the complex relationship, highly accurate modeling based on climatic and soil-related predictors is possible. These predictors are complementary, for covering different parts of the multidimensional niche. Thus, a single biome can cover a wide range of climatic space, versus a narrow range of soil types, so that its prediction is best adjusted by soil-related variables, or vice versa.