Analysis of climate change is often computationally burdensome. Here, we present an approach for intelligently selecting a sample of climates from a population of 6800 climates designed to represent the full distribution of likely climate outcomes out to 2050 for the Zambeze River Valley. Philosophically, our approach draws upon information theory. Technically, our approach draws upon the numerical integration literature and recent applications of Gaussian quadrature sampling. In our approach, future climates in the Zambeze River Valley are summarized in 12 variables. Weighted Gaussian quadrature samples containing approximately 400 climates are then obtained using the information from these 12 variables. Specifically, the moments of the 12 summary variables in the samples, out to order three, are obliged to equal (or be close to) the moments of the population of 6800 climates. Runoff in the Zambeze River Valley is then estimated for 2026 to 2050 using the CliRun model for all 6800 climates. It is then straightforward to compare the properties of various subsamples. Based on a root of mean square error (RMSE) criteria, the Gaussian quadrature samples substantially outperform random samples of the same size in the prediction of annual average runoff from 2026 to 2050. Relative to random samples, Gaussian quadrature samples tend to perform best when climate change effects are stronger. We conclude that, when properly employed, Gaussian quadrature samples provide an efficient and tractable way to treat climate uncertainty in biophysical and economic models.
