Evaluation and machine learning improvement of global flood simulations
Abstract
A warmer climate is expected to accelerate the global hydrological cycle, causing more intense precipitation and floods. Despite recent progress in global flood risk assessment, the accuracy and improvement of flood simulation is still not sufficient for most applications. Here we compare global flood simulations from five global hydrological models (GHMs) under the Inter-Sectoral Impact Model Intercomparison Project 2b (ISMIP2b) protocol within a hybrid physics-machine learning approach, against those calculated from 1032 gauging stations in the Global Streamflow Indices and Metadata Archive (GSIM) for the period of 1971-2010. A machine learning approach, namely the Long Short-Term Memory Units (LSTM) is then used to improve the performance of GHMs flood simulations within a hybrid physics-machine learning approach. We find that the GHMs perform reasonably well in terms of amplitude of peak discharge but are relatively poor in terms of their timing. The performance showed great discrepancy under different climate classifications. The large variations in performance between GHMs demonstrates that those models require improvements. LSTM used in combination with those GHMs is then shown to dramatically improve the performance of global flood simulations (especially in terms of amplitude of peak discharge), demonstrating that the combination of classical flood simulation and machine learning techniques may be a way forward for more robust and confident flood risk assessment.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H33L2122Y
- Keywords:
-
- 1847 Modeling;
- HYDROLOGY;
- 1873 Uncertainty assessment;
- HYDROLOGY;
- 1906 Computational models;
- algorithms;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS