A document by Kexin Liu. Click on the document to view its contents.

Urban drainage systems are facing major challenges with rapid urbanization and climate change, especially for developing countries. Green infrastructure (GI) is a natural-based solution expected to reduce flooding and help in water pollution control. Notwithstanding multiple research have discussed the contribution of GI to climate change adaptation. The efficiency of GI to extreme events in the context of the more frequent extreme precipitation events has received limited attention in the literature. This study aims to quantify the impact of historical and future extreme rainfalls on overflow flooding, pollutant transport, and GI’s potential in flood control by taking Phnom Penh City, the capital of Cambodia, as a case study. Firstly, we predicted the history of sub-daily extreme rainfall (1986-2005) by disaggregating outputs of various regional climate models based on the Artificial Neural Network (ANN) method. Secondly, we generated the future intensity-duration-frequency (IDF) curves based on sub-daily extreme rainfall (2026-2045) and input the design rainfall time series (based on the scenarios of RCP4.5 and RCP8.5) into a hydrological model (PCSWMM) to investigate the impact of climate change on urban drainage systems, including the flooding and pollution (suspended solids). The model successfully captured the variation of overflow flooding and transport of pollutants. Thirdly, we introduced four mitigation measures of GI and simulated their effectiveness against climate change with different extreme rainfall events. The results indicate that future climate change will increase the risk of overflow flooding and more pollutants diffuse on urban surfaces. The GI is an effective method to mitigate its impact, but the performance decreases with rainfall intensity. The effect of increased rainfall is consistent for different GI. PP is most effective in relieving the pressure of extreme rainfall for total flood,  peak flow reduction, and pollutant removal. Although GR performs reasonably well in flood control, it is the least effective in pollutant control. It can be seen that more research is needed in the implementation and optimization of GI.

Extreme rainfall can be calamitous to the ecosystem, life, society, and economy through rapidly developing (flash) floods and is likely to intensify in a warmer future climate. Such intensification is however less well understood for the rainfall in short durations (e.g., hourly; 1h) due to the coarse time-scale of climate models. This study proposes an artificial neural network (ANN) model for disaggregating coarser time-scale (i.e., 3h) rainfall datasets to finer time-scale (i.e., 1h) extreme rainfall (i.e., annual maximum series (AMS)), targeting a data-scarce county like Cambodia by using the 1h rainfall dataset and multiple meteorological covariates datasets (e.g., temperature, wind velocity, and surface latent & sensible heat flux (SLHF&SSLF)) provided by ERA5 reanalysis products. The ANN model was trained by using the information of extreme rainfall events extracted from this 1h rainfall dataset and of the associated simultaneous weather conditions signified by specific combinations of these meteorological covariates. The rationale is that future extreme rainfall patterns will resemble the historical extreme rainfall patterns if similar weather conditions exist during the extreme rainfall events. Covariate importance analysis shows that the most important covariates for the disaggregation are SLHF&SSLF and wind velocity. The proposed ANN model reproduced the observed 1h AMS satisfactorily, with R2 of 0.93 and mean absolute percentage error (MAPE) of 6.1%, averaged for the study area. This ANN model is flexible enough to be extended to other time scales (e.g., daily to sub-hourly) and can be used for similar studies globally. Future work will consider more meteorological covariates, which can be both provided by the ERA5 reanalysis products and climate models, as the predictors.