Research Publications

The future of water storage dynamics in lakes around the world is increasingly uncertain due to a changing climate and evolving anthropogenic demands. Now, more than ever, it is necessary to quantify the uncertainty in forecasts of future water supply and water levels. However, classical ensemble forecasting systems that rely on weighted repetitions of historical water flux sequences, or independently resampled then post-processed sequences, often fail to capture either the dependence between hydrological variables across space and time or the full range of plausible outcomes for these variables. In this paper, an alternative statistical approach is outlined for using high-dimensional distributions described by copulas to generate water supply ensembles which preserve the correlation structure among hydrological variables. This new approach is outlined in detail to support dispersed operational modeling efforts, demonstrated by one implementation of a copula-based model for seasonal to annual forecasting on the Laurentian Great Lakes. This model implementation performs similar to or better than current operational forecasts while extending the forecast horizon from 6 to 12 months.