Combining process-based and data-based approaches to predict high water levels : application to the city of Pontianak in the Kapuas River delta

The Kapuas River delta (KRD) is a low-lying marshy delta on the western coast of Borneo Island, Indonesia. In the last decades, palm oil cultivation and forest fires have encroached on the Kapuas water catchment areas, changing the Kapuas hydrological regime and triggering more intense flooding in the KRD. Meanwhile, climate change has also rendered flooding events in the delta less natural and more disastrous. As population density rises, flooding risks attributed to riverine and coastal surges must be properly assessed. The importance of this assessment for urban planning and coastal protection development cannot be overstated. However, as in most developing countries, local water managers in the KRD lack data and tools to assess flooding risks properly. Consequently, the mechanisms that drive the flooding events in the KRD remain poorly known. Creating robust forecasting to predict flooding events is challenging with the lack of observational data, even more so with limited computational resources. The first objective of this thesis is to investigate the impact of the interaction between tides, storm surges, and river discharges in the KRD, particularly in the densely-populated city of Pontianak, which lies on the delta. The interactions between the driving forces and the effects on compound flooding are simulated using a hydrodynamic model (SLIM). Then, based on the maximum water-level profile, compound flooding hazard zones along the Kapuas River are delineated. In the second phase, this thesis combines the process-based and data-based modeling approaches to tackle the issue caused by limited computational resources and observational data. The process starts with building a hydrodynamic model to run flood scenarios in the KRD. The next step is to create machine learning (ML) models and train them using the outputs of the hydrodynamic model to predict water levels and future floods. Several ML algorithms are evaluated to ensure that the ML model is robust. Rando
(AGRO - Sciences agronomiques et ingénierie biologique) -- UCL, 2022