Forecasting rainfall amounts at high resolutions in space and time has a variety of applications ranging from flood forecasting to transportation, but it is among the most challenging problems in hydrology and meteorology. Numerical models of the weather with better physics and enhanced resolutions, and high-quality measurements from remote sensors such as weather radar, offer a window of opportunity in this area. However, the use of available process physics, traditional statistical models, and data mining tools such as neural networks have not resulted in significant improvement over the years. This article describes how a hybrid model, which intelligently combines all these approaches to make the best possible use of the available information, was able to generate improved rainfall forecasts.