Time series harmonic regression analysis reveals seasonal vegetation productivity trends in semi-arid savannas

Abstract

Vegetation cover in savannas is characterized by high spatial heterogeneity driven by natural and anthropogenic drivers acting at multiple spatial and temporal scales. This research article identifies trends in vegetation cover and productivity in response to both land management and variable precipitation through a methodological approach that incorporates vegetation transect fieldwork with satellite imagery time series analysis. The phenological cycles of semi-arid savanna vegetation were analyzed over 27 years for both a protected area and a privately-owned property in order to capture multiple vegetation cover categories within the eastern Edwards Plateau ecoregion of central Texas. Line-intercept vegetation transects were established across these two sites in 2015, with ground cover and structural vegetation measurements collected along each transect. These measurements were classified based upon distinct vegetation cover categories: woodland, intermediate, open, and open disturbed. Time series analysis of vegetation productivity utilized 111 multi-season Landsat TM5 and Landsat 8 images that captured spring, fall, and winter seasonalities across the 27-year time period spanning from 1988 to 2015, with the Soil-Adjusted Vegetation Index (SAVI) calculated as an indicator of vegetative productivity. Harmonic regression was applied to this time series in order to determine the relationship between time of year with respect to growing season and productivity level across vegetation cover types. Fixed mean harmonic amplitudes for calendar year productivity were calculated across the entire time period as well as across three natural breaks-based sub-divisions of the 27-year period. Mean amplitudes per transect were then plotted against field-truthed tree-grass ratios in order to determine the relationship between tree and grass cover and mean harmonic SAVI amplitude. Harmonic regression identified inter- and intra-annual seasonal variability across cover types, with woodland sites exhibiting much lower variability in SAVI levels than open sites. These results provide insight into the phenological response of vegetation cover types typical of semi-arid savannas to spatially and temporally variable precipitation levels and land management schemes, allowing for better informed management and monitoring strategies for these environments.