ABSTRACT Roads are ecologically sensitive infrastructure that are highly susceptible to flooding, which can impair their functionality, serviceability, and durability. Existing assessment methods are often subjective or lack the ability to incorporate domain‐specific knowledge. Therefore, approaches that integrate expert‐driven prioritization with data‐driven optimization are needed to minimize subjectivity and improve model robustness. This study evaluated the historical and future flood vulnerability of the Limuru—Mai Mahiu—Narok road in Kenya. Using rainfall, land use/land cover, NDVI, slope, curve numbers, topographic wetness index, river density, topographic factor, landforms, soil texture, sediment transport index, and GPS data, a GIS‐ensemble framework combining Multi‐Criteria Decision Evaluation (MCDE) and Principal Component Analysis (PCA) was applied. Flood vulnerability was assessed for 1991, 2002, 2011, and 2021, while future susceptibility for 2030 was predicted using Cellular Automata–Markov (CA–Markov) modeling. Across all models, the hinterlands were consistently dominated by moderate to high vulnerability levels, covering about 62%–80% of the region. The year 2002 was the most flood‐prone, with more than 47% of the hinterlands classified as highly to extremely highly vulnerable, whereas 1991 and 2021 showed relatively lower susceptibility. Road‐specific analysis indicated that 40.89–44.80 and 30.97–38.53 km of the road fell within moderate and high vulnerability classes, with up to 73 km affected by high to extremely high vulnerability in 2002. Validation of the 2021 map produced 82% overall accuracy and a Kappa coefficient of 0.73. CA–Markov validation yielded strong performance (Kno = 0.72; Klocation = 0.73; Kstandard = 0.65). The 2030 projection shows that nearly 88% of the hinterlands and about 104 km (91.4%) of the road would be exposed to moderate to extremely high flood risk. These results highlight escalating flood threats and the need for targeted mitigation and climate‐resilient infrastructure planning.
Njeru et al. (Tue,) studied this question.