Time series of normalized difference vegetation index (NDVI) data from satellite spectral measurements can be used to characterize and quantify changes in vegetation phenology and explore the role of natural and anthropogenic activities in causing those changes. Several programs and methods exist to process phenometric data from remotely-sensed imagery, including TIMESAT, which extracts seasonality parameters from time-series image data by fitting a smooth function to the series. This smoothing function, however, is dependent upon user-defined input parameter settings which have an unknown amount of influence in shaping the final phenometric estimates. To test this, a sensitivity analysis was conducted using MODIS maximum value composite NDVI time-series data acquired for Fort Riley, Kansas during the period 2001-2012. The phenometric data generated from the different input setting files were compared against that from a base scenario using Pearson and Lin’s Concordance Correlation Analyses. Findings show that small changes to parameter settings results in insignificant differences in phenometric estimates, with the exception of end of season data and growing season length.
Next, a time-series analysis of the same MODIS NDVI data for Fort Riley and nearby Konza Prairie Biological Station (KPBS) was conducted to determine if significant differences existed in selected vegetation phenometrics. Phenometrics of interest were estimated using TIMESAT and based on a Savitzky-Golay filter with parameter settings found optimal in the previous study. The phenometrics start of season, end of season, length of season, maximum value, and small seasonal integral were compared using Kolmogorov-Smirnov (K-S) and showed significant differences existed for all phenometrics in the comparison of Fort Riley training areas and KPBS, as well as low- versus high-training intensity areas within Fort Riley. Fort Riley and high-intensity training areas have earlier dates for the start and end of the growing season, shorter growing season lengths, lower maximum NDVI values, and lower small seasonal integrals compared to KPBS and low-intensity training areas, respectively. Evidence was found that establishes a link between military land uses and/or land management practices and observed phenometric differences.
