Next-Generation IDF Curves for Hydrologic Design

Precipitation-based intensity-duration-frequency (PREC-IDF) curves are a standard tool used to derive design floods for hydraulic infrastructure worldwide. In snow-dominated regions where a large percentage of flood events are caused by snowmelt and rain-on-snow events, precipitation alone can be a poor predictor of flood risk. Yan et al. (2018) proposed next-generation intensity-duration-frequency (NG-IDF) curves, which characterize the actual water reaching the land surface (all melt events plus rainfall on snow-free ground), as an alternative to the standard PREC-IDF curves for hydrologic design.

In this study, we evaluate the impact of IDF curves on estimated flood frequency. We first compared peak design flood estimates from the National Resource Conservation Service TR-55 event-based model driven by the NG-IDF and PREC-IDF curves at 399 Snowpack Telemetry (SNOTEL) stations across the western United States, all of which had at least 30 years of high-quality records. We found that about 72% of the stations in the western United States showed the potential for underdesign, for which the PREC-IDF curves underestimated peak design floods by as much as 324%. Next, we used a physics-based hydrologic model, the Distributed Hydrology Soil Vegetation Model (DHSVM), as a standard (the "truth") to benchmark the performances of the IDF based approaches in estimating peak design floods in snow dominated locations. We found that compared to DHSVM simulation results, the PREC-IDF event-based method significantly underestimated/overestimated flood peaks; while the NG-IDF method produced results much closer to those of DHSVM. When using the PREC-IDF method, only 26-33% of the SNOTEL locations showed peak design flood errors within 20% for return periods between 10 and 50 years. Conversely, with the NG-IDF method 81-87% SNOTEL locations had peak flood errors within 20%. Due to the limitations within the event-based model (i.e., same return period between the design storm and flood), errors in design flood estimates are unavoidable and errors less than 20% are typically considered good in practice.