Flooding is a growing problem in the UK. It has a significant effect on residents, businesses and commuters in flood-prone areas. The cost of damage caused by flooding correlates closely with the warning time given before a flood event, and this makes flood monitoring and prediction critical to minimizing the cost of flood damage. This paper describes a wireless sensor network for flood warning which is not only capable of integrating with remote fixed-network grids for computationally-intensive flood modeling purposes, but is also capable of performing on-site flood modeling by organising itself as a 'local grid'. The combination of these two modes of grid computation-local and remote-yields significant benefits. For example, local computation can be used to provide timely warnings to local stakeholders, and a combination of local and remote computation can inform adaptation of the sensor network to maintain optimal performance in changing environmental conditions.