LAND COVER CLASSIFICATION BASED ON DAILY NORMALIZED DIFFERENCE VEGETATION INDEX TIME SERIES FROM MULTITEMPORAL REMOTELY SENSED DATA

Land use and land cover information is a critical parameter in various research fields affecting regional ecology and environment. This study explored the application of daily normalized difference vegetation index (NDVI) time series on land cover classification in an irrigation area. The study area is located in Wugong and Fufeng of the Baojixia irrigation area. First, the Savitzky-Golay (S-G) and harmonic analysis of time series (HANTS) filtering methods were employed in unequal-interval NDVI time series derived from the Huan Jing-1 A/B satellite. Second, daily NDVI time series data were acquired through spatial interpolation and the data was classified using a support vector machine (SVM). Finally, the parameters of the SVM classifier were optimized using a genetic algorithm (GA), which was used in the land cover classification based on daily NDVI time series. The results revealed that the S-G filtering was superior to HANTS for filtering the unequal-interval NDVI time series. Processing the daily NDVI time series through spatial interpolation substantially improved the land use classification accuracy contributing to an increase in phenological information of daily NDVI time series. Finally, the optimal parameters of the SVM classifier were determined using the GA as gamma = 0.0933 and C = 49.5643, and the overall accuracy and kappa coefficient of the classification results based on daily NDVI time series using GA-SVM were 98.24% and 0.9789, respectively, which demonstrates that GASVM had a substantial effect on land cover classification. These results could provide an empirical basis for further application of daily NDVI time series on land cover classification in irrigation areas.