Residing just to the north of the Bay of Bengal climatological precipitation maximum and at the confluence of two of the largest rivers on the planet, the Ganges and the Brahmaputra, parts of Bangladesh are flooded each year but occasionally devastating floods occur such as in 1998 when over 60% of the country was inundated for a period of over three months causing economic and societal devastation. Even the regular yearly flooding occurs irregularly in region and timing. With only river discharge data at the borders of the country, Bangladesh has only been able to issue 2-day flood forecasts for central Bangladesh.

Three years ago we embarked on a program to forecast floods and rainfall over a large range of time scales. We envisioned a system that would provide a "broad brush" seasonal outlook of river discharge into Bangladesh on time scales of 1-6 months updated every month (derived from ECMWF climate model forecasts), a "finer grain" 20-25 day forecast of Brahmaputra and Ganges discharge (using a newly developed physically-based Bayesian statistical scheme), and a "fine-grained" 1-6 day discharge forecast (using ECMWF model ensemble output). This summer, the three-phase forecast system was introduced on an experimental basis. To date the results of the forecasts are very encouraging.

We provide an overview of the three schemes and critically evaluate the forecasts made during the summer of 2003. We will present retrospective of the 2003 summer monsoon season and isolate the impacts of remote and local influences on the three time scales that we forecast. We also discuss the importance of close coordination between the providers of the forecasts and the users of the forecast. We will reiterate the importance of the development of an infrastructure within the country where the forecasts will be used.