Given the high population density, high-resolution observing and modeling capabilities are necessary for prediction of flash floods in urban areas. Continuing urbanization and climate change put such areas in a particularly vulnerable position where even a small-scale but intense rainfall event can cause deadly flash floods and extensive damages. For high-resolution observing and modeling of large urban areas, the use of weather radar and distributed hydrologic modeling is a natural progression. Quantitative precipitation estimates (QPE) from radars, however, are subject to various sources of error. High-resolution distributed modeling is subject to nonlinear growth of error due to errors in QPE and in model parameters and structures. Widely varying imperviousness in land cover, and density, capacity and complexity of storm drain networks present additional challenges. In this presentation, we describe an ongoing effort for real-time flash flood forecasting using the CASA (Collaborative Adaptive Sensing of the Atmosphere) radar network, the potential role of data assimilation in bridging the information gap, and plans for integrative sensing and prediction of urban water in DFW.