Utilizing Satellite Fusion Methods to Assess Vegetation Phenology in a Semi-Arid Ecosystem

Dryland ecosystems cover over 40% of the Earth’s surface, and are highly heterogeneous systems dependent upon rainfall and temperature. Climate change and anthropogenic activities have caused considerable shifts in vegetation and fire regimes, leading to desertification, habitat loss, and the spread of invasive species. Modern public satellite imagery is unable to detect fine temporal and spatial changes that occur in drylands. These ecosystems can have rapid phenological changes, and the heterogeneity of the ground cover is unable to be identified at course pixel sizes (e.g. 250 m). We develop a system that uses data from multiple satellites to model finer data to detect phenology in a semi-arid ecosystem, a dryland ecosystem type. The first study in this thesis uses recent developments in readily available satellite imagery, coupled with new systems for large-scale data analysis. Google Earth Engine is used with the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) to create high resolution imagery from Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS). The 250 m daily MODIS data are downscaled using the 16-day, 30 m Landsat imagery resulting in daily, 30 m data. The downscaled images are used to observe vegetation phenology over the semi-arid region of the Morley Nelson Snake River Birds of Prey National Conservation Area in Southwestern Idaho, USA. We found the fused satellite imagery has a high accuracy, with R2 ranging from 0.73 to 0.99, when comparing fusion products to the true Landsat imagery. From these data, we observed the phenology of native and invasive vegetation, which can help scientists develop models and classifications of this ecosystem. The second study in this thesis builds upon the fused satellite imagery to understand pre-and post-fire vegetation response in the same ecosystem. We investigate the phenology of five areas that burned in 2012 by using the fusion imagery (daily) to derive the normalized difference vegetation index (NDVI, a measure of vegetation greenness) in areas dominated by grass (n=4) and shrub (n=1). The five areas also had a range of historical burns before 2012, and overall we investigated the phenology of these areas over a decade. This proof of concept resulted in observations of the relationship between the timing of fire and the vegetation greenness recovery. For example, we found that early and late season fires take the longest amount of time for vegetation greenness to recover, and that the number of historical fires has little impact in the vegetation greenness response if it has already burned once, and is a grass-dominated region. The greenness dynamics of the shrub-dominated study site provides insight into the potential to monitor post-fire invasion by nonnative grasses. Ultimately the systems developed in this thesis can be used to monitor semi-arid ecosystems over long-time periods at high spatial and temporal resolution