Science news

FMI's researchers publish about 350 peer-reviewed articles annually.

In Science News we publish current information about FMI's studies on the weather, the sea and the climate.

« Back

Open-source software package can be used to produce radar-based precipitation forecasts

Open-source software package can be used to produce radar-based precipitation forecasts

Reliable rainfall forecasts facilitate proviving early warning of floods and other water-related hazards. The pysteps software package developed by a multinational team provides an open-source solution for this purpose.

The forecast model of pysteps is based on the STEPS model developed in the early 2000s in Australia and the United Kingdom. The model produces a precipitation forecast in four stages: 1. determination of the motion field from consecutive radar measurements, 2. extrapolating the precipitation patterns along the advection field, 3. a time series model for the temporal evolution of precipitation intensities and 4. stochastic simulation, which produces an ensemble of alternative forecasts and thus provides information about uncertainties.

Modeling the temporal evolution and the uncertainties are done in multiple spatial scales. This is because the predictability of precipitation depends on scale: larger structures (such as rainfall bands) have longer lifetime than smaller ones (such as convective cells).

Free use for research, weather services and applications

Pysteps has been implemented in Python. The software has been published under the BSD license, which allows free use for researchers, as well as operational weather services and commercial applications.

"The software has been designed in a modular way so that different components are easily interchangeable, which makes it a well-suited tool for researchers. On the other hand, the software can be installed via the Anaconda distribution, which together with the open-source license and comprehensive documentation makes deploying the software very easy", says the lead author Seppo Pulkkinen at the Finnish Meteorological Institute (FMI).

Development of software packages for radar-based short-term precipitation forecasting (nowcasting) started already in the 1990s. Until now, the software has been proprietary and the development has been mainly done within national weather services.

"The pysteps initiative started from a meeting with Swiss colleagues. They also expressed their interest to development of a publicly available software package developed by an international community", says Seppo Pulkkinen.

The pysteps development also resulted to a scientific article that was published in Geoscientific Model Development. The article describes the software package and the new STEPS model that brings several improvements to the original one. The use of the software is illustrated with several examples. In addition, the reliability of the nowcasts produced by the STEPS model is assessed by using radar data from several countries.

"Pysteps plays a key role in the ULJAS project that is aiming at renovation of the operational FMI nowcasting system. It is planned to replace the old TULISET and RAVAKE software packages that are based on a forecasting model developed in the early 2000s in the Finnish Meteorological Institute. In addition, the meteorological institutes in the Netherlands and India have expressed their interest to operational deployment of pysteps", says Seppo Pulkkinen.

The core development team consists of people from Finland, Switzerland, Canada and Australia. Most of the development was done during Seppo Pulkkinen's research visit to MeteoSwiss during summer 2018, and the development has actively continued since then.

Seppo Pulkkinen was funded by the Finnish Academy of Science and Letters and Finnish Meteorological Institute. His coauthors were funded by the Swiss National Science Foundation and MeteoSwiss. The grant for the lead author covers a two-year period of working abroad.

Further information

Researcher Seppo Pulkkinen, Finnish Meteorological Institute,

The article published in Geoscientific Model Development is available at

More information about pysteps at

Science news archive