de Oliveira Aparecido L.E.,São Paulo State University |
de Souza Rolim G.,São Paulo State University |
Camargo Lamparelli R.A.,The Interdisciplinary Center |
de Souza P.S.,São Paulo State University |
dos Santos E.R.,Guaxupe Ltda
Agronomy Journal | Year: 2017
Some forecasting techniques have been tested with crop models using various statistical analyses for generating future scenarios of yield (Y). Forecasting, however, can be achieved by simply using regression analysis and carefully selecting independent variables (IVs) with time displacement relative to the dependent variable. The early forecasting of Y is the vanguard of agronomic modeling, promoting improvements in planning, allowing more rational strategic decisions, and increasing food and economic security. Climatic variables are the most important factors controlling the yield and quality of coffee (Coffea arabica L.). We calibrated and tested agrometeorological models to forecast the annual Y of coffee for six traditional producing regions in the state of Minas Gerais, Brazil. We used multiple linear regressions, selecting IVs to maximize the period between the forecast of Y and the harvest for each locality. The IVs were monthly meteorological variables from 1997 to 2014: air temperature, rainfall, potential evapotranspiration, soil water storage, water deficit, and water surplus. The IVs were selected by testing all possible combinations in the domain and avoiding multicollinearity. The agrometeorological models were accurate for all regions, and the earliest forecasts were 6 and 5 mo before harvest for the producing locations of Guaxupé and Coromandel, respectively. The models for yield forecasting for Guaxupé included the water deficit in July and October and July precipitation for the high-yield season and the water deficit in April and September and October precipitation for the low-yield season. The models for yield forecasting for Coromandel included the November water surplus and February and September precipitation for the high-yield season and precipitation for January, April, and October for the low-yield season. © 2017 by the American Society of Agronomy.