Validating and Enhancing Extreme Precipitation Projections by Downscaled Global Climate Model Results and Copula Methods

Abstract

Extreme precipitation has posed a huge risk to society and the environment. It is crucial to be able to accurately analyze extreme precipitation in order to reduce its potential risk. This paper presents a validation assessment, exploration, and improvement framework and systematically studies three state-of-the-art downscaling methods applied to six different global climate model (GCM) results for extreme precipitation projection. For the purposes of illustration, the paper applies this framework to data collected from the Washington, DC, metropolitan area from 1950 to 1995. The assessment shows that existing downscaled GCMs do not adequately predict some extreme precipitation indices based on historical records, such as the annual maximum 2-day precipitation and number of days with precipitation more than 20 mm. To explore possible ways of improving the accuracy, marginal distribution and day-to-day serial dependency of extreme precipitation are studied for the downscaled GCMs and observed precipitation. The projection results are further improved by incorporating serial dependency from observed precipitation into downscaled GCM results by means of copulas. In conclusion, the proposed method provides a generic way to further improve downscaled GCMs for extreme precipitation. The analytical results validate the method and show that significant improvement can be achieved from this method.