Strategies to optimize modeling habitat suitability of Bertholletia excelsa in the Pan-Amazonia.

• Autor(a) principal: TOURNE, D. C. M.
• Data de Publicação: 2019
• Outros Autores: BALLESTER, M. V. R., JAMES, P. M. A., MARTORANO, L. G., GUEDES, M. C., THOMAS, E.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da EMBRAPA (Repository Open Access to Scientific Information from EMBRAPA - Alice)
• Texto Completo: http://www.alice.cnptia.embrapa.br/alice/handle/doc/1114461; https://doi.org/10.1002/ece3.5726
• topic: Expert knowledge; Maximum entropy; Model evaluation; Protected Amazonian species; Spatial filtering; Species distribution model; Conhecimento especializado; Entropia máxima; Avaliação de modelo; Análise de componentes principais; Filtragem espacial; Modelo de distribuição de espécie; Castanha; Principal component analysis

Aim: Amazon-nut (Bertholletia excelsa) is a hyperdominant and protected tree species, playing a keystone role in nutrient cycling and ecosystem service provision in Amazonia. Our main goal was to develop a robust habitat suitability model of Amazon-nut and to identify the most important predictor variables to support conservation and tree planting decisions. Localization: Amazon region, South America. Methods: We collected 3,325 unique Amazon-nut records and assembled >100 spatial predictor variables organized across climatic, edaphic, and geophysical categories. We compared suitability models using variables (a) selected through statistical techniques; (b) recommended by experts; and (c) integrating both approaches (a and b). We applied different spatial filtering scenarios to reduce overfitting. We additionally fine-tuned MAXENT settings to our data. The best model was selected through quantitative and qualitative assessments. Results: Principal component analysis based on expert recommendations was the most appropriate method for predictor selection. Elevation, coarse soil fragments, clay, slope, and annual potential evapotranspiration were the most important predictors. Their relative contribution to the best model amounted to 75%. Filtering of the presences within a radius of 10 km displayed lowest overfitting, a satisfactory omission rate and the most symmetric distribution curve. Our findings suggest that under current environmental conditions, suitable habitat for Amazon-nut is found across 2.3 million km2, that is, 32% of the Amazon Biome. Main conclusion: The combination of statistical techniques with expert knowledge improved the quality of our suitability model. Topographic and soil variables were the most important predictors. The combination of predictor variable selection, fine-tuning of model parameters and spatial filtering was critical for the construction of a reliable habitat suitability model.