2:45 PM - 3:00 PM
[SGD02-05] Real-time Analysis of Maritime Precipitable Water Vapor using Ship-based GNSS Measurements, and its Quality Control
Keywords:GNSS, Precipitable Water Vapor, Maritime observation, Kinematic Precise Point Positioning, Quasi-Zenith Satellite System
Shoji (2021) performed kinematic precise point positioning (PPP) with the aim of determining the optimum analysis settings for precipitable water vapor (PWV) retrieval at sea using ship-based Global Navigation Satellite System (GNSS). More in detail, we changed three analysis parameters: random-walk process noise (RWPN) sigma of zenith total delay (ZTD), analysis time width, and time-update interval of the Kalman filter state vector. GNSS-derived PWVs were compared with a mesoscale objective analysis (MA) of the Japan Meteorological Agency (JMA), thus confirming that a stronger RWPN constraint suppresses short-term fluctuations of GNSS PWV, and results in a smaller standard deviation. Nevertheless, the negative bias of GNSS PWV increases as PWV increases. When the analysis time width is wider than 1.5 h, such tendency becomes stronger as the analysis time width becomes wider. Update interval >1 Hz leads larger negative bias of GNSS PWV, along with higher estimation of GNSS antenna altitude.
We selected the following settings for real-time GNSS analysis system for maritime PWV monitoring: a) RWPN sigma corresponding to 3E-5 m/sqrt(s), b) time width of the sliding-window (Foster et al. 2005) analysis to 1.5 h, and c) update time interval to 2 s.
On March 26, 2021, we installed the system on two JMA research vessels and began permanent observations. The system utilizes the real-time satellite orbit called Multi-GNSS Advanced Demonstration tool for Orbit and Clock Analysis (MADOCA), transmitted from the Quasi-Zenith Satellite System (QZSS), received on board by GNSS receivers, and automatically performs kinematic PPP every 10 min (Fig. 1). The data acquisition rate exceeded 97% in nearly 4 months, until July 22, 2021. All retrieved PWVs have been used to make comparisons with MA, radio sonde observations, a satellite-mounted microwave radiometer. As a result, root mean square (RMS) differences <2.5 mm have been obtained, as well as an absolute bias value of ~1.0 mm or lower.
However, occasionally, larger than 10 mm difference occurred in comparison with the preceding and following analysis, and also in comparison with MA. When GNSS PWV becomes an abnormal value, the following common characteristics were seen:
(1)Anomalies have occurred since the early stages of analysis.
(2)No abnormality was found in analyzed GNSS PWV values just before and after the outlier occurred.
(3)In this analysis, the atmospheric delay gradient (MacMIllan 1995) is estimated at 2-second intervals as well as the zenith delay. When the PWV became an abnormal value, the atmospheric delay gradient exceeded 20 mm in most cases. Statistically, more than 99% of the atmospheric delay gradient is less than 20 mm. There is a tendency that the longer the delay gradient, the greater the degree of anomaly in GNSS PWV(Fig. 2).
Based on the above results, when the atmospheric delay gradient exceeds 40 mm, we set to execute an additional analysis in which the start time of the sliding-window is shifted forward by 10 minutes. It was found that this can avoid many of the outliers. However, abnormal values occur sometimes even if the gradient is less than 40 mm. The validity of setting the delay amount gradient of 40 mm as the threshold value needs to be examined in the future. Furthermore, we need to investigate the causes of abnormal values.
Acknowledgements
A part of this study is supported by JSPS KAKENHI grant 20H02420.
RTKLIB 2.4.3 b33 was used for positioning analysis.
References
MacMillan, D. S., 1995: Atmospheric gradients from very long baseline interferometry observations, Geophys. Res. Let., 22, 1041-1044.
Foster J., M. Bevis, S. Businger, 2005: GPS Meteorology: Sliding-Window Analysis, J. Atmos. Oceanic Technol., 22, 687-695.
Shoji Y., 2021: Optimization of water vapor analysis using ship-borne GNSS measurement, Japan Geoscience Union Meeting 2021, SGD01-17.
We selected the following settings for real-time GNSS analysis system for maritime PWV monitoring: a) RWPN sigma corresponding to 3E-5 m/sqrt(s), b) time width of the sliding-window (Foster et al. 2005) analysis to 1.5 h, and c) update time interval to 2 s.
On March 26, 2021, we installed the system on two JMA research vessels and began permanent observations. The system utilizes the real-time satellite orbit called Multi-GNSS Advanced Demonstration tool for Orbit and Clock Analysis (MADOCA), transmitted from the Quasi-Zenith Satellite System (QZSS), received on board by GNSS receivers, and automatically performs kinematic PPP every 10 min (Fig. 1). The data acquisition rate exceeded 97% in nearly 4 months, until July 22, 2021. All retrieved PWVs have been used to make comparisons with MA, radio sonde observations, a satellite-mounted microwave radiometer. As a result, root mean square (RMS) differences <2.5 mm have been obtained, as well as an absolute bias value of ~1.0 mm or lower.
However, occasionally, larger than 10 mm difference occurred in comparison with the preceding and following analysis, and also in comparison with MA. When GNSS PWV becomes an abnormal value, the following common characteristics were seen:
(1)Anomalies have occurred since the early stages of analysis.
(2)No abnormality was found in analyzed GNSS PWV values just before and after the outlier occurred.
(3)In this analysis, the atmospheric delay gradient (MacMIllan 1995) is estimated at 2-second intervals as well as the zenith delay. When the PWV became an abnormal value, the atmospheric delay gradient exceeded 20 mm in most cases. Statistically, more than 99% of the atmospheric delay gradient is less than 20 mm. There is a tendency that the longer the delay gradient, the greater the degree of anomaly in GNSS PWV(Fig. 2).
Based on the above results, when the atmospheric delay gradient exceeds 40 mm, we set to execute an additional analysis in which the start time of the sliding-window is shifted forward by 10 minutes. It was found that this can avoid many of the outliers. However, abnormal values occur sometimes even if the gradient is less than 40 mm. The validity of setting the delay amount gradient of 40 mm as the threshold value needs to be examined in the future. Furthermore, we need to investigate the causes of abnormal values.
Acknowledgements
A part of this study is supported by JSPS KAKENHI grant 20H02420.
RTKLIB 2.4.3 b33 was used for positioning analysis.
References
MacMillan, D. S., 1995: Atmospheric gradients from very long baseline interferometry observations, Geophys. Res. Let., 22, 1041-1044.
Foster J., M. Bevis, S. Businger, 2005: GPS Meteorology: Sliding-Window Analysis, J. Atmos. Oceanic Technol., 22, 687-695.
Shoji Y., 2021: Optimization of water vapor analysis using ship-borne GNSS measurement, Japan Geoscience Union Meeting 2021, SGD01-17.