Predição da precipitação na Amazônia legal usando modelos aditivos generalizados de locação, escala e forma (GAMLSS)

Abstract

The variability of precipitation in the Legal Amazon, particularly in areas with a high incidence of zeros in time series, is a significant challenge for climate modeling and accurate forecasting. This study aimed to model precipitation in the Legal Amazon using the Generalized Additive Model for Location, Scale, and Shape (GAMLSS), focusing on the analysis of teleconnections, wind speed, atmospheric pressure, and extreme temperatures as explanatory variables. Precipitation and other climatic variables were collected and analyzed during the quarters of 2021. The GAMLSS model was employed to identify regional precipitation patterns and the influence of teleconnections, using the Zero Adjusted Gamma (ZAGA) distribution to handle the high incidence of zeros in the time series. The results indicate that the ZAGA distribution was particularly effective, achieving R² values higher than 0.75 in 132 out of the 408 pixels analyzed. The regions of Pará, Maranhão, eastern Amazonas, and northern Tocantins, along with specific areas in Acre, Rondônia, and Mato Grosso, stood out for the high performance of the model. Pará, in particular, accounted for 44% of the models with R² values above 0.90. The conclusion highlights the robustness of the ZAGA distribution in forecasting precipitation in regions with high climate variability, providing a solid basis for future research and conservation policies in the Legal Amazon.