RESEARCH POSTER PRESENTATION DESIGN © 2015 www.PosterPresentations.com The objective of this research is to develop a new method for generating high-resolution precipitation for hydrological and inundation model using deep learning techniques in data- coarse regions. Therefore, there are three main goals to achieve in the thesis: 1. To validate the performance of Convolutional Neural Networks (CNN) for downscaling. 2. To compare the CNN results with dynamical downscaling model: Weather Research Forecast (WRF) outputs. 3. To compare the hydrological model results from in situ data and CNN. OBJECTIVES INTRODUCTION METHODOLOGY RESULTS FUTURE WORK Yi-Chia CHANG, Dr. Ralph ACIERTO, Tomoaki ITAYA, Dr. Akiyuki KAWASAKI, Dr. Ching-Pin TUNG Department of Civil Engineering, The University of Tokyo, Japan; Department of Bioenvironmental Systems Engineering, National Taiwan University, Taiwan A Deep Learning Approach to Downscaling Precipitation for Flood Risk Assessment over Data-scarce Region: Case Study of Bago River Basin, Myanmar Corresponding Email: [email protected] Future climate data are essential for conducting the hydrological assessment. Due to the scarce scale of climate data generated from General Circulation Models (GCMs), the native-scale outputs of climate models could not be directly utilized for basin-scale hydrologic models. Regional Climate Model (RCM) is a dominant technique to dynamical downscaling from the global scale to regional scale with consideration of atmospheric physical processes. However, the computational expense of RCM is costly for developing countries. Therefore, the metamodel of RCM for the specific area is eager to be designed to support decision making. This research introduces novel development and application of Convolutional Neural Network (CNN) to precipitation downscaling over Bago River Basin in Myanmar using in situ data, ERA- Interim (reanalysis data) and Coupled Model Intercomparison Project 5 (CMIP5) as input data to perform Pseudo Global Warming Method (PGWM). The regional climate model Weather Forecasting Research (WRF) is utilized in the research to conduct dynamical downscaling. Data source: • In situ data • ERA-Interim (1995-2015) • Weather Forecasting Research (WRF) 0 100 200 300 400 500 600 700 800 JAN FEB MAR APL MAY JUN JUL AUG SEP OCT NOV DEC Average Monthly Precipitation 1995-2005 Bago River Basin, Myanmar MM/Day Figure 1 – Bago River Basin and its climatology Figure 2 – Research Flowchart Figure 3 – Convolutional Neural Network Structure Figure 4 – Comparison of Ground Truth and Simulation Results Figure 5 – Rainfall-Runoff Inundation (RRI) Flood Simulation Results • Combine land cover classification from results of satellite images segmentation with flood hazard map. • Build deeper neural network architecture to enhance downscaling accuracy. • Analyze CNN outputs to improve training results and prevent model from overfitting.