Temperature profile plays an important role in the field of atmospheric research.The infrared hyperspectral vertical detector equipped with meteorological satellites can provide relatively high spatiotemporal resoluti...Temperature profile plays an important role in the field of atmospheric research.The infrared hyperspectral vertical detector equipped with meteorological satellites can provide relatively high spatiotemporal resolution data,thus being widely used for temperature profile retrieval.In recent years,deep learning,especially convolutional neural networks(CNNs),has attracted much attention in various meteorological and climate tasks,including temperature retrieval.However,it is a completely data-driven approach,which may generate results that violate physical laws.To address this issue,we propose a physical knowledge constrained CNN for temperature profile retrieval in this paper.Specifically,we take advantage of the physical knowledge from weight function and ERA5 data,and use an attention module and a loss function to guide the learning of CNN.In order to test the performance of our proposed model,we collect the geostationary interferometric infrared sounder(GIIRS)data,L-band radiosonde data,ERA5 data,and the GIIRS L2 operational product in China for experiments.It is shown that the root mean square error(RMSE)and mean bias(MB)achieved by our proposed model are 2.06 and 0.072 K,respectively,both of which are better than 2 state-of-the-art neural network-based retrieval models and the operational product.展开更多
基金supported by the National Natural Science Foundation of China(grant numbers U21B2049 and 62472230).
文摘Temperature profile plays an important role in the field of atmospheric research.The infrared hyperspectral vertical detector equipped with meteorological satellites can provide relatively high spatiotemporal resolution data,thus being widely used for temperature profile retrieval.In recent years,deep learning,especially convolutional neural networks(CNNs),has attracted much attention in various meteorological and climate tasks,including temperature retrieval.However,it is a completely data-driven approach,which may generate results that violate physical laws.To address this issue,we propose a physical knowledge constrained CNN for temperature profile retrieval in this paper.Specifically,we take advantage of the physical knowledge from weight function and ERA5 data,and use an attention module and a loss function to guide the learning of CNN.In order to test the performance of our proposed model,we collect the geostationary interferometric infrared sounder(GIIRS)data,L-band radiosonde data,ERA5 data,and the GIIRS L2 operational product in China for experiments.It is shown that the root mean square error(RMSE)and mean bias(MB)achieved by our proposed model are 2.06 and 0.072 K,respectively,both of which are better than 2 state-of-the-art neural network-based retrieval models and the operational product.
