[A-HW20] Hydrology & Water Environment

[AHW20-P04] An Efficient Surrogate Model for Extreme Hydrologic Predictions using Polynomial Chaos-Kriging

Vinh Ngoc Tran¹, *Jongho Kim¹ (1.University of Ulsan)

Keywords:Surrogate model, Hydrologic predictions, Extreme floods, Polynomial chaos-kriging, Uncertainty quantification

Due to their versatility and high-performance, surrogate models have become an attractive approach that yields the potency to break down the computational barrier of the expensive hydrological models that are often confronted with parameter uncertainties. However, the surrogate models are not qualified for accurately diagnosing extreme events that are not designed in the training data space. This study presents a state-of-the-art surrogate model, i.e., polynomial chaos-kriging (PCK), that enables us to fill the gap above effectively. The effectiveness of PCK was evidenced by comparing it with two well-known surrogate models of polynomial chaos expansion (PCE) and ordinary kriging (OK). Comprehensive analyses demonstrate that: (1) PCK outperforms others in mimicking the original model more precisely with a smaller number of design sites; (2) the extraordinary predictability of PCK is manifest by accurate predictions for extreme flood events with the flood peak about three times higher than those used to train itself, while PCE and OK failed; (3) results of sensitivity analysis verify that characteristics and behaviors of uncertain parameters in PCK are closer to those in the original model compared to PCE and OK; (4) based on a new efficient index that depicts for the robustness of the surrogate models in providing accurate predictions, PCK has shown superiority to both PCE and OK due to its accurate prognosis for extreme events, although all three surrogates speed up to about 500 times than the original model. Overall, our study reinforces a knowledge basis to perform a robust, efficient surrogate model for ensemble predictions of extreme flood events.

Acknowledgment:
This research was supported by the Water Management Research Program funded by Ministry of Environment of Korean government (127554) and by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (NRF-2019R1C1C1004833).