A noise detection method for NDVI time series data based on dixon test

The normalized difference vegetation index (NDVI) time-series data, derived from satellite sensors, has been used to support land cover change detection and monitor crops successfully, but further applications are hindered by residual noise in the NDVI data. Methods for reducing noise and constructing high-quality NDVI time-series data sets can be broadly grouped into three general types, including threshold based methods, filter fitting method and curve fitting method. All of these methods are required to set up a certain number of empirical parameters and there is no objective criteria for the definition of noise. In this paper, suspicious noise points are detected from a statistical point of view, and dixon test are used to analyze NDVI data of many consecutive years, supplemented by the Quality Assurance (QA) analysis when judging the noise point. This approach reduces the dependence on prior knowledge, and the quantitative results of noise test can be provided for data correction and data reconstruction, so that preprocessing accuracy will be improved.