HR: 0800h
AN: B41A-0094    [Abstracts]
TI: Distinguishing Inter-annual Phenological Variability from Long-term Change in the Great Basin Using a new Method of Time-Series Modeling
AU: * Bradley, B
EM: bethany\_bradley\@brown.edu
AF: Brown University, Dept. of Geology, Providence, RI 02912 United States
AU: Jacob, R
EM: robert\_jacob\@brown.edu
AF: Brown University, Dept. of Geology, Providence, RI 02912 United States
AU: Hermance, J
EM: john\_hermance\@brown.edu
AF: Brown University, Dept. of Geology, Providence, RI 02912 United States
AU: Mustard, J
EM: john\_mustard\@brown.edu
AF: Brown University, Dept. of Geology, Providence, RI 02912 United States
AB: Semi-arid vegetation communities are affected over time by a host of factors, including climate change, inter-annual climate variability, changes in hydrologic cycle, and land use-land cover change. Because vegetation exhibits inter-annual variability, it is important to determine characteristic inter-annual responses of semi-arid vegetation communities to distinguish variability from change. We explore the phenologies of five major ecosystem types located in central and northern Nevada: dry desert shrub, sagebrush steppe, annual grassland, pinyon-juniper woodland, and montane perennial grassland. By identifying the phenological characteristics of these major Great Basin land cover classes, we constrain the range of expected inter-annual and decadal variability. Time series of known vegetation types are analyzed by applying a new methodology to characterize inter-annual vegetation phenology using time series of 1 km Normalized Difference Vegetation Index (NDVI) from the Advanced Very High Resolution Radiometer (AVHRR) from 1990-2001. We characterize vegetation community response on both inter-annual and decadal time scales using a curve-fit algorithm that is flexible enough to address change over time without being influenced by error caused by sensor drift, clouds, snow, or missing data. Our methodology uses least mean squares to fit weekly and biweekly NDVI data simultaneously to a low-order baseline polynomial, which models both long-term change and mean inter-annual periodicity, and a high order annual polynomial, which allows for flexible phenologies between years. Long-term sensor drift is accounted for by removing a fit to mean NDVI from non-vegetated surfaces salt flats in Nevada and Utah prior to analysis. Anomalously low data caused by clouds or snow are removed by identifying the standard deviation of the curve-fit during the stable fall months and removing all points within one standard deviation of zero (all negative values are also excluded). Missing points removed due to clouds and snow or absent in 1994 are replaced with estimated values from the low-order baseline polynomial value at that point. Thus the model relies mainly on real data, but is stabilized by estimated data, particularly during winter and early spring when clouds and snow are common. The resulting robust, continuous model of NDVI can be used to detect subtle shifts in inter-annual vegetation response that might otherwise be masked by uncertainty in the NDVI time series. Initial analyses highlight three distinct types of change in the Great Basin linked to pinyon-juniper woodland, cheatgrass dominated grasslands and riparian marshlands. Dry desert shrub, sagebrush steppe, montane perennial grassland and pinyon-juniper do not show high inter-annual variability except in cases where green-up is offset by early spring snow. As a result, long-term change is more likely to be detectable. For example, some pinyon-juniper woodlands may show increasing greenness linked to woody expansion into neighboring ecosystems. Annual grasslands dominated by cheatgrass show a high degree of inter-annual variability in response to rainfall, making long-term change in this community difficult to identify. An example of change detected in a minor ecosystem type is observed in marshlands east of Nevada's Ruby Mountains. Here, an increasing greenness trend above the normal range of inter-annual variability may be linked to changes in water management and land use. With the combination of field localities of known land cover and a robust curve-fit NDVI time series, phenological variability can be distinguished from long-term change.
DE: 1851 Plant ecology
DE: 1640 Remote sensing
DE: 0400 Biogeosciences
SC: Biogeosciences [B]
MN: 2004 AGU Fall Meeting