Produtividade da soja no Rio Grande do Sul estimada por meio de dados obtidos por satélites

Abstract

The great soybean expansion in Rio Grande do Sul State (RS), Brazil, demands updated information regarding the crop grain yield as well as the extent of meteorological damage occurred during the season. Mathematical models used to estimate grain yield based on crop response to meteorological factors are an important tool in obtaining this information. However, its use over large regions is hampered by the limited availability of meteorological data, due to the sparse distribution of in situ weather measurement stations over a large part of the national territory. Thus, this work aimed to test a model to monitor and estimate soybean grain yield in RS coupled in a Geographic Information System (GIS), based on spectral and meteorological data obtained mainly through orbital sensors. To estimate soybean grain yield from 2016 to 2018 crop seasons the model proposed by Doorenbos and Kassam (1979) was used with rainfall data from the Global Precipitation Measurement (GPM) system, relative air humidity from the AIRS sensor and air temperature estimated from surface temperature data from MODIS sensor, along with wind speed and insolation data obtained by weather stations. We first compared estimated meteorological data from GPM, AIRS and MODIS to those observed from weather stations. Then, due to the different edaphoclimatic characteristics, we subdivided the RS into north and south portions, so that the model was distinctly adjusted between these regions. Moreover, for the entire RS we used Leaf Area Index (LAI) data obtained from literature (Agrometeorological Model - MA). For the North portion, we also used LAI values estimated by the Normalized Difference Vegetation Index (NDVI) (Agrometeorological-Spectral Model - ME) and a soybean cropped map obtained from MODIS images for both MA and ME models. Modeled grain yield data were compared with the official ones provided by the Brazilian Institute for Geography and Statistics (IBGE), in micro-region level, using the Kolmogorov-Smirnov (KS) test for =5%. For the spatio-temporal yield monitoring, we used the Penalty Index (IP). The observed bias we observed for rainfall data from GPM (5.33 mm), relative air humidity data from AIRS sensor (from - 3.30 to 4.43%) and air temperature data from MODIS sensor (-0.3°C) was suitable to their use in the estimation model. When applied to the entire RS, no significant difference was observed between MA data and those provided by the IBGE. For the north portion, we observed significant difference between MA and ME data compared to those provided by IBGE only for the 2016 crop season. Regarding the IP, the north portion was barely penalized, with the exception of IPs from 0.7 to 0.8, mainly in the West region in 2018 crop season. The south portion was very penalized, with the south and southwest borders the most affected regions, with IPs of 0.1 and 0.2 in the 2018 crop season. Overall, the proposed model showed to be a useful aid tool to monitor and estimate soybean grain yield in RS.