TropNet: a deep spatiotemporal neural network for tropospheric delay modeling and forecasting

Water vapor plays an essential role in regulating the earth’s weather and climate, and the tropospheric delays caused by water vapor are one of the error sources in space geodetic techniques. Attributing to the enhancement of the spatiotemporal resolution of satellite observations and the availabili...

Full description

Saved in:
Bibliographic Details
Published in:Journal of geodesy 2023-04, Vol.97 (4), Article 34
Main Authors: Lu, Cuixian, Zheng, Yuxin, Wu, Zhilu, Zhang, Yushan, Wang, Qiuyi, Wang, Zhuo, Liu, Yuxuan, Zhong, Yaxin
Format: Article
Language:English
Subjects:
Earth and Environmental Science
Earth Sciences
Geodetics
Geophysics/Geodesy
Meteorological satellites
Modelling
Neural networks
Original Article
Performance evaluation
Radiosondes
Resolution
Satellite observation
Satellites
Troposphere
Water vapor
Water vapour
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Water vapor plays an essential role in regulating the earth’s weather and climate, and the tropospheric delays caused by water vapor are one of the error sources in space geodetic techniques. Attributing to the enhancement of the spatiotemporal resolution of satellite observations and the availability of model simulation data, the fusion of the two datasets provides a promising opportunity to improve the performance of tropospheric delay modeling and forecasting. In this contribution, a tropospheric delay network (TropNet) model is developed based on deep learning method to forecast the zenith wet delays (ZWD) by combining information provided by the Geostationary Operational Environmental Satellite-R series and the global forecast system (GFS). The performance of the tropospheric delays predicted from TropNet is assessed with tropospheric products derived from GNSS. The results demonstrate that the TropNet predicted ZWD agree well with the GNSS-derived ZWD, and an accuracy of better than 11 mm is achieved for all the forecast lead times, showing an overall improvement of 15.5% when compared to the GFS ZWD. Moreover, intercomparisons with ZWD derived from radiosondes and Vienna Mapping Functions 3 (VMF3) are performed to further evaluate the performance of the TropNet model. Averaged RMS values equal to 14.9 mm and 13.9 mm are obtained when compared to radiosondes and VMF3. Furthermore, the TropNet model is able to forecast high-quality ZWD up to 6 h at a spatial resolution of 2 km and a temporal resolution of 1 h, which indicates a prospective potential for time-critical applications.
ISSN:0949-7714
1432-1394
DOI:10.1007/s00190-023-01722-4