The impacts of smoothing methods for time-series remote sensing data on crop phenology extraction

Abstract

Crop phenology is of critical importance to crop type classification, crop growth monitoring and yield prediction. Data smoothing is an inevitable procedure before extracting crop phenology from remote sensing data. This paper chose Guanzhong Plain in Shaanxi Province, China as the study area to investigate the impacts of the smoothing methods on the extraction of crop phenology. Results show that the double logistic function-fitting is better in describing the overall trend of crop dynamics while the Savitzky-Golay filter can retain more details in vegetation index time series. According to ground observation data, crop phenology retrieved from remote sensing data is not very consistent with the ground observations. However, as for the first growing season, crop phenology derived from the double logistic function-fitting smoothed data tended to closer to observations, and as for the second growing season, that from the Savitzky-Golay filter smoothed data provided better results.