Developing a Subseasonal-to-Seasonal Extreme Precipitation Events Database for the Contiguous United States

Drought and extreme precipitation are two natural hazards that pose significant risks on multiple spatiotemporal scales, particularly the subseasonal to seasonal (S2S) timescale. This timescale is also crucial for a wide range of stakeholders including water resources, emergency management, and energy. Yet, there exists a significant gap in forecasting skill on S2S timescales, particularly with numerical models, due to the complex nature of the weather-climate interface. While there is emerging work on drought mitigation on the S2S timescale, literature regarding extreme precipitation on the S2S timescale is less common.

This study seeks to develop an S2S extreme precipitation events database for the contiguous United States (CONUS). In order to develop robust thresholds for defining extreme precipitation, particularly in a changing climate, we employ quantile regression, a form of linear regression that incorporates a loss function such that the regression line can be fit to any part of the probability distribution. The use of quantile regression allows the quantification of nonstationarity of extreme precipitation throughout the CONUS by defining an extreme threshold as a function of time (i.e., year). Models are fit on the 95^th percentile for each point in space and use moving windows (e.g., 365 individual 14-day moving-window models). The regression is calculated on daily Livneh precipitation data spanning 1915-2011. Additionally, several other checks (e.g., ensuring at least 5 rainy days) are conducted to ensure that the database is identifying large-scale anomalous events on the S2S timescale as opposed to short-duration events. Kernel density estimation is then used to outline regions classified as “extreme” precipitation. These regions are then grouped using K-means clustering to generate regions into which individual S2S extreme precipitation events fit. Various statistics of the identified events, such as areal-averaged rainfall, will be presented as to increase understanding of an S2S extreme precipitation event. Sensitivity of the database to the threshold percentile chosen and initial composites of characteristic synoptic/global patterns leading up to and occurring during the specific S2S extreme precipitation events are also discussed.