A method for intelligent prediction of wide-area severe convective weather based on satellite data

By using the CNextP model to perform feature encoding and time-series prediction on satellite radar echo data, the problem of predicting severe convective weather at sea has been solved, high spatiotemporal resolution forecasts have been achieved, and the accuracy of severe convective weather prediction at sea and the generalization ability of the model have been improved.

CN122172348APending Publication Date: 2026-06-09THE CHINESE PEOPLES LIBERATION ARMY 92859 TROOPS
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
THE CHINESE PEOPLES LIBERATION ARMY 92859 TROOPS
Filing Date
2026-02-03
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Due to the scarcity of radar data at sea, it is difficult to make accurate forecasts of severe convective weather over a wide area.

Method used

A satellite-based intelligent forecasting method is adopted, which uses the CNextP model to perform feature encoding and time-series prediction on radar echo data. Data processing is performed through a combination network of ConvNextlayer and ConvLSTM layers. The method is trained and fine-tuned by combining data from Himawari-8 and Fengyun-4A satellites to achieve 0-2 hour forecasting of severe convective weather.

Benefits of technology

It effectively solves the problem of the difficulty in predicting severe convective weather at sea, achieves high spatiotemporal resolution forecasts, fills the gap in marine radar extrapolation forecasts, and improves the accuracy and generalization ability of forecasts.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122172348A_ABST
    Figure CN122172348A_ABST
Patent Text Reader

Abstract

The present application relates to a kind of wide-area severe convective weather intelligent prediction method based on satellite data, belong to meteorological prediction technical field.The present application obtains the space-time sequence radar reflectivity factor data obtained by satellite data inversion;Data is input to the radar echo short-term extrapolation model pre-trained, and the future 0-2 hours of forecast data is output.The radar echo short-term extrapolation model of the present application adopts the CNextP neural network architecture mixed with ConvNeXt and ConvLSTM, and is trained by the two-stage strategy of "first training on the first satellite source data, then fine-tuning on the second satellite source data", with strong spatiotemporal feature extraction capability and cross-data source generalization capability.The present application effectively solves the problem of wide sea area, less radar echo data, difficult to predict severe convective weather, fills the blank of sea severe convective extrapolation forecast.
Need to check novelty before this filing date? Find Prior Art