The Vegetation Outlook (VegOut): A New Tool for Providing Outlooks of General Vegetation Conditions Using Data Mining Techniques

Abstract

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 (R 2) 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.

A review of drought indices

“…Sperling Drought Index (SDI) H Droughtscore.com (2007) Easily understandable measure (dry < 100 < wet); uses long-term precipitation patterns, groundwater, and reservoir levels, and the Palmer drought indices for drought characterization in longer time windows. Vegetation Outlook (VegOut) Agre Tadesse and Wardlow (2007) Combines climate information and RS observations of current vegetation conditions with oceanic index data and environmental biophysical information such as land cover type, irrigation status, soils, and ecological setting to provide a future outlook of general vegetation conditions. Ratio Vegetation Index (RVI) RS Pearson and Miller (1972) RVI ¼ NIR…”

Development and evaluation of a comprehensive drought index

“…The CART analysis is a tree-building technique, which constructs a set of decision rules to build predictive models. This technique was used by Tadesse and Wardlow (2007) The regression technique estimates the linear and nonlinear behavior between the dependent and independent variables. This technique was used by Keyantash and Dracup (2004) to develop an Aggregate Drought Index (ADI) that considers meteorological, hydrological, and agricultural categories of drought.…”

“…Therefore, it was suggested that drought monitoring techniques can be improved by combining the existing indices to better capture the overall impacts of drought (Zargar et al, 2011). In general, the methods used for combining drought indices can be classified as: 1) decision matrix analysis (Svoboda et al, 2002;Balint et al, 2011;Zieses et al, 2014); 2) classification and regression tree (CART) analysis (Tadesse and Wardlow, 2007;Brown et al, 2008); and 3) regression technique (Keyantash and Dracup, 2004;Karamouz et al, 2009).…”