This study aims to explore the sources of divergences between daily and sub-daily projections of the impact of climate change on river flow metrics.
Two lumped conceptual hydrological models (GR4H and MOHYSEH) are calibrated using two climate gridded datasets (MSWEP2 and ERA5) at 3-h and 24-h time steps, on 221 10 km2 à 2 000 km2 basins from Mexico, Quebec and the USA. The resulting simulations are compared in terms of deltas of seasonal maximum and 100-year return flows) between periods of the recent past based on contrasting temperatures, and precipitation volumes and intensities. For future climates, the comparison is based on projections from seven regional climate simulations of high spatial resolution (0,11° and 0,22°). At the daily scale, 24-h flow series are compared with daily-averaged 3-h flow series, thereafter called 324 series. At the sub-daily scale, 3-h flood metrics are compared with the 24-h metrics, multiplied by a peak factor, as in standard engineering practices.
The differences between 3-h, 324 and 24-h simulations were first described. The 3-h simulations performed better to adapt to past contrasting climate conditions in median terms over the sample, but the seasonal and regional internal variability is considerable and interdependent on the modeling chain configuration (hydrological model, reference climate grid, targeted metrics). For future simulations, the uncertainty arising from the temporal resolution is small in median terms, but rises to between 20% and 70% depending on the basin, when its strong interactions with the other components of the modeling chain are taken into account. The time step impact is more pronounced in certain contexts : variation in rainfall seasonality, intensification of precipitation, diversification of flood generating process on snow-dominated basins.
A number of complementary approaches were then explored to analyze the divergences between the simulations. First, a peak factor was computed for each flood of historical series as the ratio between 3-h and 324 peaks. The inter-floods and inter-climates stability of the peak factor was analyzed. The ability of the hydrological models to simulate these observed peak factors and their variability was also assessed. The series of peak factors computed on sorted historical floods, (Fp), reveal a significant breakpoint in 55% of cases, and both positive and negative trends. These breakpoints indicate distinct flood-generating processes for common and rare floods. Also, the peak factor averaged over floods, Fp, is not stationary between past contrasting periods in 48% of cases. Last but not least, Fp depends on the time reference chosen for the daily-averaging, leading to differences of up to 83% depending on whether the 324 series is computed in UTC or local time. These findings are a strong reminder of the limited physical meaning of the peak factor. We do not recommend its use in a climate change context.
A theoretical experiment was then carried at the event scale, applying simple climate changes (variations in base flow, rainfall intensity and rainfall peak timing) to a typical rainfall event routed through a theoretical unit hydrograph. The subsequent modified flood hydrographs were daily-averaged to analyze the distortion of the climate change signal through the averaging process. The results were compared to the differences observed at the climate scale. Two cases can lead to opposite 3-h and 324 deltas for basins with a time of concentration shorter than 48-h, making it impossible to infer sub-daily delta from daily delta : 1) opposite changes of base flow and runoff flow, e.g., drier basin receiving heavier events. 2) time-to-peak in such a way that the peak is close to a bound of the averaging window in the reference period, but not in the future, or vice versa. The impact of these changes in base flow and time-to-peak depends on the model’s ability to represent them.
In addition, classification tree ensembleswere constructed based on geophysical and hydroclimatic characteristics of the basins, to identify predictors of potential divergences. The most important predictors depend on the hydrological model, the targeted metrics, and the hydrological regime of the basins. However, some are crucial in all cases. Key climate predictors are deltas in daily and sub-daily maximum precipitation (ΔRX24, ΔRX3). Key hydrological predictors are the base flow index BFI, the mean peak factor Fp and its coefficient of variation CVfp . Key geophysical predictors are the wetland and the forest percentages.
In conclusion, when sub-daily data is available, we recommend including the comparison of daily and sub-daily projections in the modeling framework. The detection of divergences then adverts for significant changes in the basin’s dominant flood-generating processes, and for the need for a thorough study of the simulations before using them for engineering purposes.
| Date | 5 Aug 2025 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Annie Poulin (Supervisor) |
|---|