Nowcasting

Nowcasting is forecasting the weather, and the water response to it, for the next few minutes up to about six hours. Over such short periods, the latest observations carry more information than a full weather model, which needs time to settle and may place storms in the wrong spot. Most rainfall nowcasts start from weather radar. The method tracks how rain cells moved over the last few radar images, estimates a motion field, and moves the current rain field forward along it. This is called advection or extrapolation. Better methods also let cells grow and decay, and produce ensembles to show uncertainty. Satellite images, lightning data and rain gauges add more information. Machine learning methods trained on years of radar images are also being tested, with mixed results for heavy, fast-developing storms. In water, nowcasts feed urban flood warnings, flash flood alerts for small, steep catchments, and real-time control of sewers and stormwater systems. For example, a sewer operator can use a one-hour rain nowcast to start emptying storage tanks before a storm arrives, reducing overflows. Small urban catchments respond in minutes, so even 30 to 60 minutes of warning has value. What it needs: frequent radar scans (often every 5 minutes), radar rainfall adjusted against gauges, and a fast link to the hydrologic or hydraulic model. Limits: skill falls quickly with lead time, often within one to two hours, because storms form, grow and die in ways that extrapolation cannot follow. New convective cells that do not yet show on radar are missed. Radar has its own errors from beam blocking, the bright band and signal loss in heavy rain. Beyond a few hours, numerical weather models usually do better, and many systems blend the two.