Information on where, when and how much rain will fall in the coming hours is crucial for the infrastructure, agriculture and water management sectors. The most disruptive events, however, are often the most challenging to forecast. These events are generally high-intensity rain showers, which can develop and dissipate in a timeframe of half an hour.
Current practice is to use numerical weather prediction models to forecast such rainfall events. These models are computationally expensive and therefore run on super computers. As a result of this, a new model run starts every couple of hours and then provides a forecast for the coming days. In the couple of hours between the new model runs, severe rainfall events can form and may therefore be totally missed by weather models.
Another approach is to use the most recent observations of remotely sensed rainfall fields to provide a better rainfall forecast up to a couple of hours ahead. This method is called nowcasting and uses a statistical extrapolation of the shape, rainfall intensity and movement of rainfall fields. This way of forecasting rainfall often outcompetes weather models for short-term rainfall forecasts (less than three hours ahead), but is not useful for longer forecasts due to the absence of physics in the nowcasting algorithms.
The method was developed to be used with observations from ground-based weather radars, because these radars give frequent updates, every five minutes for example. This rapid update cycle makes them very suitable for nowcasting of rainfall. Although nowcasting is a promising technique, not every region on earth is covered by the observations of a weather radar. Hence, this limits the applicability of radar rainfall nowcasting to specific parts of the world.
Commercial microwave links (CMLs) from cellular communication networks may offer an alternative solution. Although the purpose of CMLs, which connect telephone towers, is not to measure rainfall, the signal from one tower to the other gets attenuated by, among other factors, rainfall. Mobile network operators keep track of this attenuation in order to have insights into the reliability of their network.
For meteorologists, the information about the signal attenuation of these CMLs can help determine the amount of rain that has fallen based on the attenuation in the signal. Thus, what may look as a burden for the telecom industry, actually provides a wealth of information for the meteorological sector.
Estimating rainfall and deriving country-wide rainfall maps from the CML data is an increasingly well-studied topic, but researchers from Wageningen University & Research, Deltares and Royal Netherlands Meteorological Institute (KNMI) have gone a step further and used the rainfall estimates from the commercial microwave links for rainfall nowcasting.
A test on 12 summer events for the Netherlands showed that these nowcasts have a performance that is quite comparable to rainfall nowcasts with weather radar data, especially for high-intensity rainfall.
The study also showed some limitations to the short-term rainfall forecasts based on the CML data. Whereas radar gives observations for every square kilometer, also above water bodies, CMLs are not homogenously spread over a region and are even absent above large water bodies. The CML network density is higher in urban areas than in rural areas. This means that rainfall estimates are generally more accurate in urban areas than in areas with a sparser CML network, which in turn also affects the nowcasts. Moreover, the weather radar refreshes every five minutes, while the CML data was recorded every 15 minutes in this study. Increasing the frequency to every five minutes is, however, possible and could improve the nowcasts.
The way forward is probably to combine the data sources into a rainfall product, which is then used as input for the nowcasting algorithm. A major advantage of the CML data is that the observations take place close to the ground, whereas other remote sensing techniques, such as radar, observe rainfall at higher altitudes. The CML data could complement the radar rainfall nowcasts, when radars are present. On top of that, in regions in the world where there are no radars or even rain gauges, but which have cellular communication networks, the short-term rainfall forecasts with the CML data can offer an alternative. There, the limitations of nowcasting with CML data could be overcome by using satellite data as a complementary data source.
The study continues now for a couple of other regions: Sri Lanka, Nigeria and Papua New Guinea. In addition, the added value of these short-term rainfall forecasts also lies in applications, such as water management. These applications are already being investigated for radar rainfall nowcasting, but an important next step would be to test the CML nowcasting for these applications.