(Fig. 4). This result indicated that the temporal information of MODIS and the spatial information of Landsat ETM+ data were effectively integrated in the predicted NDVI dataset which could describe a more detailed spatial variation of NDVI at the resolution of 30 m. From the scatter plot of predicted and Landsat ETM+ NDVI data (Fig. 5), it could be seen that most of the scatter points were concentrated along the line of x = y. The determination coefficient(R2) was 0.66 at the 0.05 significant level, and the Root Mean Square Error (RMSE) was 397.64 which meant approximately 4% estimation error of predicted NDVI data (the NDVI data was scaled to 0to 10,000). Additionally, predicted NDVI data was slightly larger than Landsat ETM+ NDVI data, which might be responsible for the 5 date interval of data acquisition. Based on the comparison of predicted NDVI and Landsat ETM+ NDVI, it was indicated that STARFM in this study could effectively fuse the MODIS and Landsat ETM+NDVI data, and the fused NDVI dataset could be used for improving forest cover classification accuracy of Landsat ETM+ data.