Conceptual rainfall-runoff models driven by long climate records (continuous simulation) are an established alternative to statistical flood-frequency analysis for estimating design floods, but their parameters are uncertain and their extrapolation to rare events is poorly constrained. A study of the upper Xiangjiang basin, southern China, using two lumped models (XAJ and HBV) found that parameter equifinality dominates calibration-period uncertainty at high return periods, that roughly 15–20 years of calibration data are needed for stable design-flood estimates, and that the two models bias design floods at high return periods in opposite directions. Whether these findings generalize to other model structures (including semi-distributed and distributed models), other climatic regimes, and other basin scales — the prerequisite for generalized guidance on continuous-simulation design flood estimation — remains open.

Background

Design flood estimation — determining the flood peak (or full flood hydrograph) associated with a given return period TT, such as the 100-year flood used for spillway design and floodplain mapping — traditionally relies on statistical flood-frequency analysis of recorded annual maxima. The continuous-simulation alternative instead drives a rainfall-runoff model with long observed (or stochastically generated) rainfall and evaporation series and derives the flood-frequency relationship directly from the simulated discharge record, which allows physically based extrapolation to rare events and carries antecedent watershed conditions along with it. The rainfall-runoff models used for this purpose are often lumped conceptual models, meaning basin-averaged representations of soil-moisture, runoff-generation, and routing stores; two canonical examples are the Xinanjiang (XAJ) model, developed for humid Chinese basins and built around a tension-water capacity curve describing how unevenly soil-moisture storage is distributed over the basin, and the HBV model (from the Swedish Hydrologiska Byråns Vattenbalansavdelning), a soil-moisture and snow store model. Semi-distributed and distributed models instead subdivide a basin into sub-basins or grid cells with lateral routing, at the cost of many more parameters.

Zeng, Chen, Xu, Jie and Hou (2016) calibrated XAJ and HBV on the upper Xiangjiang basin at Hengyang station, southern China (52,150 km², humid subtropical, daily data 1965–2005), and propagated two distinct uncertainty sources into design floods estimated by continuous simulation. The first is parameter equifinality: within the GLUE framework (Generalized Likelihood Uncertainty Estimation, in which very many parameter sets are sampled and all those producing acceptably "behavioral" calibration simulations are retained and run jointly), many parameter sets fit the calibration record almost equally well yet extrapolate differently to extreme events. The second is calibration-period uncertainty: the dependence of the calibrated model on which multi-year segment of the historical record is used for calibration. They found that XAJ tends to overestimate and HBV to underestimate the design flood at high return periods (biases approaching ±10%\pm 10\% at the 100-year event); that the GLUE uncertainty band from equifinality widens with return period while the calibration-period band stays roughly constant, so equifinality dominates at high return periods; and that design-flood estimates stabilize only once calibration uses on the order of 15–20 years of data.

These conclusions came from two deterministic lumped structures and a single humid basin. The study itself states that the extent to which its findings generalize to distributed or semi-distributed models, to different climatic regimes, and to basins of different scales is unknown, and calls for applying the same uncertainty framework to a broader range of models and catchments to derive generalized guidance for design flood estimation by continuous simulation.

Problem Statement

Apply the GLUE-based uncertainty decomposition of Zeng et al. (2016) — joint quantification of parameter-equifinality uncertainty and calibration-period uncertainty in design floods derived by continuous simulation — to a broader range of hydrological model structures (including semi-distributed or distributed models) and to catchments in different climatic regimes and of different scales than the single humid subtropical basin studied in the source, and determine: (i) the calibration record length beyond which design-flood estimates stabilize, that is, the shortest length beyond which the calibration-period uncertainty — the relative band width of the design-flood estimates obtained from different calibration sub-records of that length (maximum minus minimum across sub-records, divided by the estimate from the observed record), evaluated at the design return periods examined — decreases by no more than 1 percentage point for each additional 5 years of calibration record; (ii) whether parameter equifinality, rather than calibration-period selection, dominates design-flood uncertainty at high return periods; and (iii) whether the direction of a model's design-flood bias at high return periods is systematic across structures and settings. The objective, as stated by the source study, is generalized, transferable guidance for design flood estimation using continuous simulation — not additional single-basin case studies. Meeting it requires coverage beyond the source setting: at least one semi-distributed or distributed model structure, and at least two catchments that jointly span a climatic regime and a basin scale different from the source's.

References

  1. Qiang Zeng, Hua Chen, Chong-Yu Xu, Meng-Xuan Jie, Yu-Kun Hou (2016). Feasibility and uncertainty of using conceptual rainfall-runoff models in design flood estimation. Hydrology Research 47(4):701–717. DOI 10.2166/nh.2015.069. https://doi.org/10.2166/nh.2015.069
  2. Wei Wang, Jia Liu, Chuanzhe Li, Fuliang Yu, Yuebo Xie, Qingtai Qiu, Yufei Jiao, Guojuan Zhang (2020). Assessing the applicability of conceptual hydrological models for design flood estimation in small-scale watersheds of northern China. Natural Hazards 102(3):1135–1153. DOI 10.1007/s11069-020-03949-2. https://doi.org/10.1007/s11069-020-03949-2
  3. Wushuang Yang, Runliang Xia, Hua Chen, Min Wang, Chong-Yu Xu (2022). The impact of calibration conditions on the transferability of conceptual hydrological models under stationary and nonstationary climatic conditions. Journal of Hydrology 613:128310. DOI 10.1016/j.jhydrol.2022.128310. https://doi.org/10.1016/j.jhydrol.2022.128310
  4. Phuong Cu Thi, James E Ball, Ngoc Hung Dao (2018). Uncertainty estimation using the GLUE and Bayesian approaches in flood estimation: a case study—Ba River, Vietnam. Water 10(11):1641. DOI 10.3390/w10111641. https://doi.org/10.3390/w10111641
  5. Kenechukwu Okoli, Maurizio Mazzoleni, Korbinian Breinl, Giuliano Di Baldassarre (2019). A systematic comparison of statistical and hydrological methods for design flood estimation. Hydrology Research 50(6):1665–1678. DOI 10.2166/nh.2019.188. https://doi.org/10.2166/nh.2019.188
  6. Patrick Arnaud, Philippe Cantet, Jean Odry (2017). Uncertainties of flood frequency estimation approaches based on continuous simulation using data resampling. Journal of Hydrology 554:360–369. DOI 10.1016/j.jhydrol.2017.09.011. https://doi.org/10.1016/j.jhydrol.2017.09.011
  7. James E Ball (2020). An assessment of continuous modeling for robust design flood estimation in urban environments. Frontiers in Earth Science 8:124. DOI 10.3389/feart.2020.00124. https://doi.org/10.3389/feart.2020.00124
  8. Gianni Vesuviano, Adam Griffin, Elizabeth Stewart (2022). Flood frequency estimation in data-sparse Wainganga Basin, India, using continuous simulation. Water 14(18):2887. DOI 10.3390/w14182887. https://doi.org/10.3390/w14182887