Distributed modeling has been advanced significantly during the last decade, and has become a hot topic in the hydrologic research and forecasting communities. However, a number of critical questions need to be answered before distributed watershed models become a basic building element of real-time forecasting systems. Our paper identifies the most critical scientific and operational issues that must be overcome for distributed models to become operationally feasible. It establishes a framework, the Distributed Model Intercomparison Project (DMIP), in which research models are exposed to operational-quality data typically used in hydrological forecasting. This exposure serves to highlight the need for continued improvements in the estimation of rainfall, as well as providing researchers with a rigorous opportunity to further improve their models.

The paper describes the DMIP, a scientific comparison of many new advanced methods for river and flash-flood modeling and forecasting. The purpose of the comparison was to provide guidance to the research program of the National Oceanic and Atmospheric Administration's National Weather Service (NOAA/NWS) as to which model or modeling approach might be best suited to accomplish the NOAA/NWS water resources forecasting mission. Rather than evaluating one model at a time, the NOAA/NWS's strategy was to design, lead, and participate in a multi-institutional comparison of leading approaches from around the world, inviting scientists to test their models on common data sets.

Twelve groups from Denmark, New Zealand, China, Canada, and the US participated in DMIP. Based on the success of DMIP, we are now planning for another model intercomparison (named DMIP 2), which will cover areas in the Sierra Nevada mountains. We hope to launch DMIP 2 by October 1, 2005.

As far as we know, DMIP was the first published intercomparison of distributed and lumped hydrologic models. Lumped models basically average the rainfall and watershed properties such as soil types, land cover, land use, and topographic features. On the other hand, distributed models account for the spatial variability of precipitation and watershed properties. The NOAA/NWS received valuable guidance for its research directions. Moreover, DMIP benefited the greater scientific community by allowing research-quality models to be tested in a pseudo-operational environment, providing valuable guidance to the developers.