Predicting future flood hazards in a changing climate requires adopting a stochastic framework due to the multiple sources of uncertainties (e.g., from climate change scenarios, climate models, or natural variability). This requires performing multiple flood inundation simulations which are computationally costly. Data-driven models can help overcome this issue as they can emulate urban flood maps considerably faster than traditional flood simulation models. However, their lack of generalizability to both terrain and rainfall events still limits their application. Additionally, these models face the challenge of not having sufficient training data. This led state-of-the-art models to adopt a patch-based framework, where the study area is first divided into local patches (i.e., broken into smaller terrain images) that are subsequently merged to reconstruct the whole study area prediction. The main drawback of this method is that the model is blind to the surroundings of the local patch. To overcome this bottleneck, we developed a new deep learning model that includes patches' contextual information while keeping high-resolution information of the local patch. We trained and tested the model in the city of Zurich, at spatial resolution of 1 m. The evaluation focused on 1-hour rainfall events at 5 min temporal resolution and encompassing extreme precipitation return periods from 2- to 100-year. The results show that the proposed CNN-attention model outperforms the state-of-the-art patch-based urban flood emulator. First, our model can faithfully represent flood depths for a wide range of extreme rainfall events (peak rainfall intensities ranging from 42.5 mm h-1 to 161.4 mm h-1). Second, the model's terrain generalizability was assessed in distinct urban settings, namely Luzern and Singapore. Our model accurately identifies water accumulation locations, which constitutes an improvement compared to current models. Using transfer learning, the model was successfully retrained in the new cities, requiring only a single rainfall event to adapt the model to new terrains while preserving adaptability across diverse rainfall conditions. Our results suggest that by integrating contextual terrain information with local terrain patches, our proposed model effectively generates high-resolution urban pluvial flood maps, demonstrating applicability across varied terrains and rainfall events.

Fast urban pluvial flood models are necessary for a range of applications, such as near real-time flood nowcasting or processing large rainfall ensembles for uncertainty analysis. Data-driven models can help overcome the long computational time of traditional flood simulation models, and the state-of-the-art models have shown promising accuracy. Yet the lack of generalizability of data-driven urban pluvial flood models to both unseen rainfall and distinctively different terrain, at the fine resolution required for urban flood mapping, still limits their application. These models usually adopt a patch-based framework to overcome multiple bottlenecks, such as data availability and computational and memory constraints. However, this approach does not incorporate contextual information of the terrain surrounding the small image patch (typically 256mx256m). We propose a new deep-learning model that maintains the high-resolution information of the local patch and incorporates a larger surrounding area to increase the visual field of the model with the aim of enhancing the generalizability of data-driven urban pluvial flood models. We trained and tested the model in the city of Zurich (Switzerland), at a spatial resolution of 1 m, for 1 h rainfall events at 5 min temporal resolution. We demonstrate that our model can faithfully represent flood depths for a wide range of rainfall events, with peak rainfall intensities ranging from 42.5 to 161.4 mmh-1. Then, we assessed the model's terrain generalizability in distinct urban settings, namely, Lucerne (Switzerland) and Singapore. The model accurately identifies locations of water accumulation, which constitutes an improvement compared to other deep-learning models. Using transfer learning, the model was successfully retrained in the new cities, requiring only a single rainfall event to adapt the model to new terrains while preserving adaptability across diverse rainfall conditions. Our results indicate that by incorporating contextual terrain information into the local patches, our proposed model effectively simulates high-resolution urban pluvial flood maps, demonstrating applicability across varied terrains and rainfall events.

Climate change is expected to affect precipitation, streamflow, and sediment transport. These changes are particularly relevant in mountainous environments that play a crucial role in water resources and sediment supply for downstream reaches. We investigated the impact of climate change on hydrology and geomorphology in the upper Emme catchment (127 km2) in the Swiss pre-Alps by simulating its hydromorphological response to present climate and three climate scenarios at the end of the century using the distributed CAESAR-Lisflood landscape evolution model. The mean seasonal changes, intensification of short-duration rainfall extremes, and snow processes were explicitly modeled. The results highlight the importance of the intensity of rainfall events to predict sediment transport at the outlet, while changes to snow processes are predominant to understand the seasonal hydrological shift. For the highest emission scenario (RCP8.5), the sediment yield at the outlet increased by 6% despite a reduction in precipitation by 7% compared to the present climate, as a result of heavy precipitation intensification. On a seasonal scale, discharge increased in winter while it decreased in spring in all scenarios due to changes in snow accumulation and melting. Furthermore, we found that erosion and deposition will change spatially by the end of the century, with a shift from erosion- to deposition-dominated valleys.

Climate change is affecting the hydromorphological system. In many places, changes in sediment dynamics are closely correlated to changes in precipitation frequency and magnitude. However, in nivo-pluvial regimes, the hydromorphological response to climate is more challenging to predict as it is not only the amount and occurrence of precipitation that is changing. The changes in precipitation type (i.e., snow or rain), snow accumulation, and snowmelt rates will also have a significant effect on the catchment net precipitation (composed of direct precipitation plus snowmelt contribution), and this may affect overland flow, erosion, stream discharge, sediment transport, and deposition. We investigated the impacts of climate change on hydrology and geomorphology in a small catchment (Emme, 127 km2) located in the Swiss pre-Alps by simulating the difference between the hydromorphological response to net precipitation in the present climate and in three climate scenarios at the end of the century using the CAESAR-Lisflood landscape evolution model. For the most extreme climate scenario (RCP8.5), simulations showed that despite the reduction in net precipitation (by 7 %) and discharge (by 4 %), sediment yield at the outlet of the catchment increased by 6 %. This is not only because precipitation falls more as rain than snow during the cold months, but also because heavy precipitation is expected to intensify. On a seasonal scale, we found that the amount of net precipitation, discharge, and erosion will increase in winter at the end of the century, while it will decrease in spring. In all three climate scenarios, net precipitation is projected to decrease in summer, but sediment yields may both decrease or increase. Autumn is the season with the greatest changes in erosion, while net precipitation remains constant or only slightly increases. Furthermore, we found that erosion and deposition patterns are changing spatially, with more erosion in mid-elevations and more deposition in valleys. Although our results are specific to the study site, we expect similar trends in other catchments of the pre-Alpine region.