HR: 0830h
AN: H11F-0914    [PDF]
TI: Using Decision Trees to Examine Relationships between Inter-Annual Vegetation Variability, Topographic Attributes, and Climate Signals
AU: * White, A B
EM: abwhite@uiuc.edu
AF: University of Illinois, 205 N. Mathews Ave., Urbana, IL 61801 United States
AU: Kumar, P
EM: kumar1@uiuc.edu
AF: University of Illinois, 205 N. Mathews Ave., Urbana, IL 61801 United States
AB: The objective of this research is to develop KDD (knowledge discovery in databases) techniques for spatio-temporal geo-data, and use these techniques to examine inter-annual vegetation health signals. The underlying hypothesis of the research is that the signatures of inter-annual variability of climate on vegetation dynamics as represented by the statistical descriptors of vegetation index variations depend upon a variety of attributes related to the topography, hydrology, physiography, and climate. NDVI (normalized differential vegetation index) is enlisted to represent vegetation health and relationships between this index and topographic attributes such as elevation, slope, aspect, compound topographic index (CTI), and the proximity to a stream, are analyzed. Several scientific questions related to the identification and characterization of the inter-annual variability ensue as a consequence of our hypothesis. Investigations were performed using 13 years of 1-km resolution NDVI data from the AVHRR instrument on NOAA's POES (polar-orbiting operational environmental satellite) over the continental U.S. Various temporal change indices were used in order to identify anomalous inter-annual behavior in the NDVI index, including maximum absolute and relative deviations from the 13-year mean and positive and negative persistence indices (after Zhou et al., 2001). The KDD technique used in this research is the decision tree, which falls under the classification and prediction division of data mining techniques. The algorithm is similar to c4.5 and id3, but can handle continuous input and output values without binning and is optimized to determine the minimum error. Future work will incorporate clustering algorithms (both distance and density-based) and association rule algorithms (constraint-based) adapted for spatial-temporal data. Investigations will also be performed at smaller spatial scales, integrating higher resolution data. Throughout the growing season, elevation and slope are dominant factors associated with increased vegetation variability. From May to September slope prevails at high, rather than low elevations, although in the beginning and end of the growing season (April and October) this is not the case. This may possibly be due to the lack of vegetation at higher elevations at the fringes of the growing season. In general, the lower the slope, the greater the relative change in vegetation, thus linking zones of moisture convergence typically associated with low slopes to increased changes in vegetation over time. The relative change in vegetation is greater at mid-range elevations in April through June, high elevations in July through September, and low elevations in October. Zones of sub-surface flow convergence, as captured by the CTI, play an important role in July through October; however, the influence alternates from low elevations in July to high elevations in August, reverting back to low elevations in September and October.
DE: 1640 Remote sensing
DE: 1694 Instruments and techniques
DE: 1803 Anthropogenic effects
DE: 1851 Plant ecology
DE: 1894 Instruments and techniques
SC: Hydrology [H]
MN: 2003 Fall Meeting