Abstract:

【OBJECTIVE】It is important to get the accurate information of crop spatial distribution and areas for yield assessment and crop pattern optimization.【METHOD】In this research, time-series MODIS-NDVI was used to improve the accuracy of winter wheat classification. MODIS reflectivity data was used to calculate NDVI firstly. The Savitzky-Golay smooth filter has been developed to reduce the noise and enable the reconstruction of high quality MODIS-NDVI time-series. Based on the reconstructed MODIS-NDVI, as well as crop phenology, a simple but reasonable model was built to distinguish winter wheat classification in Hebei province.【RESULTS】The result showed that the Savitzky-Golay smooth filter model can eliminate efficiently the cloud contaminated data and abnormal data in time-series VI data, and the reconstruct data can capture the crop time profile curve better. There was a notable difference for the classification accuracy between counties. The lowest was 85.9% for Handan county, and the highest was 99% for Cangzhou county. The total accuracy of winter wheat for Hebei province was over 95%. The distribution of winter wheat in Hebei province in 2004 has the distinct features of zonality, it distributed mainly over southeastern Handan, Xintai, central Shijiazhuang and eastern Baoding.【CONCLUSION】It was possible to extract planting spatial distribution information of winter wheat by using time-series MODIS-NDVI data for large areas. The accuracy also showed the results were high in reliability especially in plain areas.