Improving Predictive Drought Models with Sensitivity Analysis

Drought is a common consequence of climate variability in the south-central U.S., but they are expected to occur more often and become more intense with climate change. Natural resource managers can improve their planning efforts with advance warnings of impending drought. Using input from resource managers in the Chickasaw Nation, this research team previously created models that forecast droughts up to 18 months in advance with information about their expected timing and intensity. Developed for all climate divisions in Louisiana, New Mexico, Oklahoma, and Texas, these drought models rely on input from predictor variables associated with global weather patterns like El Niño and La Niña. However, it is unclear how uncertainty associated with the predictor variables might affect the model’s drought forecasts. 

The purpose of this project is to conduct a sensitivity analysis of the predictor variables used in the drought models, specifically in the existing model for Oklahoma Climate Division 8, wherein is much of the Chickasaw Nation. Sensitivity analysis works by changing a single input variable in the model while holding the others constant, providing an unbiased estimate of how changes to that single variable changes model output. 

The sensitivity analyses performed in this project will allow researchers to quantify uncertainty in their drought models and to refine the models to make better predictions of future drought. Accurate forecasts of droughts with extended lead time are valuable for water managers to plan for and address issues related to water stress.