An application of Sparse Variational Gaussian Process with Bernoulli likelihood for flood inundation risk mapping

As climate change intensifies, urban flooding has become a growing threat to cities worldwide, especially in low-lying, densely populated areas. Accurate flood risk prediction is essential for disaster readiness, yet it remains a challenge due to the complex, dynamic nature of extreme weather patterns. This study applies Gaussian Process modelling, a powerful probabilistic method capable of capturing spatial and temporal correlations, to forecast flood inundation risks in Semarang, Indonesia. By integrating diverse data sources, such as weather station records, satellite data, and aerial imagery, the model generates detailed flood risk maps for areas prone to recurrent inundation. Semarang, where up to 40% of lowland regions are affected annually, serves as a critical testbed. The resulting maps provide local authorities with actionable insights to identify high-risk zones, allocate resources, and implement more targeted mitigation strategies. This approach demonstrates how advanced machine learning techniques can enhance urban resilience and inform proactive policymaking in the face of climate uncertainty.