MEDUSA - Model Error Detection by Using Simulated Satellite Images

The project MEDUSA aims at an objective quantitative decision tool to judge simulated imagery vis-à-vis reality, enabling a profound selection of the most appropriate forecast model for the day.

Short Description

Errors arising from mathematical/computational restrictions (approximation of non-linear differential equations, the chaotic nature of weather, the spatial resolution of the model), from assumptions in the underlying physics and lack of observational data are responsible for occasional conspicuous deviations in numerical weather forecasting models, compared with the truth represented in meteorological satellite images.

One way to make model output directly comparable to the satellite data are "simulated satellite images", i.e. hypothetical observations by the satellite sensor if the model-predicted three-dimensional distributions of temperature and moisture were perfectly correct.

Two independent detection techniques were elaborated for erroneously forecasted propagation speeds of meteorological systems (namely: computation of shift vectors between truth and simulation, objective recognition of typical stripe-shaped patterns in difference images).

Moreover, large average discrepancies within a segment (being defined as an area where both the real satellite image and the simulated counterpart show fairly homogeneous patterns) are detected automatically, alerting the forecaster on meteorological systems being underestimated / overestimated / absent in the numerical weather chart.

Project Partners


Zentralanstalt für Meteorologie und Geodynamik - Alexander Jann


European Centre for Medium-Range Weather Forecasts - Jean-Noël Thépaut

Contact Address