High-resolution Average Forest Phenology and Annual Residuals for Quantifying the Start of Spring and Summer Leaf-area Dynamics

Land surface phenology (LSP) is the seasonal pattern of vegetation dynamics that occur each spring and fall. Multiple drivers of spatial variation in LSP and its variation over time have been analyzed using satellite remote sensing. Until recently, these observations have been restricted to moderate- and low-resolution data, as it is only at these spatial resolutions for which temporally continuous data is available. However, understanding small scale variation in LSP over space and time may be key to linking pattern to process, and in particular, could be used to understand how ecological processes at the stand level scale to landscapes and continents. Through utilization of the large, and now free, Landsat record, recent research has led to the development of robust methods for calculating average phenological patterns at 30-m resolution by stacking two decades worth of data by acquisition day of year (DOY). Here we have extended these techniques to calculate the deviation from the average LSP for any given acquisition DOY-year combination. We model the average LSP as two sigmoid functions, one increasing in spring and a second decreasing in fall, connected by a sloped line representing gradual summer leaf area changes (see Figure). Deviation from the average LSP is considered here to take two forms: (1) residual vegetation cover in mid- to late-summer represent locations in which disturbance, drought, or (alternatively) better than average growing conditions have resulted a separation (either negative or positive) from the average vegetation cover for that DOY, and (2) climate conditions that result in an earlier or later onset of greenness, exhibited as a separation from the average spring onset of greenness curve in the DOY direction (either early or late.) Our study system for this work is the deciduous forests of the mid-Atlantic, USA, where we show that late summer vegetation cover is tied to edaphic properties governing the site specific soil moisture balance. Additionally, we show that climatic factors (mostly related to topography) strongly influence the average start of spring. Annual deviations in the start of spring do not always scale linearly suggesting a spatially complex relationship between climate and the onset of spring. Model fit for a single pixel of mid-Atlantic deciduous forest. Shades of gray represent the weight each datum has on the model fit (increasing, white to black). Data weights account for variable atmospheric conditions between acquisitions.