Introduction

In the United States, federal flood frequency guidance has been codified since 1967 (England et al., 2019). The most recent versions of these documents, Guidelines for Determining Flood Flow Frequency Bulletin 17B and Guidelines for Determining Flood Flow Frequency Bulletin 17C, were published by the Advisory Committee on Water Information Hydrology Subcommittee in 1982 and 2019, respectively. The Subcommittee on Hydrology recognized the need for “identification and treatment of mixed distributions, including those based on hydrometeorological or hydrological conditions” in the Future Studies section of both documents (Hydrology Subcommittee Interagency Advisory Committee on Water Data, 1982; England et al., 2019).

The United State Geological Survey (USGS) and the U.S. Army Corps of Engineers (USACE) are in year 4 of a 4-years Interagency Reimbursable Work Agreement (IRWA) to support the Federal Emergency Management Agency (FEMA) in meeting its goals of developing a risk informed National Flood Insurance Program (NFIP) and creating valuable flood hazard and risk information, the Risk Management Directorate implemented the Future of Flood Risk Data (FFRD) initiative. 

USGS and USACE have collaborated to achieve four priorities which help enable improved flood frequency methods needed to support FFRD:

  • Identify improved mixed population flood frequency analysis methods for use at select sites with long-term streamgage data in the United States based on physical causal mechanisms.
  • Develop and test semi-automated methods for determining flood type classifications at USGS streamgages across the nation.
  • Identify changes in the proportion of floods originating from different combinations of meteorological drivers and antecedent watershed conditions.
  • Prescribe a plan for national implementation, including training of federal employees.

Three large-scale "meta regions" representing relatively uniform hydroclimatic conditions within the U.S. were identified to develop flood frequency analysis and flood type classification methodologies: 1) Mountainous, 2) Central / Plains, and 3) Humid East. Two watersheds were selected within each meta region (for a total of six watersheds): one for calibration and one for validation. The pilot watersheds are shown in the following figure.

Selected Pilot Regions

Development of Flood and Storm Type Classification Methods

Manual and automated flood and storm type classification algorithms were developed for the six pilot watersheds. Floods were assigned primary and, when appropriate, secondary causal mechanisms, distinguishing among rain-only, mixed rain and snow, and snowmelt-only events, with further subdivision by storm type (e.g., atmospheric rivers, frontal systems, convective storms, tropical cyclones/tropical storm remnants, monsoon, long-duration rainfall) and modifiers such as frozen or saturated ground, ice jams, and wildfire. The classification relied on a combination of historical narrative sources—USACE and USGS post-flood reports, newspaper articles, water control manuals, storm data from the National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Information (NCEI), and Daily Weather Maps (1871–present)—and time series extracted from gridded datasets that describe precipitation, snow water equivalent (SWE), temperature, soil moisture, atmospheric rivers, convective available potential energy, and synoptic-scale circulation. Time series were extracted using the HEC-Vortex Grid-to-Point tool, and these multiple lines of evidence were synthesized to infer the dominant flood-generating processes and to provide a foundation for developing and testing flood typing procedures in mixed population settings.

An automated flood typing procedure was developed for each pilot watershed using a threshold-based approach, as shown in the following figures. 

Upper Colorado River Region Flood Classification Flow Chart

The goal of the automated procedure was to reproduce the results of the manual flood typing procedure using zonal statistic time series from gridded data sources and tropical storm track information. Performance of the automated flood typing algorithms was evaluated by comparing the results obtained through automated procedures against those that were derived manually at gages within each of the 6 pilot regions. Confusion matrices and summary statistics were used to quantify model performance compared to manual results.

Improved Mixed Population Flood Frequency Analysis Methods

When flood events are caused by different types of meteorologic or hydrologic phenomena, special treatment may be required in order to make adequate estimates of the potential for flooding. Starting in the mid 20th century, multiple federal agencies began applying special treatment to flood events occurring within mid-Atlantic states where tropical cyclone/tropical storm remnant (TC/TSR) events strongly influence the perceived flood threat (Beard, 1962). In such cases, flood frequency relationships have been developed using samples that originate from two or more separate causal factors, often referred to as a mixed population flood frequency analysis (hereafter referred to simply as mixed population analysis). An example of an annual maximum series (AMS) comprised of multiple flood producing mechanisms is shown below. In this instance, flood frequency relationships must be specially derived in order to accurately estimate quantiles of interest, especially when extrapolating beyond the range of observed data.

Example of an Annual Maximum Series Comprised of Multiple Flood Mechanisms

Multiple approaches were then utilized to estimate flood frequency.

First, a "baseline" approach to estimating flood frequency was employed which used Bulletin 17C procedures (England et al., 2019). This baseline approach disregarded all flood mechanism information and treated the entire AMS as a single homogenous sample.


Engineer Manual 1110-2-1415 (EM 1415, USACE, 1993) recommends that floods originating from two or more hydrologically independent causative mechanisms be separated, probability distributions be fit to each sample, and combined using the probability of union approach.

Example Baseline Results


Example EM 1415 Results


Example Comparing Baseline and EM 1415 Results

For locations that exhibited large disparities in magnitudes between the various flood mechanism-specific AMS, extrapolations beyond the range of observed data were notably different between the EM 1415 and baseline methods. These extrapolations to rare AEP were afforded by the use of flood mechanism-specific AMS which were disregarded when using the baseline method. Notice that the combined curve computed using the EM 1415 method has a noticeable inflection upwards (as compared to the baseline results) and follows the upward trend of the empirical plotting positions for the largest observed annual maxima.

Identification of Changes in the Magnitude and Frequency of Flood and Storm Types Over the Historical Past

The classification algorithms were used to determine how flood and storm types (and their flood magnitudes) vary in both time and space for the historical period within each pilot region. Partial duration series (PDS) were extracted using the Peaks Over Threshold analysis within the Hydrologic Engineering Center's Statistical Software Package (HEC-SSP). Annual peak flows and partial duration series were used to assess spatial and temporal variation in flood generating mechanisms. 

Spatial Variations in Flood Mechanisms

Flood generating mechanisms were found to vary geographically within the pilot regions.  For instance, within the Delaware River watershed, floods driven by mixed events are more common in the headwaters (northern latitudes) of the watershed than further south as a result of colder temperatures and increased precipitation falling as snow or remaining on the ground as snow. Conversely, a greater percentage of flood events are driven by rain-only floods in the southern portion of the watershed. In addition, TC/TSR driven flood events are more prevalent in the southern portion of the basin, which is closer to the Atlantic Ocean, as shown in the following figure.

Delaware River Spatial Distribution of AMS Flood Classifications

Within the Puget Sound region, differences in runoff mechanisms are primarily caused by the Cascade Mountain range and its influences on the predominant weather patterns affecting this region. As warm, moist air from the Pacific Ocean moves inland, the air tends to cool and release significant precipitation as it reaches the Cascade Mountains. The presence of this mountain range causes average annual precipitation in this region to vary significantly. On the western side of the Cascade Mountains, approximately 30 inches of precipitation falls per year near Seattle, WA. This average annual accumulation increases to over 90 inches along the crest of the Cascade Mountains. On the eastern side of these mountains, a rain shadow effect is noticeable with average annual accumulations of 20 inches per year near the Columbia River to 70 inches near Lake Chelan. Also, precipitation form (i.e., rain or snow) is highly dependent upon elevation due to the tendency for air to cool as it rises. Consequently, as elevations increase, snow can accumulate to a significant degree while rain is a much more form of precipitation in lower elevations. This leads to rain-only events being much more common in the western portion of the Puget Sound region while snowmelt-only floods are more prevalent in the eastern portion of the watershed, as shown in the following figure.

Within the Red River of the North, the proportion of snowmelt-only flood events is highest in the northwestern portion of the basin (northern latitudes) as a result of colder temperatures and increased precipitation falling as snow or remaining on the ground as snow. The proportion of mixed flood is highest in the southeastern portion of the basin, as shown in the following figure.

The prevalence of mixed, rain-only, and snowmelt-only floods has slight variations throughout the Upper Colorado River region, as shown in the following figure. Typically, the more the watershed contains higher elevations, the more the flow is dominated by snowmelt and mixed flood mechanisms. Lower elevations, particularly smaller watersheds, contain a higher proportion of rain-only events, as these not only have a smaller seasonal snowpack, but are also more sensitive to runoff from small, localized storms.

Upper Colorado River Region Spatial Distribution of AMS Flood Classifications

Seasonal Timing in Flood Mechanisms

The seasonal timing and temporal (decadal) variation in flood generating mechanisms were examined. Within the Trinity River watershed, the primary driver of flood events are frontal and convective rainfall, which most frequently occur in the spring and summer months, as shown within the following figure. Tropical storms impact the Trinity River basin in the late spring through early fall (June through October). 

Within the Puget Sound Region, atmospheric rivers (AR) are the predominant source of moisture delivery, rain-only and mixed flood events are more likely to occur in the late fall/winter/early spring months roughly spanning October through March each year. Thus, flooding within the portion of the Puget Sound region located west of the Cascade Mountains is more likely to occur in those months, as shown within the following figure.

Conversely, snowmelt-only flooding, which is more prevalent within the portion of this region located east of the Cascade Mountains, requires snow to accumulate and then be subjected to elevated temperatures over an extended time period. These conditions are more likely to occur in the late spring/early summer months. Thus, this runoff mechanism occurs much more frequently in the late spring/early summer months, as shown within the following figure.

Temporal Variations in Flood Mechanisms

Changes in the seasonal timing and frequency of floods over the historical period were examined. Within the Delaware River watershed, the percentage of all flood peaks occurring in each month were determined for two 60-year periods: 1900–1960 and 1960–2020. Flood timing has shifted significantly between the early period (1900–1960) and the recent period (1960–2020). This shift has manifested in a significant decrease in the prevalence of flood events in the months of March, July, August, November, and December and an increase in AMS events in the month of April, May, June, and September, as shown within the following figures. This change in seasonal timing of peak flow events is likely attributable to several disparate physical factors. However, a possible explanation is the large-scale construction of reservoirs throughout the Delaware River region in/around the 1960s. 

In addition, the frequency (as indicated by the average number of flood events per year, or arrival rate) of floods has shifted the historical period in the Delaware River region. The arrival rates of mixed flood events decreased from water year (WY) 1901 to 2020 while the arrival rates of rain-only flood events increased from 1901 to 2020, as shown in the following figures.

Plan for National Upscaling and Dissemination of Results

The team has delivered numerous webinars and conference presentations, and published several journal articles (links provided below). In the fourth and final year of the work unit, the USGS and USACE team are developing technology transfer and training materials and software features to support operationalization of methodologies developed in the work unit. In addition, HEC will be delivering updated mixed population related material in the Flood Frequency Analysis PROSPECT course in April 2026. 

This page will be updated with links to Mixed Populations related webinars, presentations, and conference materials.

Webinars

Conference Presentations

American Meteorological Society 2026

American Geophysical Union Annual Meeting 2025

American Geophysical Union Annual Meeting 2024

SEDHYD 2023

Journal Publications

References

England, J.F., Jr., Cohn, T.A., Faber, B.A., Stedinger, J.R., Thomas, W.O., Jr., Veilleux, A.G., Kiang, J.E., and Mason, R.R., Jr., (2018). Guidelines for Determining Flood Flow Frequency - Bulletin 17C (ver. 1.1, May 2019): U.S. Geological Survey Techniques and Methods, book 4, chap. B5. Reston, VA: U.S. Geological Survey.

Hydrology Subcommittee Interagency Advisory Committee on Water Data. (1982). Guidelines for Determining Flood Flow Frequency Bulletin 17B. Reston, VA: U.S. Geological Survey.

U.S. Army Corps of Engineers. (1993). Hydrologic Frequency Analysis. Engineer Manual 1110-2-1415. Washington, D.C.: Department of the Army.