Multi-Level Circulation Pattern Classification Based on the Transfer Learning CNN Network

Deep learning artificial intelligence technology, which has the advantages of nonlinear mapping ability, massive information extraction spatial-temporal modeling and so on, provides new ideas methods for further improving accuracy weather climate extreme event prediction. A transfer CNN (Convolutional Neural Networks) classification model is established to classify circulation patterns, along with newly reconstructed dataset regional persistent historical heavy rain events, daily rainfall data 2474 observational stations, NCEP/NCAR global reanalysis geopotential height field in 1981–2018. Different from previous classifications, usually one level variable, here, addition 500 hPa heights, 200 zonal winds 850 meridional over key areas are also considered model. The results show that multi-level pattern based on network a higher independent test than single-level model, reaching 92.5% (while only 85% model). spatial correlation coefficient precipitation between each typical mode related patterns obtained by greater CNN, variance heights associated CNN. These performance better study helpful develop classifications large-scale or disaster events then provide physical basis forecast effect extending valid time through combining numerical products.