SATFRONT
SATFRONT is a machine-learning-based framework for atmospheric front detection from satellite data. It identifies cold fronts, warm fronts and occluded fronts, with satellite observations as the primary input and reanalysis data as an optional auxiliary input.
Timeline (last 90 days)
SATFRONT is a machine learning framework aimed at detecting cold fronts, warm fronts, and occluded fronts over Europe from geostationary satellite observations, using satellite observations as the primary input and reanalysis data as an optional auxiliary input.
Using only satellite observations, SATFRONT achieves a critical success index (CSI) of 0.428 at a 150 km tolerance.
When ERA5 reanalysis data is used as an additional input, SATFRONT's CSI rises to 0.591, exceeding the upper limit of consistency among manual analyses from different agencies.
SATFRONT's training data comes from the front analysis products of five European national meteorological agencies, and evaluation uses the UK Met Office's analysis results.
SATFRONT's evaluation scope is limited to the European region, and the model's generalization ability in other climate regions remains to be verified.
SATFRONT demonstrated that cold fronts, warm fronts, and occluded fronts can be detected directly from geostationary satellite observations, laying the foundation for high-frequency automated front analysis.
If the highest accuracy is pursued, the combination of satellite plus reanalysis is the best choice; if the application scenario requires low latency, satellite-only input can already provide usable detection results.
All Facts (7)
SATFRONT is a machine learning framework aimed at detecting cold fronts, warm fronts, and occluded fronts over Europe from geostationary satellite observations, using satellite observations as the primary input and reanalysis data as an optional auxiliary input.
50%UnverifiedUsing only satellite observations, SATFRONT achieves a critical success index (CSI) of 0.428 at a 150 km tolerance.
50%UnverifiedWhen ERA5 reanalysis data is used as an additional input, SATFRONT's CSI rises to 0.591, exceeding the upper limit of consistency among manual analyses from different agencies.
50%UnverifiedSATFRONT's training data comes from the front analysis products of five European national meteorological agencies, and evaluation uses the UK Met Office's analysis results.
50%UnverifiedSATFRONT's evaluation scope is limited to the European region, and the model's generalization ability in other climate regions remains to be verified.
50%UnverifiedSATFRONT demonstrated that cold fronts, warm fronts, and occluded fronts can be detected directly from geostationary satellite observations, laying the foundation for high-frequency automated front analysis.
50%UnverifiedIf the highest accuracy is pursued, the combination of satellite plus reanalysis is the best choice; if the application scenario requires low latency, satellite-only input can already provide usable detection results.
50%