Estimation of extreme flood quantiles with the SCHADEX method in projected climatic conditions

Extreme floods estimation in a changing climate is a challenge facing two major methodological difficulties: extrapolation to low-probabilities events, in a non-stationary climate. Despite its complexity, such an estimation is a key input for the safety assessment,and the design of high-risk infrastructures (dam, nuclear powerplant), built for decades and supposed to withstand the future climate.

When significant, the trend on observed floods is dependent on the climatology and scale of the considered catchment and cannot be directly transposed to the high quantiles required for safety assessment. In the literature, three modeling frameworks can be distinguished to compute these estimations in a non-stationary climate:

• Non-stationary Flood Frequency Analysis performed with time or a climate covariable.
• Dedicated extreme flood estimation method (like those being based on rainfall-runoff stochastic simulation), integrating the time or a climate covariable.
• Complete climate and hydrology modeling chains (combining GCM, RCM and hydrological models).

The approach proposed here falls into the second category, with the application of the SCHADEX method in an evolving climate. The SCHADEX method  it is based on a semi-continuous rainfall–runoff simulation process which allows the generation of an exhaustive set of crossings between precipitation and soil saturation hazards.

In this study a regional surface temperature, modeled by 10 different GCMs from the CMIP5 project for the RCP 4.5 and 8.5 scenarios, is used to drive a non-stationary, temperature-varying, distribution of extreme rainfall. The temperature-quantile models are calibrated season by season thanks to the observations of the 1950-2019 period where trends are statistically significant. Another method downscales the same GCM models thanks to the analog method to generate projected series of areal rainfall and basin average temperature. These series are used as future climatological input of the SCHADEX method. For several 35-years windows up to 2099, and for the RCP 4.5 and 8.5 scenarios, SCHADEX computes the estimation of extreme flood quantile based on both the projected extreme rainfall distribution and climatology. These estimations are compared to the 1985-2019 reference period to assess the evolution of estimated high quantiles.

The study is based on a dataset of seven catchments ranging from 1200 to 7000 km² located in various regions of the South-East half of France with contrasted climates. Only the significant rainfall trends are modeled, assuming a stationary extreme rainfall distribution otherwise.

The most significant changes in extreme rainfall are for basins under Mediterranean influence. Due to the non-linearity on the catchment’s response to heavy rain, the changes in extreme flood estimation are generally higher than the changes in extreme rainfall. In most catchments, drier future pre-flood conditions do not significantly dampen the increase of rainfall. For mountainous catchments, increased temperatures lead to higher rain-snow limit during intense events in autumn, and higher pre-flood snowmelt in spring, globally increasing the efficiency of heavy rainfall.

Some perspectives of this study are drawn, among them the need for another climate variable in the models, aside the surface temperature, which could account for the cyclogenesis evolutions. Some discrepancies between the two modeling chains (extreme rainfall distribution and climatology) are also to be tackled.