The increasing frequency and intensity of extreme weather events due to climate change highlight the need for accurate hazard mapping for heavy rainfall. Land use and land cover (LULC) maps, along with digital elevation models (DEMs) and meteorological data, are essential to construct these hazard maps. For this purpose, LULC maps must be of high resolution to capture small hydrodynamically relevant structures and up-to-date to reflect constant changes driven by human activities. Uncrewed aerial vehicles (UAVs) are ideal for acquiring these high-resolution images because of their cost effectiveness, flexibility, and detail. However, no publicly available datasets and approaches currently provide urban LULC maps at this level of detail (<5 cm) while considering classes relevant for constructing heavy rainfall hazard maps. Therefore, this article presents a novel approach for LULC classification using high-resolution UAV imagery to enhance the generation of heavy rainfall hazard maps. We develop a comprehensive processing pipeline that includes UAV data collection, orthophoto creation, and LULC classification using advanced deep learning architectures. Our method tackles challenges such as differing between 22 LULC classes with objects of varying sizes and managing imbalanced datasets. Additionally, we create a high-resolution UAV image dataset with more than 12K annotated images. A comprehensive evaluation of 29 deep learning architectures reveals that CFNet with an EfficientNetB4 backbone achieves the best performance, with an accuracy of 0.911 and a Mean Intersection over Union (IoU) of 0.738. The results demonstrate that our approach effectively classifies hydrodynamically relevant structures, such as different roof types, street materials, and drainage facilities. This work establishes a solid foundation for future research and practical applications in the automated generation of accurate heavy rainfall hazard maps, ultimately aiming to improve risk management strategies in the context of climate change.