Uma abordagem híbrida dinâmico-estatística para a previsão climática do nordeste do Brasil.

Abstract

Climate forecasting is gaining more and more prominence in planning activities related to different economic sectors, such as agriculture, energy, water resources, transport and tourism. There are different climate models and different techniques to convey this information to society, which require variations to enable decision-making. However, despite advances, it is still necessary to research methodologies that can improve and increase the reliability of specific climate variation in models. Therefore, the main objective of this research was to apply a post-processing technique to the outputs of the climate models that make up the North-American MultiModel Ensemble (NMME) project. The study area is the Northeast region of Brazil (NEB), and the technique used to calibrate the variations was the Canonical Correlation Analysis (CCA). The CCA was used to calibrate the monetary variations of the accumulated variations for the NEB, using the model forecast as the predictor variable and the observed occurrences as the predictor, comparing a hindcast of 30 years of forecast with observations from the same period, from 1991 to 2020. The results show that the direct outputs of NMME model corrections present significant views when comparing model and observed climatology, and that after flexibility with ACC, these views are corrected and correlations between observations and observations increase significantly, in addition. there will be a reduction in variation errors. After this step, a lag of up to three months of forecasting for a given month was provided, which models presented the best results, in order to combine them into a limited set of models to generate more specific efficiency for the NEB. Such ensemble media (ensembles) showed an even greater demonstration than that of individual models even after post-processing with CCA.