Xiao Lou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Xiao Lou has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Deming Zhao
Key Laboratory of Regional Climate and Environment for Temperate East Asia Institute of Atmospheric Physics Chinese Academy of Sciences Beijing China
1 shared publication - Jiangjiang Xia
Temperate East Asia Research Center on Global Change, Institute of Atmospheric Physics, Chinese Academy of Sciences , Beijing 100029
1 shared publication - Wencong Cheng
Beijing Aviation Meteorological Institute, Beijing 100085, China
1 shared publication - Xiaoqi Yu
Key Laboratory of Regional Climate-Environment for Temperate East Asia & Center for Artificial Intelligence in Atmospheric Science, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China
1 shared publication - Yan Yan
93110 Troops, Beijing, 100871, China
1 shared publication - Zhibin Wang
DAMO Academy, Alibaba Group, Beijing 100102, China
1 shared publication
Latest Publications
- Open Access
- 12 June 2023
- Published by MDPI AG in Remote Sensing
- Vol. 15 (12) , 3065
Radar Echo Reconstruction in Oceanic Area via Deep Learning of Satellite Data
A conventional way to monitor severe convective weather is using the composite reflectivity of radar as an indicator. For oceanic areas without radar deployment, reconstruction from satellite data is useful. However, those reconstruction models built on a land dataset are not directly applicable to the ocean due to different underlying surfaces. In this study, we built reconstruction models based on U-Net (named STR-UNet) for different underlying surfaces (land, coast, offshore, and sea), and evaluated their applicability to the ocean. Our results suggest that the comprehensive use of land, coast, and offshore datasets should be more suitable for reconstruction in the ocean than using the sea dataset. The comprehensive performances (in terms of RMSE, MAE, POD, CSI, FAR, and BIAS) of the Land-Model, Coast-Model, and Offshore-Model in the ocean are superior to those of the Sea-Model, e.g., with RMSE being 5.61, 6.08, 5.06, and 7.73 in the oceanic area (Region B), respectively. We then analyzed the importance of different types of features on different underlying surfaces for reconstruction by using interpretability methods combined with physical meaning. Overall, satellite cloud-related features are most important, followed by satellite water-related features and satellite temperature-related features. For the transition of the model from land to coast, then offshore, the importance of satellite water-related features gradually increases, while the importance of satellite cloud-related features and satellite temperature-related features gradually decreases. It is worth mentioning that in the offshore region, the importance of satellite water-related features slightly exceeds the importance of satellite cloud-related features. Finally, based on the performance of the case, the results show that the STR-UNet reconstruction models we established can accurately reconstruct the shape, location, intensity, and range of the convective center, achieving the goal of detecting severe convective weather where a radar is not present.
- 12 June 2023
- Published by MDPI AG in Remote Sensing
- Vol. 15 (12) , 3065
Open AccessRadar Echo Reconstruction in Oceanic Area via Deep Learning of Satellite Data
A conventional way to monitor severe convective weather is using the composite reflectivity of radar as an indicator. For oceanic areas without radar deployment, reconstruction from satellite data is useful. However, those reconstruction models built on a land dataset are not directly applicable to the ocean due to different underlying surfaces. In this study, we built reconstruction models based on U-Net (named STR-UNet) for different underlying surfaces (land, coast, offshore, and sea), and evaluated their applicability to the ocean. Our results suggest that the comprehensive use of land, coast, and offshore datasets should be more suitable for reconstruction in the ocean than using the sea dataset. The comprehensive performances (in terms of RMSE, MAE, POD, CSI, FAR, and BIAS) of the Land-Model, Coast-Model, and Offshore-Model in the ocean are superior to those of the Sea-Model, e.g., with RMSE being 5.61, 6.08, 5.06, and 7.73 in the oceanic area (Region B), respectively. We then analyzed the importance of different types of features on different underlying surfaces for reconstruction by using interpretability methods combined with physical meaning. Overall, satellite cloud-related features are most important, followed by satellite water-related features and satellite temperature-related features. For the transition of the model from land to coast, then offshore, the importance of satellite water-related features gradually increases, while the importance of satellite cloud-related features and satellite temperature-related features gradually decreases. It is worth mentioning that in the offshore region, the importance of satellite water-related features slightly exceeds the importance of satellite cloud-related features. Finally, based on the performance of the case, the results show that the STR-UNet reconstruction models we established can accurately reconstruct the shape, location, intensity, and range of the convective center, achieving the goal of detecting severe convective weather where a radar is not present.