Search for: All records

Abstract Accurately delineating both pluvial and fluvial flood risk is critical to protecting vulnerable populations in urban environments. Although there are currently models and frameworks to estimate stormwater runoff and predict urban flooding, there are often minimal observations to validate results due to the quick retreat of floodwaters from affected areas. In this research, we compare and contrast different methodologies for capturing flood extent in order to highlight the challenges inherent in current methods for urban flooding delineation. This research focuses on two Philadelphia neighborhoods, Manayunk and Eastwick, that face frequent flooding. Overall, Philadelphia, PA is a city with a large proportion of vulnerable populations and is plagued by flooding, with expectations that flood risk will increase as climate change progresses. An array of data, including remotely sensed satellite imagery after major flooding events, Federal Emergency Management Agency’s Special Flood Hazard Areas, First Street Foundation’s Flood Factor, road closures, National Flood Insurance Program claims, and community surveys, were compared for the study areas. Here we show how stakeholder surveys can illuminate the weight of firsthand and communal knowledge on local understandings of stormwater and flood risk. These surveys highlighted different impacts of flooding, depending on the most persistent flood type, pluvial or fluvial, in each area, not present in large datasets. Given the complexity of flooding, there is no single method to fully encompass the impacts on both human well-being and the environment. Through the co-creation of flood risk knowledge, community members are empowered and play a critical role in fostering resilience in their neighborhoods. Community stormwater knowledge is a powerful tool that can be used as a complement to hydrologic flood delineation techniques to overcome common limitations in urban landscapes. 

Federated learning (FL) aims to collaboratively train a global model using local data from a network of clients. To warrant collaborative training, each federated client may expect the resulting global model to satisfy some individual requirement, such as achieving a certain loss threshold on their local data. However, in real FL scenarios, the global model may not satisfy the requirements of all clients in the network due to the data heterogeneity across clients. In this work, we explore the problem of global model appeal in FL, which we define as the total number of clients that find that the global model satisfies their individual requirements. We discover that global models trained using traditional FL approaches can result in a significant number of clients unsatisfied with the model based on their local requirements. As a consequence, we show that global model appeal can directly impact how clients participate in training and how the model performs on new clients at inference time. Our work proposes MaxFL, which maximizes the number of clients that find the global model appealing. MaxFL achieves a 22-40% and 18-50% improvement in the test accuracy of training clients and (unseen) test clients respectively, compared to a wide range of FL approaches that tackle data heterogeneity, aim to incentivize clients, and learn personalized/fair models. 

Abstract Studying the centennial or millennial timescale response of large rivers to changing patterns in precipitation, discharge, flood intensity and recurrence, and associated sediment erosion is critical for understanding long‐term fluvial geomorphic adjustment to climate. Long hydrographs, maintaining reliable Flow Duration Curves (FDCs), are a fundamental input for such simulations; however, recorded discharge series rarely span more than a few decades. The absence of robust methodologies for generating representative long‐term hydrographs, especially those incorporating coarse temporal resolution or lacking continuous simulations, is therefore a fundamental challenge for climate resilience. We present a novel approach for constructing multi‐century hydrographs that successfully conserve the statistical, especially frequency analysis, and stochastic characteristics of observed hydrographs. This approach integrates a powerful combination of a weather generator with a fine disaggregation technique and a continuous rainfall‐runoff transformation model. We tested our approach to generate a statistically representative 300‐year hydrograph on the Ninnescah River Basin in Kansas, using a satellite precipitation data set to address the considerable gaps in the available hourly observed data sets. This approach emphasizes the similarities of FDCs between the observed and generated hydrographs, exhibiting a reasonably acceptable range of average absolute deviation between 6% and 18%. We extended this methodology to create projected high‐resolution hydrographs based on a range of climate change scenarios. The projected outcomes present pronounced increases in the FDCs compared to the current condition, especially for more distant futures, which necessitates more efficient adaptation strategies. This approach represents a paradigm shift in long‐term hydrologic modeling.