NRCS curve number method: A comparison of methods for estimating the curve number from rainfall-runoff data

Authors: Moglen, Glenn; SADEQ, H. - University Of Maryland; HUGHES, L. - Howard University; MEADOWS, M. - University Of South Carolina; MILLER, J. - Desert Research Institute In Las Vegas; RAMIREZ-AVILA, J. - Mississippi State University; TOLLNER, E. - University Of Georgia

Submitted to: Journal Hydrologic Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/10/2022
Publication Date: 8/8/2022
Citation: Moglen, G.E., Sadeq, H., Hughes, L., Meadows, M., Miller, J.J., Ramirez-Avila, J., Tollner, E. 2022. NRCS curve number method: A comparison of methods for estimating the curve number from rainfall-runoff data. Journal Hydrologic Engineering. https://doi.org/10.1061/(ASCE)HE.1943-5584.0002210.
DOI: https://doi.org/10.1061/(ASCE)HE.1943-5584.0002210

Interpretive Summary: The Natural Resources Conservation Service (NRCS) Curve Number method has been used to model runoff volume from precipitation events since the 1950s. A central piece of this method is the longstanding approach to subtract an "initial abstraction" of 20 percent of watershed storage from the precipitation depth, before runoff can occur. In recent years, this 20 percent figure has been challenged as too large, with 5 percent often cited as more appropriate. This work uses historically-collected Agricultural Research Service (ARS) rainfall-runoff observations to compare with modeled runoff. Also examined were the calibration method and the value of limiting observations to those precipitation events greater than 25.4 mm (1 inch). Results show that a least squares and an asymptotic model both perform well as calibration methods and that a censoring threshold for precipitation events had little impact on calibrated curve number or goodness-of-fit. The main point, as to the better initial abstraction ratio, showed that both ratios performed similarly. This work suggests that retaining the 20 percent figure will minimize confusion and provide a greater safety factor for future drainage infrastructure design.

Technical Abstract: A data set comprised of rainfall-runoff data gathered at 31 Agricultural Research Service experimental watersheds was used to explore curve number calibration. This exploration focused on the calibrated value and goodness-of-fit as a function of several items: calibration approach, precipitation event threshold, data ordering approach, and initial abstraction value. Calibration methods explored were least squares, the National Engineering Handbook (NEH) median, and an asymptotic approach. Results were quantified for events exceeding two precipitation thresholds: 0 mm and 25.4 mm. Natural and frequency-matched data ordering methods were analyzed. Initial abstraction ratios of 0.05 and 0.20 were examined. Findings showed that the least-squares calibration approach applied directly to rainfall-runoff data performed best. Initial abstraction ratios clearly influenced the magnitude of the calibrated curve number. However, neither ratio outperformed the other with regards to goodness-of-fit of predicted runoff to observed runoff. Precipitation threshold experiments produced mixed results with neither threshold level producing clearly superior model fit. Frequency-matching was not considered a valid analysis approach but was contrasted with naturally-ordered results indicating a bias towards producing calibrated curve numbers that were 5 to 10 points larger.