10:45 AM - 11:00 AM
[STT39-01] Accuracy evaluation of L-Band InSAR PWV estimation using GNSS Observations in Japan
Keywords:InSAR, ALOS2, PWV estimation, GNSS
In recent years, Japan has been experiencing damage caused by heavy rainfall that develops locally and in a short period of time. The current precipitation forecasting system has some problems such as underestimating the amount of precipitation. Thus, it is necessary to improve the forecasting accuracy for heavy rainfall. The use of high-density water vapor observation data for precipitation prediction is one of the methods that can improve the prediction accuracy. In this study, we focused on Interferometric synthetic aperture radar (InSAR) as a method to observe high-spatial density water vapor information. InSAR can measure not only the surface deformation, but also the information of troposphere at the high spatial resolution. This is caused by the propagation delay effect similar to the Global Navigation Satellite System (GNSS), which is approximately proportional to the zenith total delay (ZTD) or the precipitable water vapor (PWV). By data assimilating with InSAR PWV, it is expected for meteorological models to reproduce initial values being closer to reality. Previous researches have shown that the forecast accuracy for precipitation was improved in a hindcast experiment using data assimilation with InSAR PWV. Before data assimilation, it is first necessary to develop a method to estimate PWV at a single epoch from InSAR data. However, in the previous study, InSAR PWV were obtained only from C-band radar, and there were no Japanese precipitation cases that had been studied [for example, Mateus et al. 2016 and Miranda et al. 2019]. Here we developed a single epoch PWV estimation method from L-band InSAR data by use of GNSS atmospheric observations. We evaluated the accuracy of the developed model by comparing with independent GNSS PWV data.
To estimate InSAR PWV at a single epoch, at first, we calculated the distribution of the zenith total delay (ZTD) at secondary SAR observation time using InSAR data and ZTDs observed by GNSS. Next, we estimated zenith hydrostatic delay (ZHD) from the ellipsoid height and pressure data, and we obtained the zenith wetting delay (ZWD) subtracting ZHD from ZTD. Then, we calculated PWV multiplying ZWD and the proportional constant estimated from the elevation and temperature data. In this way, the single epoch PWV distribution can be estimated from InSAR data. We evaluated the accuracy of the single epoch InSAR PWV by comparing with the PWV estimated from GNSS data. We used the temperature and pressure data observed at 10-minute intervals by AMeDAS, provided by the Japan Meteorological Agency and the GNSS Earth Observation Network System (GEONET) ZTD data processed with the 5-minutes PPP analysis by the Nevada Geodetic Laboratory at University of Nevada, Reno. Although this dataset includes the estimated PWV data, we only used the ZTD data to unify the PWV estimation method with the InSAR PWV processing.
In the InSAR processing, all interferograms were generated from level-1.1 Single Look Complex (SLC) images using the Radar Interferometry Calculation Tools (RINC) software ver. 0.41r. SAR data used were derived from the ALOS-2/PALSAR-2 stripmap mode. We adopted the Split-Spectrum Method for compensating the differential ionospheric path delay in L-band interferograms. We used 46 InSAR scenes collected over four areas in Japan: southern Ibaraki, western Tokyo and Kanagawa, Osaka, and southern Kyushu, spanning from 2014 to 2020. The multilook processing resulted in reduction of the decorrelation noise with sacrificing the spatial resolution to 100×100 m.
As a result of using 46 InSAR scenes observed in 4 areas, the mean of residuals between GNSS PWV and InSAR PWV is -0.241 mm, and the standard deviation of residuals is 1.315 mm. RMSE between GNSS and InSAR PWV is 1.337 mm. Considering the error of GNSS PWV, the error of InSAR PWV estimated by this method is about 3.10 mm. This error value is slightly larger than the error of 2.97 mm for L-Band InSAR PWV evaluated by Matsuzawa and Kinoshita (2021). However, since the value range of PWV is from 0 mm to 60 mm in general, our PWV estimation method showed significant accuracy. In the future, we plan to perform hindcast experiments using the mesoscale meteorological model with the data assimilation of InSAR PWV estimated from this method and the observation error from Matsuzawa and Kinoshita (2021).
To estimate InSAR PWV at a single epoch, at first, we calculated the distribution of the zenith total delay (ZTD) at secondary SAR observation time using InSAR data and ZTDs observed by GNSS. Next, we estimated zenith hydrostatic delay (ZHD) from the ellipsoid height and pressure data, and we obtained the zenith wetting delay (ZWD) subtracting ZHD from ZTD. Then, we calculated PWV multiplying ZWD and the proportional constant estimated from the elevation and temperature data. In this way, the single epoch PWV distribution can be estimated from InSAR data. We evaluated the accuracy of the single epoch InSAR PWV by comparing with the PWV estimated from GNSS data. We used the temperature and pressure data observed at 10-minute intervals by AMeDAS, provided by the Japan Meteorological Agency and the GNSS Earth Observation Network System (GEONET) ZTD data processed with the 5-minutes PPP analysis by the Nevada Geodetic Laboratory at University of Nevada, Reno. Although this dataset includes the estimated PWV data, we only used the ZTD data to unify the PWV estimation method with the InSAR PWV processing.
In the InSAR processing, all interferograms were generated from level-1.1 Single Look Complex (SLC) images using the Radar Interferometry Calculation Tools (RINC) software ver. 0.41r. SAR data used were derived from the ALOS-2/PALSAR-2 stripmap mode. We adopted the Split-Spectrum Method for compensating the differential ionospheric path delay in L-band interferograms. We used 46 InSAR scenes collected over four areas in Japan: southern Ibaraki, western Tokyo and Kanagawa, Osaka, and southern Kyushu, spanning from 2014 to 2020. The multilook processing resulted in reduction of the decorrelation noise with sacrificing the spatial resolution to 100×100 m.
As a result of using 46 InSAR scenes observed in 4 areas, the mean of residuals between GNSS PWV and InSAR PWV is -0.241 mm, and the standard deviation of residuals is 1.315 mm. RMSE between GNSS and InSAR PWV is 1.337 mm. Considering the error of GNSS PWV, the error of InSAR PWV estimated by this method is about 3.10 mm. This error value is slightly larger than the error of 2.97 mm for L-Band InSAR PWV evaluated by Matsuzawa and Kinoshita (2021). However, since the value range of PWV is from 0 mm to 60 mm in general, our PWV estimation method showed significant accuracy. In the future, we plan to perform hindcast experiments using the mesoscale meteorological model with the data assimilation of InSAR PWV estimated from this method and the observation error from Matsuzawa and Kinoshita (2021).