Monitoring Brazilian soybean production using NOAA/AVHRR based vegetation condition indices

This study explores the potential application of NOAA/AVHRR based satellite indices to estimate the soybean yield for Brazil. NOAA AVHRR GVI (Global Vegetation Index) weekly maximum composite NDVI (Normalized Difference Vegetation Index) data sets with a resolution of 16 km for the period of 1985 to 1998 (except 1995 due to missing data) provided by NOAA/NESDIS were used in this study. Nine soybean yield models, including eight principal production states and the country, were constructed using observed yield data and NDVI and/or TCI (Temperature Condition Index) data from 1985 to 1995. The data period of 1996 to 1998 was used to evaluate the model performance. The crop yield is generally affected by technological improvements through time and by the annual weather fluctuation. The contribution of technological improvements was approximated by trend term and weather-related fluctuations of yield around the trend were estimated through AVHRR-based indices. The results showed that four states had significant technological trend contribution with slopes ranging from 0.49 to 0.86 and R2 ranging from 0.22 to 0.62. In most of the models, yield variation around the trend was sensitive to TCI (Temperature Condition Index), during the period of the grain filling stage (end of January for northern states and mid February for the southern states). For most of the models, the determination coefficient (R²) was higher than 0.6 and the root mean square error (RMSE) was lower than 10%. The results of model validation showed that the absolute errors were lower than 10% in 21 out of 27 cases tested. It is concluded that the satellite indices are useful for crop production monitoring. In the Brazilian soybean production region, where the summer crop season coincides with the rainy season, the temperature-based index is more informative about possible fluctuation of soybean yield and production in Brazil. It is suggested that a combination use of satellite and in situ data may likely improve the yield estimate.