Articles | Volume 18, issue 7
Hydrol. Earth Syst. Sci., 18, 2645–2656, 2014
https://doi.org/10.5194/hess-18-2645-2014
Hydrol. Earth Syst. Sci., 18, 2645–2656, 2014
https://doi.org/10.5194/hess-18-2645-2014

Research article 17 Jul 2014

Research article | 17 Jul 2014

Evaluation of Mekong River commission operational flood forecasts, 2000–2012

T. C. Pagano T. C. Pagano
  • Bureau of Meteorology, 700 Collins Street, Docklands VIC 3008, Australia

Abstract. This study created a 13-year historical archive of operational flood forecasts issued by the Regional Flood Management and Mitigation Center (RFMMC) of the Mekong River Commission. The RFMMC issues 1- to 5-day daily deterministic river height forecasts for 22 locations throughout the wet season (June–October). When these forecasts reach near flood level, government agencies and the public are encouraged to take protective action against damages. When measured by standard skill scores, the forecasts perform exceptionally well (e.g., 1 day-ahead Nash–Sutcliffe > 0.99) although much of this apparent skill is due to the strong seasonal cycle and the narrow natural range of variability at certain locations. Five-day forecasts upstream of Phnom Penh typically have 0.8 m error standard deviation, whereas below Phnom Penh the error is typically 0.3 m. The coefficients of persistence for 1-day forecasts are typically 0.4–0.8 and 5-day forecasts are typically 0.1–0.7. RFMMC uses a series of benchmarks to define a metric of percentage satisfactory forecasts. As the benchmarks were derived based on the average error, certain locations and lead times consistently appear less satisfactory than others. Instead, different benchmarks were proposed and derived based on the 70th percentile of absolute error over the 13-year period. There are no obvious trends in the percentage of satisfactory forecasts from 2002 to 2012, regardless of the benchmark chosen. Finally, when evaluated from a categorical "crossing above/not-crossing above flood level" perspective, the forecasts have a moderate probability of detection (48% at 1 day ahead, 31% at 5 days ahead) and false alarm rate (13% at 1 day ahead, 74% at 5 days ahead).