The Value of Forecasters-in-the-Loop in Real-Time Flood Forecasting in the Age of Machine Learning

Vinh Ngoc Tran; Donghui Xu; Phong Le; Jongho Kim; Tam Van Nguyen; Giang Tien Nguyen; Pedro J. Restrepo; Valeriy Y. Ivanov

doi:10.1029/2025GL118317

The Value of Forecasters-in-the-Loop in Real-Time Flood Forecasting in the Age of Machine Learning

Vinh Ngoc Tran
, Donghui Xu
, Phong Le
, Jongho Kim
, Tam Van Nguyen
, Giang Tien Nguyen
, Pedro J. Restrepo
, Valeriy Y. Ivanov

Watershed Systems Modeling

Research output: Contribution to journal › Article › peer-review

1 Scopus citations

Abstract

Machine learning (ML) applications in hydrological forecasting are increasingly prevalent and show great potential. However, many previous studies have only evaluated performance through reanalysis or retrospective simulations compared to simplified baselines. This study provides the first assessment of ML performance against actual operational forecasting systems operated by the California Nevada River Forecast Center (CNRFC), which combines the Community Hydrologic Prediction System (CHPS) with forecasters-in-the-loop. Results demonstrate that forecasters-in-the-loop systems consistently outperform ML models in both general forecasts and flood alerting across lead times up to 96 hr, even when ML models use observed forcings, while CNRFC operational process relies on biased weather forecasts. Our analysis reveals that forecaster expertise maintains forecast reliability despite inaccurate precipitation inputs, with human-guided systems showing superior performance degradation characteristics at extended lead times. These findings highlight the irreplaceable value of human expertise in operational forecasting and caution against overstating current ML capabilities in real-world applications.

Original language	English
Article number	e2025GL118317
Journal	Geophysical Research Letters
Volume	53
Issue number	8
DOIs	https://doi.org/10.1029/2025GL118317
State	Published - Apr 28 2026

Funding

V. N. Tran and V. Y. Ivanov acknowledge the support of the U.S. Department of Defense, Department of the Navy, the Office of Naval Research award #N00014‐23‐1‐2735. J. Kim was supported by a 2022‐MOIS63‐002(RS‐2022‐ND641012) of Cooperative Research Method and Safety Management Technology in National Disaster funded by Ministry of Interior and Safety, South Korea. D.X. was supported by the Scientific Discovery through Advanced Computing 5 (Capturing the Dynamics of Compound Flooding in E3SM), funded by the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research. V. N. Tran was supported by Dan Lu's Early Career Project, sponsored by the Office of Biological and Environmental Research in the U.S. Department of Energy (DOE). This manuscript has been authored by staff from UT‐Battelle, LLC, under contract DE‐AC05‐00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid‐up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan ( http://energy.gov/downloads/doe‐public‐access‐plan ). V. N. Tran and V. Y. Ivanov acknowledge the support of the U.S. Department of Defense, Department of the Navy, the Office of Naval Research award #N00014-23-1-2735. J. Kim was supported by a 2022-MOIS63-002(RS-2022-ND641012) of Cooperative Research Method and Safety Management Technology in National Disaster funded by Ministry of Interior and Safety, South Korea. D.X. was supported by the Scientific Discovery through Advanced Computing 5 (Capturing the Dynamics of Compound Flooding in E3SM), funded by the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research. V. N. Tran was supported by Dan Lu's Early Career Project, sponsored by the Office of Biological and Environmental Research in the U.S. Department of Energy (DOE). This manuscript has been authored by staff from UT-Battelle, LLC, under contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).

Keywords

forecasters-in-the-loop
machine learning
real-time flood forecasting

Access to Document

10.1029/2025GL118317

Cite this

@article{319d95924f324dfa970d9417caf38efe,

title = "The Value of Forecasters-in-the-Loop in Real-Time Flood Forecasting in the Age of Machine Learning",

abstract = "Machine learning (ML) applications in hydrological forecasting are increasingly prevalent and show great potential. However, many previous studies have only evaluated performance through reanalysis or retrospective simulations compared to simplified baselines. This study provides the first assessment of ML performance against actual operational forecasting systems operated by the California Nevada River Forecast Center (CNRFC), which combines the Community Hydrologic Prediction System (CHPS) with forecasters-in-the-loop. Results demonstrate that forecasters-in-the-loop systems consistently outperform ML models in both general forecasts and flood alerting across lead times up to 96 hr, even when ML models use observed forcings, while CNRFC operational process relies on biased weather forecasts. Our analysis reveals that forecaster expertise maintains forecast reliability despite inaccurate precipitation inputs, with human-guided systems showing superior performance degradation characteristics at extended lead times. These findings highlight the irreplaceable value of human expertise in operational forecasting and caution against overstating current ML capabilities in real-world applications.",

keywords = "forecasters-in-the-loop, machine learning, real-time flood forecasting",

author = "Tran, \{Vinh Ngoc\} and Donghui Xu and Phong Le and Jongho Kim and \{Van Nguyen\}, Tam and Nguyen, \{Giang Tien\} and Restrepo, \{Pedro J.\} and Ivanov, \{Valeriy Y.\}",

note = "Publisher Copyright: {\textcopyright} 2026. The Author(s).",

year = "2026",

month = apr,

day = "28",

doi = "10.1029/2025GL118317",

language = "English",

volume = "53",

journal = "Geophysical Research Letters",

issn = "0094-8276",

publisher = "Wiley Open Access",

number = "8",

}

TY - JOUR

T1 - The Value of Forecasters-in-the-Loop in Real-Time Flood Forecasting in the Age of Machine Learning

AU - Tran, Vinh Ngoc

AU - Xu, Donghui

AU - Le, Phong

AU - Kim, Jongho

AU - Van Nguyen, Tam

AU - Nguyen, Giang Tien

AU - Restrepo, Pedro J.

AU - Ivanov, Valeriy Y.

PY - 2026/4/28

Y1 - 2026/4/28

N2 - Machine learning (ML) applications in hydrological forecasting are increasingly prevalent and show great potential. However, many previous studies have only evaluated performance through reanalysis or retrospective simulations compared to simplified baselines. This study provides the first assessment of ML performance against actual operational forecasting systems operated by the California Nevada River Forecast Center (CNRFC), which combines the Community Hydrologic Prediction System (CHPS) with forecasters-in-the-loop. Results demonstrate that forecasters-in-the-loop systems consistently outperform ML models in both general forecasts and flood alerting across lead times up to 96 hr, even when ML models use observed forcings, while CNRFC operational process relies on biased weather forecasts. Our analysis reveals that forecaster expertise maintains forecast reliability despite inaccurate precipitation inputs, with human-guided systems showing superior performance degradation characteristics at extended lead times. These findings highlight the irreplaceable value of human expertise in operational forecasting and caution against overstating current ML capabilities in real-world applications.

AB - Machine learning (ML) applications in hydrological forecasting are increasingly prevalent and show great potential. However, many previous studies have only evaluated performance through reanalysis or retrospective simulations compared to simplified baselines. This study provides the first assessment of ML performance against actual operational forecasting systems operated by the California Nevada River Forecast Center (CNRFC), which combines the Community Hydrologic Prediction System (CHPS) with forecasters-in-the-loop. Results demonstrate that forecasters-in-the-loop systems consistently outperform ML models in both general forecasts and flood alerting across lead times up to 96 hr, even when ML models use observed forcings, while CNRFC operational process relies on biased weather forecasts. Our analysis reveals that forecaster expertise maintains forecast reliability despite inaccurate precipitation inputs, with human-guided systems showing superior performance degradation characteristics at extended lead times. These findings highlight the irreplaceable value of human expertise in operational forecasting and caution against overstating current ML capabilities in real-world applications.

KW - forecasters-in-the-loop

KW - machine learning

KW - real-time flood forecasting

UR - https://www.scopus.com/pages/publications/105036545199

U2 - 10.1029/2025GL118317

DO - 10.1029/2025GL118317

M3 - Article

AN - SCOPUS:105036545199

SN - 0094-8276

VL - 53

JO - Geophysical Research Letters

JF - Geophysical Research Letters

IS - 8

M1 - e2025GL118317

ER -

The Value of Forecasters-in-the-Loop in Real-Time Flood Forecasting in the Age of Machine Learning

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this