Extreme Precipitation in the San Francisco Bay Area: Comparing Downscaling Methodologies’ Skill in Representing Extreme Precipitation in Hindcasts and Differences in Their Projections

Abstract:

Despite the growing availability of high-resolution datasets of spatially downscaled CMIP5 projections, few studies have explored the differences in extreme precipitation events that stem from the choice of downscaling method, or from the specific climatological datasets that are used for the bias correction and spatial disaggregation. Here we take three different statistically downscaled methods applied to CMIP5 global climate models and analyze their extreme precipitation events, hindcasted and projected, for the location of NASA Ames Research Center, in South San Francisco Bay. The downscaling methods analyzed are: i) Bias Correction Spatial Disaggregation (BCSD), ii) Bias Correction Constructed Analogs (BCCA), and iii) Extreme-value model based on synoptic climate predictors. We fit a generalized extreme value distribution (GEV) to datasets i and ii and use statistical tests to determine the significance of differences in the fitted GEV parameters. We explore the implications of the GEV parameter differences by comparing the daily precipitation values corresponding to 100-year, 500-year and 1,000-year return periods in the three datasets. The implications of how extreme daily values are assumed to change with spatial scale, from the gage location (a point location), to a small grid cell (1 km) or a larger grid cell (12 km), are explored.

From our preliminary results, BCCA and BCSD projections predict that extreme precipitation events will be on the rise, and may have the potential to cause flooding at NASA Ames, and in the surrounding Bay Area. These downscaling methods can be studied in further detail in different regions of the contiguous US, and be used by local water resource management agencies in planning infrastructural adaptations.