A Comparison and Evaluation of Performances among Crop Yield Forecasting Models Based on Remote Sensing: Results from the Geoland Observatory of Food Monitoring

In the context of the GEOLAND EC FP6 project the comparison of different remote sensing based approaches for yield forecasting over large areas in Europe are tested and results inter-compared. In particular the methods tested include the ones in use within the MARS-Crop Yield Forecasting System as the results from the Crop Growth Monitoring System model and vegetation indicators derived from Low Resolution SPOT-VGT and NOAA Images, METEOSAT based yield forecasting and ERS-Scatterometer Crop Performance Index. Performances of the different models were tested in Spain, Belgium and Poland. The inter-comparisons of the crop yield forecasts were mainly based on the forecasting error obtained from the different approaches based on the Root Mean Square Forecast Error (RMSFE). This error was derived by comparing the predicted yields of the different models with the official yield as from official statistics (EUROSTAT). The comparison of the RMSFE was used to verify the convergence of results from the different models, the reliability of the information, i.e. precision and bias, and its precocity compared to the crop cycle. The results showed that the indicators are able to give reliable information with some differences: remote sensing indicators are more precise and accurate in southern areas (less cloud cover) while in northern areas good results are obtained under the use of better local calibrations of traditional crop yield forecasting systems and/or the use of additional information for instance remote sensing data as inputs into advanced crop modelling systems. Furthermore, in order to take care of the different time series length available, a qualitative indicator called Performance Score (Ps) was introduced. The analysis of the Ps showed that when a long time series of observation is available greater advantages are obtained from RS rather than from more advanced crop models.