Spatio-Temporal Processing of Time-Series of Satellite Data for Vegetation Phenology
Jönsson, Per1; Eklundh, Lars2
1Malmö University, SWEDEN; 2Lund University, SWEDEN
Large amounts of high temporal resolution data derived from satellites demand new and computationally efficient methods for extracting information. The TIMESAT software package was developed for processing time-series of satellite images. TIMESAT fits smooth mathematical functions (least-squares fitted asymmetric Gaussian, double logistic functions, or Savitzky-Golay filter) to time-series of satellite data in order to reduce the influence of noise. It then extracts phenological metrics (beginning and end of the growing season, length of the season, amplitude, integrated value, asymmetry of the season etc.) for each image pixel and growing season. The program fits functions to the upper envelope of the data in order to handle negatively biased noise. It also weights each observation in accordance with data quality labels, such as the MODIS QA flags. The package has been widely applied for data smoothing and extraction of land surface phenology during the last ten years. However, several challenges remain to be solved regarding how to best process the data and how to interpret the resulting parameters. One such challenge is to utilize the spatial as well as the temporal domain when smoothing data. Based on the assumption of some spatially uncorrelated noise a spatio-temporal smoother will be effective in generating stable measure of seasonality. We have implemented a fast and general method for data smoothing and seasonality extraction in TIMESAT that utilizes the spatial neighborhood of a pixel and that balances data smoothness with fidelity to the original time-series. The method is validated against in-situ data from spectral measurement towers, and is shown to be superior for describing the seasonal trajectory of vegetation index data in comparison to previously tested methods. We show examples of fitted data and extracted seasonality parameters for selected test regions.