[M-GI26] Data-driven approaches for weather and hydrological predictions

[MGI26-09] Uncertainty Quantification of Flood Hazards Focusing on Parameters of a Rainfall-Runoff-Inundation Coupled Model

*日浦 直紀¹、澤田 洋平¹ (1.東京大学)

It is important to consider uncertainties in the estimation of flood hazards to take reasonable measures against floods. There are many unknown parameters in hydrodynamic models which simulate processes from rainfall to inundation, and these unknown parameters are the source of uncertainty in model-estimated flood hazards such as inundation depth. In many practices, the parameters of hydrodynamic models are calibrated by river discharge observation since the details of inundation processes are not directly observed. However, it is often the case that the quantity of information from the observation of river discharge is too small to fully constrain the parameters, and there are many combinations of parameters which can equally reproduce observed river discharge (equifinality), which leads to uncertainty in estimated parameters. Such parametric uncertainties bring uncertainties in estimated inundation depth. Here we aim to quantify this uncertainty and its sensitivity to observation of river discharge. We used the rainfall-runoff-inundation (RRI) model, which analyzes the whole process from rainfall to inundation. The probability distribution function of the parameters was calculated by the Markov Chain Monte Carlo method (MCMC). To reduce the computational cost of MCMC, we constructed a surrogate model which can efficiently mimic the relationship between parameters and likelihood. Then, the parameters were sampled from the posterior distribution, and an ensemble simulation was performed to obtain the distribution of inundation depth. We find that the estimation of inundation depth has considerable uncertainties, which makes it infeasible to discuss the risk for human damage. We also find that the increase in observation error and/or the decrease in the number of observation gauges significantly increases the uncertainty of the estimation of inundation depth. This study implies that the flood risk assessment based on hydrodynamic models calibrated only by river discharge observations has large uncertainty, and it is necessary for the accurate risk assessment to obtain the detailed observation of flood inundation.