An Adaptive Noise Reduction Method for NDVI Time Series Data Based on S–G Filtering and Wavelet Analysis

Abstract

In order to reduce the noise in advanced very high-resolution radiometer global inventory modeling and mapping studies normalized difference vegetation index (NDVI) version 3g time series data, we propose an adaptive noise reduction method based on the Savitzky–Golay filter and wavelet analysis and on curve-fitting and spectrum analysis. The noise reduction effect was analyzed and evaluated. Studies have shown that the denoising of data using asymmetric Gaussian and double logistic curve-fitting methods results in the loss of details of the NDVI changes and is not conducive to the extraction of vegetation phenotypic characteristics. The wavelet function, wavelet decomposition layer number, and the threshold determination method have a large influence on the noise reduction effect, and the adaptive method has high noise reduction efficiency and effectively reduces the noise in the NDVI data. The proposed method does not require the determination of the smoothing window size and threshold for each year, which represents an advantage for processing large amounts of data.