STRENGTH OF THE RELATIONSHIP BETWEEN FLOODS AND EXTREME RAINFALL EVENTS IN THE UNITED STATES

The frequency and intensity of heavy rainfall events have increased throughout the United States over the past 100 years and are projected to continue to increase in the future (Karl et al., 2009; National Climate Assessment, 2014). Despite the consistent trends in precipitation, trends in flooding are not as clear due to additional complex flood generation mechanisms such as soil moisture and snowmelt as well as the impacts of land use change and watershed regulation and diversion (Berghuijs et al., 2016; Collins et al., 2014; Villarini et al., 2009; Vogel et al., 2011). This study evaluates the strength of the relationship between extreme rainfall and flooding within the contiguous United States (CONUS) over the past 30 years, for 5,268 watersheds from the Geospatial Attributes of Gages for Evaluating Streamflow, version II (GAGES-II) dataset. A flood set for each watershed is developed from historical daily streamflow records, which are compared with daily gridded precipitation data to evaluate the strength of the relationship between flood magnitude (Q) and precipitation accumulation (P). The role of antecedent conditions and watershed characteristics on the P and Q relationship strength is evaluated and regional differences in relationship strength are examined. Extreme rainfall is found to be a relatively poor predictor of flood magnitude within the CONUS, with P explaining at least 50% of the variation in Q for less than 25% of study watersheds. The relationship strength is stronger in regions that typically experience little to no snowfall, such as the Southeast and Southcentral United States. The exclusion of winter flood events increases relationship strength in some regions that experience substantial snowfall. The relationship between extreme rainfall and flooding increased slightly with an increasing percentage of urban area for watersheds that had a change in percent urban area from 1992 to 2011 of less than 5%. A multiple linear regression with seven individual days of precipitation as predictors showed improved relationship strength over a simple linear regression using three day total precipitation as a predictor of Q