ANNOUNCEMENTS
This dissertation explores how urbanisation affects flooding in Assam, India, over the period 2000 - 2024. It emphasises the influence of proxy indicators, such as electricity usage, urban road development, drainage system expansion, and changes in land use on the frequency of floods. The goal is to quantify these relationships and derive explainable insights relevant to policy through a data-driven modelling framework.
The research begins with the creation of a combination of statistical and machine learning models, such as ARIMA, ARIMAX, Generalised Linear Models (GLMs), Generalised Additive Models (GAMs), and Random Forests, to predict flood occurrences based on urbanisation proxy indicators. Model performance is evaluated using accuracy metrics, and interpretability is enhanced through SHAP summary plots, SHAP-over-time heatmaps, rule extraction, and waterfall additive breakdowns.
The results show that higher electricity consumption and increased built-up land consistently raise flood risk, while the roles of drainage infrastructure and GDP are more intricate or vary by model. Additionally, a supplementary survey of 70 residents from Guwahati strengthens these findings, capturing local perceptions of risk, preparedness behaviours, and indices of social vulnerability. The convergence of model predictions and community perceptions emphasises significant deficiencies in institutional preparedness and urban planning.
This research concludes with a series of data-driven policy suggestions, promoting improved drainage maintenance, community preparedness initiatives, and the creation of a centralised, high-resolution data framework. These findings are intended to inform climate-resilient flood management approaches in Assam and other similarly urbanising areas.
Keywords: Assam Flood Risk Assessment; Urbanisation Flood Drivers; ARIMA; GAM; Random Forest; SHAP; Drainage & Road Infrastructure; Land Use; Policy Recommendations.
