Some major advances in both theory and operations have been driven by new technology: Weather Surveillance Radar 1988 Doppler (WSR-88D). Known as the Next Generation Weather Radar (NEXRAD) system, a network of about 160 of these radars will be deployed by the NWS during the 1990s to cover all contiguous states. With a rainfall detection range of about 230 km and an array of products generated from the return signal, these radars are changing the view of operational forecasters on detection of local storms, estimation of rainfall coverage and intensity, and ultimately forecasting headwater floods and flash floods.
Potential gains from using weather radar in flood
forecasting have been studied by James et al. [1993]. A
distributed rainfall-runoff model was applied to a 785 km
basin equipped with two rain gauges and covered by a radar. Data
recorded during a past storm provided inputs for computing three
flood hydrographs from (1) rainfall recorded by rain gauges, (2)
radar estimates of rainfall, and (3) combined rain gauge
measurements and radar estimates. The hydrograph computed from
the combined input was the closest to the observed hydrograph.
Advantages and caveats of using radar rainfall estimates in operational forecasting have been well illustrated in a case study by Amburn and Fortin [1993]. A series of summer thunderstorms developed over northeastern Oklahoma. Reports from 11 rain gauges (only two of which were located inside a drainage basin) indicated maximum storm total of 4.0 inches, and basin average rainfall of 1.44 inches, producing a forecast of the river stage below the flood stage. On the other hand, radar estimated a maximum of 9.1 inches, and an average of 5.2 inches, resulting in a forecast of a record flood. The NWS forecasters subjectively combined the two sets of estimates, taking into consideration reports from spotters which suggested that the storm centers missed the rain gauges, while hail contaminated the radar reflectivity field. The subjective estimate of basin average rainfall of 2.8 inches produced a quite accurate forecast of the rising limb of the hydrograph. The public received a timely alarm: a flood watch provided 2.5 hours of lead time to the flood stage, and a warning provided 13.5 hours of lead time to the 7 feet crest.
This experience is symptomatic of the current technology: while the weather radar provides superior information about the spatial and temporal resolution of the rainfall event, estimates of rainfall accumulation may have phenomenal biases. Novel debiasing techniques should be researched. From an operational point of view, such techniques should recognize that (1) the bias in radar estimates is conditional on spatially varying storm characteristics, (2) rain gauges may be too sparse to properly debias the radar estimates, (3) valuable additional information may be in the qualitative form, such as verbal reports by spotters, and therefore (4) the forecaster's judgment remains vital in assigning credibility weights to information sources. Ergo, techniques for combining information from multiple sources should leave considerable latitude to the forecaster. Advances in systems engineering could be harnessed in order to amalgamate the power of statistical combining methods with judgmental capabilities of forecasters into a next generation of human-computer estimation systems.