Performance of Smoothing Methods for Reconstructing NDVI Time-Series and Estimating Vegetation Phenology from MODIS Data

摘要

Many time-series smoothing methods can be used for reducing noise and extracting plantphenological parameters from remotely-sensed data, but there is still no conclusive evidence infavor of one method over others. Here we use moderate-resolution imaging spectroradiometer(MODIS) derived normalized difference vegetation index (NDVI) to investigate five smoothingmethods: Savitzky-Golay fitting (SG), locally weighted regression scatterplot smoothing (LO), splinesmoothing (SP), asymmetric Gaussian function fitting (AG), and double logistic function fitting (DL).We use ground tower measured NDVI (10 sites) and gross primary productivity (GPP, 4 sites) toevaluate the smoothed satellite-derived NDVI time-series, and elevation data to evaluate phenologyparameters derived from smoothed NDVI. The results indicate that all smoothing methods can reducenoise and improve signal quality, but that no single method always performs better than others.Overall, the local filtering methods (SG and LO) can generate very accurate results if smoothingparameters are optimally calibrated. If local calibration cannot be performed, cross validation is away to automatically determine the smoothing parameter. However, this method may in some casesgenerate poor fits, and when calibration is not possible the function fitting methods (AG and DL)provide the most robust description of the seasonal dynamics.