Scenario-based vegetation outlook (S-VegOut): Predicting general vegetation condition using different scenarios over the central U.S

Predicting vegetation conditions over large geographic areas is imperative for a wide range of applications such as crop and rangeland condition assessments, drought monitoring, fire risk potential, and ecological studies. However, such predictions are very challenging given the complexity of climate-vegetation interactions and diversity of land use practices across large geographic areas. Improvements in our predictive capabilities in this area are becoming possible with the increasing availability of many high quality, environmental data sets (e.g., climate, ocean, and remote sensing observations), the longer historical records (i.e., > 20 years) of these data sets, and the emergence of statistical data mining techniques.

Time-series, satellite-based vegetation index (VI) observations from moderate to coarse-resolution global Earth imagers, such as the Advanced Very High Resolution Radiometer (AVHRR) and the more recent Moderate Resolution Imaging Spectroradiometer (MODIS), have been widely used to monitor vegetation conditions at regional to global scales. Several studies have firmly established our capability to monitor ‘current' vegetation conditions across the landscape using remotely-sensed VI data from satellite-borne sensors. However, few efforts have focused on forecasting or predicting ‘future' vegetation conditions. In this study, a new predictive tool called the Scenario-based Vegetation Outlook (S-VegOut), which is being developed at the National Drought Mitigation Center (NDMC), is presented and tested for 2009 growing season.

Scenario-VegOut predicts seasonal greenness for different climatic episodes (i.e. dry, normal, and wet conditions) using regression tree-based modeling that integrates historical vegetation condition information from satellite-based VI observations with climate-based drought indicators, oceanic indices, and general biophysical information about the environment (e.g., land use/land cover type and soil characteristics). Three hypothetical scenarios [i.e., drier than normal precipitation (< 75% of average precipitation received), normal (between 75 and 125% of average), and above normal (>125% of average) precipitation] are used to generate 1-kilometer (km) resolution vegetation condition outlook maps at multiple time-steps into the future (ranging from 2 to 6 weeks). This experimental tool predicts general vegetation conditions based on the analysis of historical patterns among the satellite, climate, and oceanic data over a 20-year period (1989 to 2008) by applying the different scenario-based regression tree models to each 1-km grid cell for a series of bi-weekly periods (14-day intervals) across the growing season.

In this study, we present the method and predictive ability of Scenario-VegOut maps for 2-, 4-, and 6-week vegetation condition outlooks across the 2009 growing season for a 15-state region of the central United States. Statistical analysis of the accuracy of the models and evaluation of the S-VegOut maps including the initial feedback from potential stakeholders will also be discussed.