Untangling Robust Decision Outcomes Given Correlations in Climatic and Economic Parameters

The resiliency of planned infrastructure and policies can turn on climatic thresholds. Robust decision frameworks, a growing approach to infrastructure planning, identify these thresholds by using exploratory modeling that simulates many combinations of uncertain climatic and economic parameters. Statistical clustering algorithms are used to tease apart the determinants of conditions that result in unacceptable performance. Often, however, uncertain variables are correlated, complicating the statistical analysis and casting doubt upon the significant variables identified. For example, increased temperature is associated with lower crop yields, which reduces the economic return per unit of water (net variable benefit, NVB). To evaluate the impact of these correlations in uncertainty parameters, we induced correlations between temperature and key climatic and economic parameters that drive exploratory modeling simulations used in robust decision making analysis. We tested four configurations of these uncertainty parameters, inducing correlations of 0%, 30%, 60%, and 90% between temperature and the absolute value of precipitation, coefficient of variation (COV), and surface downward solar radiation (rsds), and negative correlations between temperature and NVB and the discount rate. We used a calibrated simulation model of a dam cascade system centered on Lake Tana, Ethiopia to compute the agricultural supply of the reservoirs. We perform statistical clustering of the agricultural and economic performance of the reservoirs using the Patient Rule Induction Method to identify climatic and economic scenarios that cause failure. We then determine how well these economic and climatic parameters can be untangled, via scenario discovery using agricultural water supply and net present value as two indicators of reservoir performance. Thus, we identify key thresholds linked to the climatic and economic parameters in each correlation setup, allowing us to compare the parameters and thresholds identified in each setup and determine the reliability of robust analysis in an interconnected climatic and economic system. This analysis highlights the degree to which correlations in uncertain parameters can impact the robust decision framework results and suggests methods to avoid confounding results.