Date of this Version
Published in Seventh IEEE International Conference on Data Mining Workshops (ICDMW 2007), IEEE, pp. 667–672.
The integration of climate, satellite, ocean, and biophysical data holds considerable potential for enhancing our drought monitoring and prediction capabilities beyond the tools that currently exist. Improvements in meteorological observations and prediction methods, increased accuracy of seasonal forecasts using oceanic indicators, and advancements in satellite-based remote sensing have greatly enhanced our capability to monitor vegetation conditions and develop better drought early warning and knowledge-based decision support systems. In this paper, a new prediction tool called the Vegetation Outlook (VegOut) is presented. The VegOut integrates climate, oceanic, and satellite-based vegetation indicators and utilizes a regression tree data mining technique to identify historical patterns between drought intensity and vegetation conditions and predict future vegetation conditions based on these patterns at multiple time steps (2-, 4-, and 6-week outlooks). Cross-validation (withholding years) revealed that the seasonal VegOut models had relatively high prediction accuracy. Correlation coefficient (R2) values ranged from 0.94 to 0.98 for 2-week, 0.86 to 0.96 for 4-week, and 0.79 to 0.94 for 6-week predictions. The spatial patterns of predicted vegetation conditions also had relatively strong agreement with the observed patterns from satellite at each of the time steps evaluated.