Estimation of Coffee Yield from Gridded Weather Data

Abstract

Core Ideas Use of gridded weather data [European Center for Medium‐Range Weather Forecasts (ECMWF) and NASA] gave similar estimations of coffee yield to the models that use surface‐observed weather data. A new calibration was proposed to the model to the penalties and the coefficients calibrated from the different data sources. The calibration of the coefficients for the ECMWF and NASA data was efficient. The tests with independent data confirmed the improvement of the calibration. Coffee yield can be estimated using the ECMWF and NASA grid data. Agrometeorological models have been applied to crop yield estimations, providing information for management. A limiting factor of these models is the input data, which come mostly from surface meteorological stations (SMS). We propose the use of gridded weather data provided by the European Center for Medium‐Range Weather Forecasts (ECMWF) and the NASA as a viable and innovative alternative for coffee (Coffea arabica L.) yield estimation in Brazil. We made modifications in the coefficients of the model proposed by Santos and Camargo, regarding the penalties for extreme temperatures and those related to different gridded data sources. The accuracy, measured by the mean absolute percentage error (MAPE), was 23.76, 24.61, and 22% for the calibrations with ECMWF, NASA, and SMS data, respectively. These high levels of MAPE were the result of the high biennality of the crop. The average tendency, measured by the systematic root mean square error, was an overestimation (or underestimation) of ±465 kg ha−1 by the ECMWF, ±411 kg ha−1 by NASA, and ±653 by the SMS models. The mean precision, measured by the nonsystematic root mean square error, was 186, 190, and 280 kg ha−1 for the ECMWF, NASA, and SMS models, respectively. These results indicate that coffee yield for São Paulo and Minas Gerais can be calibrated and estimated with the ECMWF and NASA gridded data.