HR: 1340h
AN: B43B-0273    [Abstracts]
TI: Using a curve-fitting methodology on remotely sensed time series to detect subtle patterns of land surface phenology
AU: * Bradley, B
EM: bethany_bradley@brown.edu
AF: Brown University, Department of Geological Sciences Box 1846, Providence, RI 02912 United States
AU: Mustard, J
EM: john_mustard@brown.edu
AF: Brown University, Department of Geological Sciences Box 1846, Providence, RI 02912 United States
AU: Jacob, R
EM: robert_jacob@brown.edu
AF: Brown University, Department of Geological Sciences Box 1846, Providence, RI 02912 United States
AU: Hermance, J
EM: john_hermance@brown.edu
AF: Brown University, Department of Geological Sciences Box 1846, Providence, RI 02912 United States
AB: Annual, inter-annual, and long-term trends in Land Surface Phenology (LSP) using NDVI time series from AVHRR and MODIS can be used to distinguish between natural ecosystem dynamics and land cover change. However, the full potential of long-term NDVI time series is often hampered by poor quality data caused by instrumentation problems, atmospheric conditions (e.g. clouds or haze), ground conditions (e.g. snow), and inter-annual variability of land cover. These effects make LSP difficult to identify, and may mask subtle shifts in inter-annual ecosystem response resulting from land use or other anthropogenic forcing. In order to maximize LSP detection, we use a curve fitting methodology useful for long-term time series across a range of phenologies. This approach is minimally affected by sensor error, clouds, and snow, and requires neither spatial nor temporal averaging to reduce noise. This methodology employs a spline-based curve, which is fit iteratively so that positive residuals are upweighted to capture the upper envelope of NDVI values. Here, we apply the curve fitting methodology to weekly AVHRR NDVI data (1990-2000) and biweekly MODIS NDVI data (2000-2005) at 1 km pixel resolution for the Great Basin desert of the western U.S. The spatial and temporal patterns of known ecosystems may then be assessed in order to identify anomalous trends in regional LSP. We compare both spatial and temporal variability of four known ecosystem types surveyed in 2004: sagebrush steppe, cheatgrass grassland, pinyon-juniper woodland, and montane shrubland. Average onset of greenness (using a timing of half max technique) occurred on Apr 14 (+/- 6 days; sagebrush), Apr 9 (+/- 8 days; cheatgrass), Apr 17 (+/- 8 days; pinyon-juniper) and May 24 (+/- 5 days; montane). The small standard deviation observed in similar ecosystems distributed throughout the Great Basin indicates that the phenologies are spatially robust in any individual year. However, there is considerable temporal variability within a time series. Average onset of greenness between 1990-2000 occurred at day Apr 15 (+/- 35 days; sagebrush), Apr 8 (+/- 20 days; cheatgrass), Apr 20 (+/- 29 days; pinyon-juniper), and May 21 (+/- 17 days; montane) for typical pixels from the four ecosystems. The higher standard deviations in temporal averages than spatial averages result from inter-annual variability in LSP. Temporal variability can also be observed at the regional scale and may be related to climate patterns. In 1996, after a dry winter, average regional onset of greenness occurred on March 31. In this year, many desert shrublands had such a low phenological response that NDVI amplitude was less than 0.05. In 1998, after a particularly wet winter, average regional onset of greenness occurred on May 2. In this year, peak NDVI was higher, and phenologies were shifted to later in the season. This curve fitting methodology increases our ability to measure inter-annual trends in LSP. Great Basin phenology as a whole is strongly influenced by regional weather patterns, but in any given year ecosystems respond similarly. These types of analyses will lead to better understanding of local and regional land cover trends.
DE: 0439 Ecosystems, structure and dynamics (4815)
DE: 0466 Modeling
DE: 0476 Plant ecology (1851)
DE: 0480 Remote sensing
SC: Biogeosciences [B]
MN: Fall Meeting 2005