Sugarcane yield estimation in São Paulo State - Brazil

This paper presents a method for crop yield forecast based on remote sensing and official data. The method uses a statistical approach to extract different pixels of smoothed NDVI data derived from MODIS sensor to be used as proxies for sugarcane yield estimation at a municipal scale. From 368 municipalities with yield's historical data from 2003 to 2012, three groups were created based on acreage percentile and then 30 municipalities was randomly selected, 10 for each group. Two municipalities with extreme acreage values (minimum and maximum) were discarded from each group and two different approaches were tested to normalize yield data and NDVI: Zscore and Rscore. In addition, two methods were used as selection criteria: RMSE and Pearson's correlation. Results showed that municipalities with large sugarcane acreage tended to present better agreement between observed and estimated yield, which reinforce the potential of this method to be operationally used for sugarcane yield forecast over large areas. Moreover, the proposed method may estimate yield early in crop season, whereas official statistics are usually published late after harvest.