[대기] 초청세미나 (Prof. Hironobu Iwabuchi, Tohoku University)
일시 : 2025-09-26(금) 10:30 ~ 11:30
연사 : Prof. Hironobu Iwabuchi
소속 : Tohoku University
문의 : 02-880-6723
장소 : 501동 504호 세미나실
[Title] All-weather atmospheric profiling from geostationary satellite
[Abstract]
Water vapor transport in the low/middle troposphere over ocean is a major source of torrential rain in coastal areas in countries surrounded by ocean. Observations of thermodynamic state of atmosphere is limited over ocean, and frequent and dense observations, if available, is potentially improve weather forecast, particularly in heavy rain cases. Satellite observation of temperature and moisture has been performed by hyperspectral infrared sounders and microwave radiometers/sounders. Even using a combination of infrared and microwave sounders, temporal interval, accuracy, and horizontal and vertical resolutions are limited, and spatiotemporal variation with high accuracy are not available from satellites currently in space. To fill this gap, all-weather atmospheric profiling of air temperature and moisture from measurements by a multispectral imager onboard Japanese geostationary satellite Himawari is realized for the first time. The Machine Learning/Artificial Intelligence (ML/AI) take the spatial distribution of infrared brightness temperatures at infrared channels with ancillary information, such as satellite geometries and climatology of various meteorological variables, as input. The ML model is fitted to the radiosonde measurements by the maximum likelihood to jointly diagnose uncertainty estimates. As the ML model can be applied regardless of cloud, highly accurate temperature and moisture retrieval is obtained under all weather conditions, which is certainly feasible by using spatial features imprinted in the multispectral image. A case study shows that lower tropospheric precipitable water is about 10% different from ERA5 reanalysis. Large-scale distribution of the humidity in the middle and lower troposphere in deep convective cloud system in a typhoon is reasonably retrieved when compared with ERA5. The root-mean-square error in estimated relative humidity, on average, is 12% with respect to radiosonde measurements. Estimated humidity is significantly better accurate as that of the hyperspectral infrared sounder. The accuracy is stably high even in deeper part of cloud and in the middle/lower troposphere over ocean. The remote sensing community has not well explored the use of spatial features because there has been no theory to model the spatial features, but spatial features provide rich information on the atmospheric state and are worth to be explored in more applications.
[Abstract]
Water vapor transport in the low/middle troposphere over ocean is a major source of torrential rain in coastal areas in countries surrounded by ocean. Observations of thermodynamic state of atmosphere is limited over ocean, and frequent and dense observations, if available, is potentially improve weather forecast, particularly in heavy rain cases. Satellite observation of temperature and moisture has been performed by hyperspectral infrared sounders and microwave radiometers/sounders. Even using a combination of infrared and microwave sounders, temporal interval, accuracy, and horizontal and vertical resolutions are limited, and spatiotemporal variation with high accuracy are not available from satellites currently in space. To fill this gap, all-weather atmospheric profiling of air temperature and moisture from measurements by a multispectral imager onboard Japanese geostationary satellite Himawari is realized for the first time. The Machine Learning/Artificial Intelligence (ML/AI) take the spatial distribution of infrared brightness temperatures at infrared channels with ancillary information, such as satellite geometries and climatology of various meteorological variables, as input. The ML model is fitted to the radiosonde measurements by the maximum likelihood to jointly diagnose uncertainty estimates. As the ML model can be applied regardless of cloud, highly accurate temperature and moisture retrieval is obtained under all weather conditions, which is certainly feasible by using spatial features imprinted in the multispectral image. A case study shows that lower tropospheric precipitable water is about 10% different from ERA5 reanalysis. Large-scale distribution of the humidity in the middle and lower troposphere in deep convective cloud system in a typhoon is reasonably retrieved when compared with ERA5. The root-mean-square error in estimated relative humidity, on average, is 12% with respect to radiosonde measurements. Estimated humidity is significantly better accurate as that of the hyperspectral infrared sounder. The accuracy is stably high even in deeper part of cloud and in the middle/lower troposphere over ocean. The remote sensing community has not well explored the use of spatial features because there has been no theory to model the spatial features, but spatial features provide rich information on the atmospheric state and are worth to be explored in more applications.
첨부파일 (1개)
- seminar_abstract_Iwabuchi.pdf (133 KB, download:4)
