Modelo estatístico de previsão de produtividade de soja e arroz para o Rio Grande do Sul

Abstract

In this PhD thesis, the usefulness of the insertion of climatic indicators in a statistical model for predicting rice and soybean yield in Rio Grande do Sul is presented. Initially, yield data of the two crops were separated into groups of homogeneous behavior in term of average yield. For soybean, the study period was from 1974 to 2013, excluding the 1983 crop because it was not included in the database. The northeast region of the State with the highest yield is highlighted, while the municipalities located in the northwest present series with lower average yield. In the rice crop, the study comprised the years 1990 to 2013 and the western and southern regions of the State show the highest average yield during the study period. In the municipalities of the central depression and near the Patos Lagoon the lowest average yield is observed. After this stage, lagged correlations were made between climatic indicators and a mean yield of each of the homogeneous groups in order to identify teleconnection patterns that influence the interannual variability of rice and soybean yield in the State. For soybean, the climatic indicators that presented the highest correlations were the Arctic Oscillation, North Atlantic Oscillation in addition to a region in the South Atlantic Ocean between 20°S/30°S and 20°W/40° W. Rice, in general, presented higher correlations than soybean. This fact highlighting mainly the indices referring to the oceanic and atmospheric components of the phenomenon El Niño Southern Oscillation and the index referring to Pacific Decadal Oscillation. To the highest correlation indexes with each homogeneous group, such as rice and soybean cultures, areas of Sea Surface Temperature with a high production correlation were added. Thereby a statistical regression model to crop forecast in Rio Grande do Sul may be elaborated. Through the Principal Component Regression method, the predictors for each group and culture were selected with the purpose of providing in October a yield estimate based on indicators obtained up to the month of September. October is when a major part of the soybean and rice are sown. The model shows good results, including as a support tool in the planning of rice and soybean harvest in the State of Rio Grande do Sul. As the advance in planting and crop development occurs the model can be updated with the inclusion of new index. It is also useful as a crop tracking tool and as an aid to eventual corrections of estimates that need to be made.