Real-time Flood Forecasting
We’ve learned to predict typhoons. What is required to predict the floods they bring?
Extension of Forecast Time
The model described so far allows fairly reliable predictions for several forecast hours. But in the face of dangerous floods, earlier warning is immensely valuable. To extend the length of forecast time, a few special techniques, mostly stochastic, some deterministic, have been developed. For example, for the upstream boundaries, where the influence of the tides is normally absent, we have devised a rainfall-stage model that relates the rate of rainfall upland/upstream directly to the rate of change of the river stage at a boundary point in the river channel, which is a more direct evaluation for our purposes than the more commonly employed rainfall-discharge (or rainfall-runoff) model, which relates rainfall to the discharge at the boundary point. We can, in turn, link the upper end of the rainfall-stage model to another such model, named the typhoon-rainfall model, for yet another extension. In this case, we fetch key parameters about the approaching typhoon, such as the location of its center, its radius and maximum near-center wind speed, and we correlate these parameters with the expected rainfall at the upland rain-gage stations. Using these techniques, we are able to extend the flood forecast to as much as 24 hours.
For the river-mouth boundary, a highly accurate physics-based model was employed, one using harmonic analysis to assemble a sufficient number of astronomical tide components. (Timing of tides according to tide tables may be quite precise, but the tables are accurate only for specific locations. Accurate data for other locations often require interpolation, the more precise the better.) The tide data may be extended by a stochastics-based storm-surge model if the storm surge is expected to be significant, resulting in composite boundary stage data.