Identification of summer hailstorm from MSG data using Python

Severe storms with hail precipitation are one of the most common meteorological risks in Europe. They can provoke grave economic losses and even affect the population. Identifying these types of episodes is one of the most habitual demands of the population in general, and specifically, for risk management organisms. The Group for Atmospheric Physics (GAP) from the University of León (ULE), has developed a model that allows for identification of hailstorms in real-time (nowcasting), using information gathered from a Meteostat Second Generation satellite (MSG).

The new generation of geostationary MSG satellites includes sensors that have a spatial-temporal and spectral resolution that is better than that of its predecessors. One of these sensors is a radiometer named Spinning Enhanced Visible and InfraRed Imager (SEVIRI). This sensor has twelve spectral channels with a sampling distance of 3 km and also has a high-resolution visible (HRV) channel with a sampling distance of 1km. The temporal resolution of SEVIRI is 15 minutes. These characteristics translate to a substantial improvement in following meteorological phenomenon with fast rates of growth and a relatively short lifetime, such as, for example, flash-floods, or convective storms, which makes MSG an extremely useful tool in following this type of phenomenon.

The information gathered by the SEVIRI radiometer from the MSG satellite presents a format distributed by EUMETSAT (European Organisation for the Exploitation of Meteorological Satellites) named HRIT. This information can be read and managed from Python using MIPP (meteorological ingest processing package) and MPOP (meteorological post processing package) packages. There are also tools that transform the information in an HRIT format to other more friendly formats, such as NetCDF, which can also be treated using Python, via the NetCDF4 package (python/numpy interface to netCDF library).

On the other hand, calculating hail probability in real-time using the algorithm proposed by the GAP requires various steps. The first step is the definition of the geographical area of study. It is necessary to discretize this area in order to study it. In order to do so, a two-dimensional matrix to cover it is defined. We refer to this discretized model as a grid. Each point in the grid is called a pixel. The larger the number of pixels in the grid, the greater the resolution of the results obtained will be. Additionally, the computation time will also be greater.

The next step consists of calculating hail probability in each pixel. In order to do so, a multinominal logistic regression model is applied in two phases, using the Stepwise method for incorporating variables. In the first phase, the clouds of interest to the study are separated from the rest of the cloud formations. Specifically, well-developed convective formations are detected, which are known as cumulonimbus clouds. In the second phase, the probability that hail precipitation will exist within these clouds is calculated.

Lastly, it is necessary to make a series of corrections to the results obtained in the previous step. The first correction consists of adjusting the probability of each pixel, taking into account its solar zenith angle, since the algorithm has been optimized for the calculating probability in day-time conditions.

Later, it is necessary to correct the coordinates of each pixel due to the Parallax effect. This distortion in the coordinates is produced because of the position of the satellite in relation to the curvature of the Earth. The corrections are carried out according to the method proposed in Vicente et al. (2002).

Finally, following all of the abovementioned steps, the probability of hail precipitation for each pixel on the grid for the defined study area is obtained.

The resulting information can be directly represented graphically using Python. Additionally, the results can be exported to other formats, such as, for example, NetCDF, readable by any tool that allows for the analysis and representation of this information: GrADS, NCL, Matlab, R, ncview, etc.

In summary, the methodology presented allows to construct graphic representations in which, for a chosen area of study, it is possible to know the probability that hail precipitation is being produced in real-time (nowcasting).

References

PyPI - the Python Package Index: Meteorological ingest processing package. [Available online at http://pypi.python.org/pypi/mipp.]

PyPI - the Python Package Index: Meteorological post processing package. [Available online at http://pypi.python.org/pypi/mpop.]

PyPI - the Python Package Index: python/numpy interface to netCDF library. [Available online at http://pypi.python.org/pypi/netCDF4.]

Vicente, G.A., J.C. Davenport, and R.A. Scofield, 2002: The role of orographic and parallax corrections on real time hight resolution satellite rainfall rate distribution. Int. J. Remote Sens., 203, 221-330.

Acknowledgements

The authors would like to thank the Junta de Castilla y León for their economic support via the LE220A11-2 Project.