RAINSAT: Precipitation measurements and nowcasts for Africa

Flash floods are yearly returning problem in many African cities due to heavy rainfall and poor drainage. This problem is exacerbated by rapid urbanization and climate change. Early warning is an effective method for damage reduction. Early warning for urban (flash) floods requires real-time rainfall observations and short-term forecasts (nowcasts) at high spatial and temporal resolutions. Unfortunately, weather stations do not offer sufficient spatial coverage in many African countries. Most satellite observations, such as GPM, are only available with a few hours delay, the so called latency. For flood events, which typically occur within a few hours, this information comes too late.

HKV and KNMI developed RAINSAT (, a website and app disclosing precipitation measurements for Africa every 15 minutes with a resolution of approximately 3 × 3 km and a latency of 45 minutes. HKV is currently working on a nowcasting component, predicting up to 3 hours ahead. Precipitation information with this resolution and frequency is unique to most places in Africa and certainly not available for free. RAINSAT allows for earlier warning of extreme precipitation events at higher spatial resolution!

RAINSAT discloses precipitation estimates based on infrared measurements from the Meteosat Second Generation SEVIRI instrument for free. These measurements come from a geostationary satellite, making measurements available every 15 minutes for all of Africa and Europe. Infrared measurements are also available at night; this is an advantage over images in the visual spectrum. The KNMI converts the infrared measurements into an estimate of the precipitation amount. The nowcasting model is based on Machine Learning methods and trained on the same Meteosat Second Generation SEVIRI data. This allows us to predict extreme downpours up to 3 hours ahead and to warn of possible flooding even earlier. RAINSAT information can be integrated into flood warning systems as well.

The Machine Learning techniques we use can also be of value in the Netherlands. In collaboration with KNMI and TU Delft, we are investigating the application of these techniques to radar measurements of precipitation. It is expected that these techniques can predict the increase and decrease of rainfall intensities and the circulation of a shower better than current methods. We try to predict precipitation in the short term even more accurately!