Extreme-value analysis for the characterization of extremes in water resources: A generalized workflow and case study on New Mexico monsoon precipitation

Water managers need non-stationary tools to better characterize precipitation extremes. Statistical approaches based on extreme value theory (EVT) are increasingly being used, but few end-to-end generalized workflows are available. In this paper, a step-by-step framework is demonstrated for developing an EVT model that considers the influence of dominant weather patterns on precipitation extremes in a watershed. Specifically, the Point Process (PP) model is utilized, which is a unified statistical framework for modeling the frequency and magnitude of extremes above a threshold. Because threshold selection can be subjective, a demonstration of how to go about selecting a threshold is provided; in particular, by examining a range of thresholds. The workflow is applied to daily precipitation from the Rio Grande watershed in New Mexico. In this arid watershed, extreme precipitation events substantially contribute to total runoff. An improved understanding of the drivers and extent of changes in extreme precipitation is essential for water resource and risk management. In addition to a stationary PP model without covariates, several covariates are examined for inclusion in the location and scale parameters. The significance of including the covariates is assessed, as well as several additional criteria, including if the covariate (s) make intuitive sense and if it is a good candidate for statistical downscaling (i.e., methods that relate largescale variables to the local scale). A final PP model is selected that includes the wet weather types in the location and scale parameters. This model is applied in a downscaling context using a large ensemble of climate projections, which shows that the frequency of exceeding a high threshold increases after 2050, but the conditional likelihood of exceeding the maximum observed precipitation stays relatively constant.