A framework for quantifying climate-informed heavy rainfall change: Implications for adaptation strategies

Abstract

To understand the influence of climate change on heavy rainfalls and reduce the consequential multidimensional risks, we develop a climate-informed and adaptation strategies-related framework by using the information on heavy rainfalls and various socioeconomic factors. For this purpose, we firstly quantify the spatiotemporal characteristics of heavy rainfalls with various durations (1 h to multiple days) and return periods (2-year to 50-year) for the flood-prone country Cambodia, as a case study, during the historical period (1980-2005), mid-century (2040-2065), and late-century (2070-2095), using the latest three hourly climate model datasets under RCP 8.5 and 1 hourly ERA5 reanalysis datasets. A novel conditional artificial neural network (CANN) model is employed for temporal disaggregation to obtain the monthly maximum of 1 hourly rainfall in the future periods and subsequently, a zero-inflated generalized extreme value function (ZIGEV) is applied for extreme value analysis (EVA) to obtain rainfall intensity with different return periods. Secondly, the province-level flood risk change maps are developed based on a novel flood risk change index. The combination of CANN and ZIGEV performs better in EVA than traditional approaches by reducing the uncertainty from the stationarity assumption of temporal disaggregation and bias in the disaggregated rainfall. Rainfall intensity is projected to increase more in higher return periods and shorter durations towards the late-century, predominantly over Southern and Central Cambodia. Projected rainfall intensity-duration-frequency (IDF) curves in the capital city, Phnom Penh, reveal that the occurrence frequency of heavy rainfall in a given duration (e.g., 48 h) is likely to become ~10-fold in the mid-century. Results of province-level flood risk change maps indicate that Southeastern and Northwestern regions should be prioritized for employing adaption strategies. Our results will assist the policymakers in further mapping the flood susceptibility and vulnerability in different spatiotemporal scales across various communities and localities in the country and beyond.

Keywords: Adaptation strategies; Artificial neural network; Climate change; Conditional temporal disaggregation; Future flood risk change map; Heavy rainfalls.