Improvement of the probabilistic overflow hazard modelling chain in France
(1) Thomas Onfroy - Modelling, R&D Department, CCR
(2) Etienne Leblois - Riverly-Lyon Research Unit, National Research Institute for Agriculture, Food and the Environment (INRAE)
(3) Jean-Philippe Naulin - Modelling, R&D Department, CCR
(4) David Moncoulon - Modelling, R&D Department, CCR
INTRODUCTION
Modelling the overflow of rivers during a flood is an essential step in modelling subsequent insured damage. Based on the claims recorded by CCR, it is estimated that almost 35% of the claims, representing 60% of the loss amounts, are located in the overflow areas of the main rivers. The remaining damages are the result of runoff phenomena, network overflow, rising water tables or the overflow of small rivers or sections located at the head of the basin.
As part of an ongoing process of improving Nat Cat models, the overflow simulation chain has been updated to strengthen the estimation of flood flows and the spread of water levels within the major river beds. This improvement allows for an updated exposure map as well as an estimate of the likelihood of damage due to overflow. The modelling of the probabilistic overflow hazard is based on two sources of rainfall-runoff models used in a complementary manner: (1) the INRAE model which is used to model slow and fast floods (2) the CCR rainfall-runoff model, adapted to the simulation of fast floods.
Both models use ARPEGE-Climat data from Météo-France as input. This data is available for current climate, but also for future climate under the IPCC RCP 4.5 and 8.5 scenarios. These can later be used to assess the impact of climate change on large rivers.
Three main areas of development were explored as part of this research:
- The consideration of the flow modelling carried out by INRAE on major rivers;
- The improvement of rainfall-runoff modelling on rivers and secondary catchments;
- the conversion of flows from the two methodologies into water levels and their propagation on the DTM using a hydraulic model.
METHODOLOGY
Flow modelling on large rivers (INRAE)
This modelling was carried out in the framework of a partnership with INRAE to work specifically on large rivers that are affected by slow flooding. The study was based on meteorological data provided by Météo-France as part of the climate study conducted in 2018 [1]. This includes simulations of 400 years with constant climate over which rainfall, atmospheric pressure and wind speed data have been continuously estimated by the ARPEGE-Climat model. The precipitation outputs of the ARPEGE model are available on an hourly basis with a spatial resolution of 8 km. Three versions of the 400 years are available. A version with current climate conditions, a version corresponding to the IPCC RCP 4.5 and another with RCP 8.5 scenario.
The INRAE rainfall-runoff model is a daily model, providing flow estimates on a network of 1054 hydrographic stations. It operates over the entire 400-year catalogue, taking into account the state of saturation of the soil on a continuous basis, but also the snowmelt on rivers with a snow regime. Thus, this model is particularly suitable for slow floods and large gauged rivers. It is thus more limited for fast floods that last a few hours and involve small rivers that do not have hydrographic stations. The selection of events is based on a ten-year flow threshold. All events that exceed this threshold are simulated by the hydraulic model.
Figure 1 - Block diagram of overflow modelling
Rainfall-runoff model on secondary rivers (CCR)
To estimate flow rates for rapid flood events, the CCR rainfall-runoff model was used. In contrast to the INRAE model, this is an event-based model with sub-hourly time steps. It is therefore particularly well adapted to quick events but, as it does not consider the evolution of soil moisture over time and the base flow, it is less efficient for slow floods. The rainfall-runoff simulations are carried out on all events where the rainfall exceeds the ten-year threshold. This modelling chain was applied to a catalogue of hypothetical events to determine the exposure of the French territory to the most extreme flood events and to calculate return periods. These hypothetical events are derived from 400-year continuous simulations of the ARPEGE-Climat model of Météo-France. These simulations were already used in the study [1] to assess the impact of climate change.
# overflow
# floods
# large rivers
# climate change
Spreading model
Once the water depths have been estimated, the overflow is simulated with the spreading model. This is based on the hydraulic equations of the Lisflood-FP model [2] which are also based on the Saint Venant equations. The depth spread can thus operate in a non-steady state, i.e., it takes into account time in the calculation of flows. It can thus be fed with a flow that fluctuates over time. The Lisflood model estimates a flow rate based on surface roughness and waterline slope. The flows are propagated in 4 directions on the DTM: up, down, right and left. In the end, the model provides an estimate of the maximum height of water reached during the event on each DTM grid cell, based on the 6 hours of flow during which the flood was the greatest.
This modelling chain is now slower than the one developed previously. However, this increase in computing time seems reasonable in view of the improved accuracy of the simulations.
Figure 2 - Example of topographic transect data in the Paris region
Figure 3 - Example of a simulated water level map for floods in the Languedoc region in 2018.
Figure 4 - Map of probabilistic hazard in the Paris region.
Figure 5 - Map of maximum water levels reached during a 100-year return period event in the Paris region. The extension of the 1910 Seine flood is shown in white.
THE PARTNERS
INRAE is a French research organisation that specialises in agriculture, food and the environment. It includes more than 10,000 agents in 18 research centres located across France.
RESULTS
Evaluation of the new overflow model
The model was validated on 5 significant historical events by comparing the simulated flooded areas with the observed flooded areas. The events selected are the Gard flood in 2002, the Var flood in 2010, the Seine and Loire floods in 2016, the Seine and Marne floods in 2018 and the Languedoc flood in 2018. This range of events includes both fast and slow floods. Figure 3 shows the comparison between simulations and observations for the Languedoc event in 2018. The observed areas (in red on the map) come from satellite data of the event with 10m resolution (ESA Sentinel-1 and -2). The results showed that the new overflow model detects 76% of the areas flooded during an event compared to 30% of areas that are wrongly detected as flooded (false detections). These limitations are partly explained by the average resolution of the 25m DTM. For a more accurate estimation of flooded areas and water levels, a DTM with a resolution of less than 25m will eventually be integrated into the flood modelling chain.
Probabilistic overflow hazard mapping
These simulations based on 400 years of hypothetical precipitations allow to carry out a mapping of the hazard on the entire metropolitan territory of which examples are presented in Figures 4 and 5. The first type of output isa probabilistic hazard map where each grid cell contains an estimate of the return period of the event necessary for a flood. A red grid cell, for example, may be flooded every 20 years.
The second type of output corresponds to the simulated water levels for a given return period. Figure 5 showsthe simulated water levels for an event with a return period of 100 years in the Seine catchment. The results obtained are consistent with the extension of the 1910 flood of the Seine, which is shown in white.
CONCLUSION
These developments to update the overflow modelling chain have increased the accuracy of the estimated flooded areas and water levels when simulating flooding in both major rivers and secondary catchments.
The developments have been incorporated to recalibrate the internal damage model. Probabilistic modelling allows the production of a map of return periods and associated water levels to measure the exposure of the mainland to flooding under different climate scenarios. This modelling can be replicated by using Météo-France datasets corresponding to the IPCC RCP 4.5 and 8.5 climate scenarios to assess the impact of climate change on rivers in France. This work is part of an ongoing process to improve the models and other developments may be carried out in the future, such as the use of precipitation data from models or radar observations, the use of better resolved altimetric data, or the improvement of the consideration of flood defences and underground networks./
REFERENCES
1. CCR (2018), The consequences of climate change on the cost of Natural Catastrophes in France by 2050.
2. Bates et al. 2010. A simple inertial formulation of the shallow water equations for efficient two-dimensional flood inundation modelling. Journal Of Hydrology
CITATION
Onfroy et al, Improvement of the probabilistic overflow hazard modelling chain in Metropolitan France. In CCR 2022 Scientific Report; CCR, Paris, France, 2022, pp. 06-09