DEVELOPMENT OF REFINED SATELLITE LAND SURFACE PHENOLOGY DETECTION APPROACHES FOR ROBUST VEGETATION MONITORING

Abstract

Land surface phenology (LSP) from remote sensing data can serve as an integrative indicator of climate change impacts on terrestrial ecosystems. Accurate evaluations of changes in the number and timing of phenological events derived from LSP, such as the start of the growing season, are needed to support climate change mitigation and adaption. Yet, current LSP detection approaches are not fully capable of characterizing sub-annual phenological events across biomes. This dissertation aims to develop approaches that enhance robust detection of seasonal and sporadic phenological events from satellite vegetation indices (VIs) to support improved monitoring of vegetation signals. To address this goal, I conducted three studies focused on flux tower sites located across diverse biomes in the United States from 2003 to 2015. First, I evaluated the conditions under which the LSP detection algorithm developed for the Moderate Resolution Imaging Spectroradiometer (MODIS) robustly quantifies the inter-annual variability of growing season length derived from MODIS, Visible Infrared Imaging Radiometer Suite (VIIRS), and in-situ tower VI timeseries. Second, I developed a refined, adaptive LSP detection method and evaluated its ability to detect the start of the growing season across biomes. Algorithm performance was assessed by comparing signals derived from MODIS and VIIRS VIs to those from flux tower estimates of gross primary productivity. Finally, I applied this refined algorithm to detect the number and timing of phenological events across dryland sites with high intra- and inter-annual phenological variability. To identify a robust approach for quantifying such events in such dryland sites, I compared several VIs and phenological transition date detection approaches. Together, results suggest that the MODIS algorithm captures the inter-annual variability of major phenological events but often fails to detect sporadic events. The refined algorithm improved the detectability of irregular events commonly observed in dryland sites, especially when applied to the Enhanced Vegetation Index to quantify the number and peak of phenological events, and to water VIs to detect the start and end of events. The improved approaches to detect LSP developed here are expected to result in more accurate assessments of how climate change affects vegetation in regions with high current or projected future phenological variability.