Multiple linear regression to forecast decendial weather for agricultural purposes

Abstract

Predicting climate conditions helps in decision-making due to its great influence on crops, enabling more efficient production strategies and reducing damage, especially in the most critical phases of corn cultivation that determine its productive potential. A multiple linear regression model (RLM) was developed to forecast meteorological elements at least 2 months in advance for 15 locations that are prominent in corn production in Brazil. A set of daily data on average, minimum and maximum air temperature, wind speed, relative humidity and global radiation provided by the NASA/POWER system and precipitation data obtained from the National Water Agency (2003 to 2019) was used, organized into decennials (DEC) depending on the average corn cycle and grouped into two types of climate (Am and Aw). Forecasts using 14 DEC in both climate types showed, on average, high accuracy for all elements. The results indicated that the MLR has high accuracy in predicting these variables, especially in estimating wind speed. For the more volatile elements, such as precipitation, the model was still able to maintain acceptable reliability despite the complexities inherent in its prediction. In addition, the performance of the Camargo model to calculate the water balance was highlighted. This model required less input data to calculate water storage in the soil, which is decisive for planning corn cultivation, being an effective tool for predicting climatic elements on a ten-day scale.