Previsão histórica das safras de culturas de ciclo curto do Estado de Alagoas com modelos do CMIP6

Abstract

Rainfall variability is one of the most important factors in the success or failure of agricultural crops grown under rainfed conditions. This study sought to understand the relationship between accumulated rainfall in the wettest semester and the predictability of bean, corn and cassava harvests in the three mesoregions of the state of Alagoas. Due to the small territorial extension of the state when compared to other Brazilian states, and based on scientific literature, it was found that the main growing season for these crops is between the months of April and September in the three mesoregions: sertão, agreste and leste. Production data provided by the IBGE for each crop between 1974 and 2014 was used to calibrate a forecasting system based on the relationship between predictor: accumulated rainfall in the AMJJAS semester, and predictor: agricultural production, using the multivariate statistical method known as canonical correlation analysis (CCA). This technique was used in two ways, firstly as a simple method for purely statistical forecasting, and secondly as a bias correction method for calibrating the rainfall estimated by eight CMIP6 models for the Brazilian Northeast (NEB). As well as the observed rainfall data in the NEB, rainfall obtained from the ACCESS-CM2 (Australia), BCC CSM2-MR (China), CNRM-CM6-1-HR (France), GFDL-ESM4 (United States), MIROC6 (Japan), NORESM2-MM (Norway), INM-CM5-0 (Russia) and TaiESM1 (Taiwan) models in the AMJJAS semester was used with and without bias correction to predict crop production. The rainfall from the models shows strong biases in relation to the observed rainfall, but they were consistently corrected with ACC, considerably reducing the bias, the errors and increasing the correlation in relation to the observed rainfall. However, crop simulations were only slightly more efficient using the corrected rainfall from the models. Finally, by applying the ensemble concept, the best simulation of crop production came from the average of the simulations obtained from the models' original rainfall, without bias correction with ACC, which became the most important result of this research.