Characterizing phenological vegetation dynamics amidst extreme climate variability in Australia with MODIS VI data

Australia's climate is extremely variable with inter-annual rainfall at any given site varying by 5- or 6-fold or more, across the continent. In addition to such inter-annual variability, there can be significant intra-annual variability, especially in monsoonal Australia (e.g. the wet tropical savannas) and Mediterranean climates in SW Australia where prolonged dry seasons occur each year. This presents unique challenges to the characterization of seasonal dynamics with satellite datasets. In contrast to annual reoccurring temperature-driven phenology of northern hemisphere mid-latitudes, vegetation dynamics of the vast and dry Australian interior are poorly quantified by existing remote sensing products. For example, in the current global-based MODIS phenology product, central Australia is covered by ~30% fill values for any given year. Two challenges are specific to Australian landscapes: first, the difficulty of characterizing seasonality of rainfall-driven ecosystems in interior Australia where duration and magnitude of green-up and brown down cycles show high inter annual variability; second, modeling two phenologic layers, the trees and the grass in savannas were the trees are evergreen but the herbaceous understory varies with rainfall. Savannas cover >50% of Australia. Australia's vegetation and climate are different from other continents. A MODIS phenology product capable of characterizing vegetation dynamics across the continent is being developed in this research as part of the AusCover national expert network aiming to provide Australian biophysical remote sensing data time-series and continental-scale map products. These products aim to support the Terrestrial Ecosystem Research Network (TERN) serving ecosystem research in Australia. The MODIS land surface product for Australia first searches the entire time series of each Climate Modeling Grid pixel for low-high-low extreme point sequences. A double logistic function is then fit to each of these sequences allowing identification of growth periods with different magnitudes and durations anywhere in the time series. Results show that the highest absolute variability in peak greenness occurred in cropped areas while the highest relative variability (coefficient of variation) occurred in interior Australia particularly around Lake Eyre, the center of a closed drainage basin in the dry interior of the continent. Across the desert interior, the timing of the green-up onset and the peak greenness was correlated with the landfall of cyclones and the inland penetration and strength of the north Australian summer monsoon (represented by TRMM data). The variability of Australian land surface phenology magnitude and timing was found to be strongly correlated with the swings between La Nina and El Nino events. The information on vegetation dynamics represented here is critical for land surface, fuel accumulation, agricultural production, and permanent ecosystem change modeling in relation to climate trends. A unique research opportunity is provided by recent climate variability: in 2010 a persistent El Nino has given way to a strong two-year La Nina breaking a decade long drought that was followed by record-breaking rainfall across most of the continent and extensive flooding followed by sustained greening.