Estimativa de safra do algodoeiro com produtos de sensoriamento remoto

Abstract

Brazil has remained among the five largest cotton producers in the world. Due to high investments in cotton crops, the field monitoring on spatial and temporal scales is essential for forecasting harvests and planning the entire production chain. This work aimed to: develop an equation to estimate cotton productivity using only spectral vegetation indices (VI), validating it with field production values; and to assess the estimate of biomass by the Monteith Method using, in addition to the vegetation index, evapotranspiration images. For this purpose, production field data from cotton grown in the Cerrado of Bahia State were used to validate the estimates. Images from Sentinel-2A and 2B satellites were processed in the Google Earth Engine platform to determine the average and maximum values of the NDVI, SAVI and EVI vegetation indices, in the cycle of each cultivated plot, which were used in the empirical modeling. Landsat 8 satellite products (NDVI and Evapotranspiration) were obtained from the GEESebal application for implementation in the Monteith Model. Estimates were compared with field data, using Tukey's Test, and evaluated according to the following indicators: Mean Absolute Error (MAE), Mean Ratio (MR); Root Mean Square Error (RQME) coefficients of determination (R²), Pearson's correlation coefficient (r) and Willmott performance index (d). The estimates with mean and maximum IVs of the crop cycles showed a median performance: means of estimated and observed values were statistically equal, but R² < 0.5 and RQME around 500 kg ha-1. The models derived with average EVI and with maximum SAVI presented the best performance indices. Despite a good correlation with the productivity measured in the field, the estimate using the Monteith method showed an overestimation of about 50%, which may be related to the low image coverage in the cycle of the plots studied.