InSAR Observation and Numerical Modeling of the Water Vapor Signal during 2008 Seino Heavy Rain Event, Central Japan
Kinoshita, Youhei¹; Shimada, Masanobu²; Furuya, Masato^1
1Natural History Sciences, Hokkaido University, JAPAN; ²Earth Observation Research Center, Japan Aerospace Exploration Agency, JAPAN

Interferometric Synthetic Aperture Radar (InSAR) is a technique for mapping the phase difference between two observations and used for detecting ground deformation associated with, for instance, earthquakes and volcanic activities. Like GPS, the phase signal in InSAR is affected by the earth's atmosphere that has a different refractive index relative to vacuum. In the neutral atmosphere, water vapor has the largest impact on InSAR phase signals due to its spatiotemporal heterogeneity. However, we can take advantage of this to detect an extremely detailed spatial distribution of precipitable water vapor in the absence of any ground displacements (Hanssen et al., 1999).
On 2-3 September 2008, a torrential rain known as Seino heavy rain occurred over wide areas of central Japan. We generated an InSAR image from ALOS/PALSAR emergency observation data to examine the details of the heterogeneous water vapor distribution during the heavy rain event. Near Ibi River, we detected localized but large-amplitude signals that were associated with the water vapor distribution during the heavy rain, and indicate a radar line-of-sight (LOS) change of approximately 130 mm with a horizontal scale of 10 km. Comparing this with other interferograms over the same region, we verified that the localized signal is due to neither any ground displacements, DEM errors nor ionospheric effects, but due to water vapor in the troposphere. In addition, the weather radar image at the moment of the SAR image acquisition indicates a small area with high rainfall intensity at the very location of the signal. The high rainfall intensity indicates that there was a well-developed thundercloud at that location, and thus, strongly supports that the signal is due to the presence of water vapor during the heavy rain event.
Based on the interferogram that includes the water vapor-related localized signal, we estimated three-dimensional (3D) water vapor distribution during the heavy rain with the ray tracing technique (Hobiger et al., 2008). The 3D refractivity field is a function of dry air density, temperature and water vapor density (Aparicio and Laroche, 2011), and is needed to estimate the delay amount by ray tracing technique. As an initial estimate for the 3D distributions of partial pressure of dry air, temperature and partial water pressure of water vapor, we used data sets from the 10-km mesh Meso-scale Model (MSM) data from JMA. Then, a water vapor distribution of the location of the InSAR observation was constructed by trial and error, in order to clarify the signal in the interferogram.
Although MSM data indicated an extremely humid lower troposphere (relative humidity, RH, greater than 90 %) and a relatively dry upper troposphere (RH less than 60 %), there were no spotty and saturated (100 % RH) signals corresponding to the signal. The inferred 3D refractivity distribution indicates saturated air from the surface to 9000 m above ground level in 10 km of the spotty signal, and the inferred delay can reproduce the spotty signal. Although the distribution of water vapor cannot be uniquely estimated, it is clear that humidity levels in the middle troposphere were extremely high above the spotty signal.
In order to clarify what local weather systems caused the localized delay signal in the interferogram, a numerical experiment of the heavy rain was conducted using a high-resolution non-hydrostatic NWM, the Weather Research and Forecasting (WRF) model (Skamarock et al., 2008). The data for initial and boundary conditions were constructed from the global analyses of NCEP with the horizontal resolution of 1 degree. Four nested grid domains were used in the current simulations, and the horizontal resolution of the finest domain is 333 m, which can significantly resolve small-scale phenomena on the order of 1 km. The 30-second (~1 km) resolution global topography model (GTOPO30) was used for the topography of the simulations.
As a result of the WRF simulations, we could successfully reproduce a deep convection cell that passed through the location of the spotty signal at 14:35 Coordinated Universal Time (UTC), an hour after the PALSAR observation time. The zenith PWV at the location of signal A was 15 mm higher than that around the signal. Both the maximum amplitude and the spatial scale of the simulated propagation delay are comparable to those in the interferogram, and the interior of the convection was saturated with water vapor. This simulation clearly demonstrates that a deep convective system causes a localized delay signal whose amplitude is greater than ~10 cm in an interferogram.
The simulation results indicate that with a background of southerly wind, the convection was generated on the leeward side of the Yoro Mountains, which are approximately 1000 m high. To confirm this, another WRF simulation was conducted by removing the Yoro Mountains from GTOPO30, which did not reveal the deep convective system found in the original simulation. During the Seino heavy rain event, the lower layer of the troposphere was extremely humid and higher layers of the troposphere were dry, suggesting that the conditions on a synoptic scale were very unstable. Therefore, these simulations strongly suggest that the deep convection leading to the spotty signal was initiated by the orographic lift of the Yoro Mountains and in unstable conditions with a background southerly wind.