The weather radar - Utilisation - Forecasting
By its remote detection of precipitation, and the possibility to observe the rain motion, the weather radar is a privileged tool for short range rainfall forecasting (range below two hours, called "nowcasting").
Whatever the forecasting method used, forecasting assume strong hypotheses concerning the quality of radar rainfall measurements and the evolution of rain fields. The limits of rainfall forecasting are directly linked to the accuracy of these assumptions. We can list a few of them:
· The measurement of the past and current values of rainfall is of good quality: if this hypothesis is not true, and if the treatment of the main sources of error in the radar measurement is not efficient, the quality of the rainfall forecasts is directly reduced.
· The rain motion is well estimated by the analysis of the last radar images, particularly the two last images.
· This motion will persist in the future during all the forecast time range.
· The rainy activity will remain unchanged in the future during all the forecast time range. This seems to us the more important limitation to the radar rainfall forecasting, and this limit depends on: the type of rain; the spatial size of the forecasts (mean rainfall over a great basin, punctual forecast); the type of variable forecasted.
Therefore, it is important in first stage to verify that the area of interest and the zone upstream for the rain motion, is well covered by the radar network available. The maps of measurement quality of the French weather radar network could be used first. But we have demonstrated [Faure et al. 2001, in French], that the accuracy of the forecasts for a basin not only depends on the location of the basin from the radar, but also on the trajectory of the rain compared to the location of the main effects of the source of errors in the radar measurement for these images. Also, for convective rain, the best forecasting was not realised for the same rain events for different locations of the basin (with the same radar images and the same forecast software).
To be simple, we could estimate for no convective rain, that the quantitative precipitation forecasts (QPF) on a basin may be of good quality for time range exceeding 1h30, even for small basins, and the quality of the radar data is the principal limitation.
For convective rain, including small heavy rain cells evolving very quickly (life time shorter than 30 to 45 minutes), the smaller the size of the basin is, the more the quantitative forecasting is limited. For very small catchment areas, the limit of the forecasting is controlled by the behaviour of each intense rain cell, and for areas of few km˛ the limit for the acceptable forecast range can be reduced to few minutes. For spatial mean forecasts over large basins, the limit of the forecasting is controlled by the behaviour of large rain structures, and local variations are filtered by the spatial integration.
Finally, the type of variable forecasted is important: if it is difficult to forecast precisely the rainfall value from T+30mn to T+45mn, the forecasting of the cumulated value for T+0to+45mn is generally more reliable (but uniquely by the average of the good quality at short range and the poor quality at long range), and the forecasting of the possibility of rain (possibly the date of the rain beginning) is even more reliable.
The strategy for use rainfall forecasting, therefore, must be adapted to the actual need of information, the forecast lead-time required, the surface associated to the forecast values, and the limits above mentioned. For a user who is not an expert, we can hardly recommend to be assisted by an expert, and before an operational use, to realise validation studies analysing the quality of the forecasts provided by a software with real data. These studies should compare radar forecasts and corresponding radar observation, as well as radar forecasts and rain gauge data.