Extreme Precipitation Event Catalogue (EPEC)

The EPEC is a catalogue of labelled extreme precipitation events resolved at 0.03332 degrees (~4km) hourly across the Contiguous United States (CONUS) from January 2002 through May 2025. This event-based approach offers novel insights into the space and time evolving structure of extreme precipitation, enabling understanding of extreme precipitation event characteristics and trends at local scales. EPEC includes 31,140 storms labelled as microscale (n = 2,340), mesoscale (n = 28,034), synoptic (n = 554), and Tropical Cyclone (n = 248). The intended audience includes flood hazard research scientists interested in modeling hydrologic and hydraulic response from extreme precipitation events, civil engineers, city managers, and watershed planners interested in understanding recent trends in extreme precipitation events at scales relevant to planning design, and private industry including banking, insurance, and reinsurance organizations interested in reducing portfolio-scale risks.

EPEC is created by combining NOAA Analysis of Record for Calibration (AORC), NOAA Atlas 14 Annual Recurrence Intervals (ARIs), unsupervised clustering using Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and supervised classification using Random Forest (RF). Each event within the EPEC includes both a PNG file and a GeoTIFF file. Each included PNG file plots the spatial extent of the event and is colored by the maximum exceeded return period at that location. The start and end date times are also included in the subtitle of each PNG image. Each included GeoTIFF file is named using the convention 'YYYY-MM-DD-XXX.tif' where YYYY-MM-DD is the year-month-day associated with the first day of the precipitation event and XXX is the storm ID number within that day. GeoTIFF layers are named using the convention 'tYYYYMMDD_HHMMSS' which represents the date time for each time step. Raster values represent hourly precipitation derived from NOAA AORC and are in units of millimeters per hour (mm/hr). A csv file is included (cluster_characteristics_label_all_predict.csv) which describes individual event characteristics and the probability of each event's classification as one of four storm types (microscale, mesoscale, synoptic, and tropical cyclone). Storm types were determined using a Random Forest classification model trained (80%) and tested (20%) using a 10-fold cross-validation on a 10% sample of hand-labelled events from the catalogue. Hand-labeled events are defined under the 'label' column and all event predictions are defined under the 'predicted_class' column where 'MICRO' ~ microscale, 'MCS' ~ mesoscale, 'SYN' ~ synoptic, and 'TC' ~ tropical cyclone.