Data assimilation experiment for reproducing localized delay signals derived from InSAR

Thursday, 18 December 2014
Youhei Kinoshita and Masato Furuya, Hokkaido University, Sapporo, Japan
InSAR phase signals are affected by the Earth's atmosphere like those of the GNSS. Therefore, InSAR can detect water vapor distribution with unprecedented spatial resolution if there are neither surface deformation signals or other errors, and thus is potentially useful for meteorological applications. However, there has been only a few studies using InSAR as a water vapor sensor (e.g. Hanssen et al., 1999, Kinoshita et al., 2013).
We reported six case studies that detected localized water vapor signals with InSAR based on ALOS/PALSAR data (Kinoshita et al., JpGU 2013), some of which reached over 20 cm in the LOS direction within 10 km2. Each signal located at the very location of high rainfall intensity in the weather radar data. Such localized signals strongly suggest the existence of developed convective systems at the SAR observation time.
To investigate the mechanism of the localized delay signals in the interferogram, we performed the WRF simulation in the case of Niigata on 25 August 2010. We used the JMA MSM data and the NCEP high-resolution SST data as the initial values. Two nested domains were used and horizontal resolutions of them were 3 km and 1 km, respectively. The WRF simulation could reproduce the convective system that extended east and west, and the shape of the reproduced convective system was similar to the localized signals in the interferogram, whereas the location of the reproduced convective system was about 30 km north of that of observed signals.
To improve the location of the reproduced convective system, we performed the 4DVAR experiment implemented in the WRF data assimilation model. In this study, we assimilated zenith total delay data derived from the Japanese GNSS network, GEONET. Due to the limitation of the computational resource, we performed the 4DVAR data assimilation in the coarser domain (10 km) and then we downscaled the assimilated initial value to the finer domain (2.5 km). The simulation with the assimilation could reproduce the convective system whose shape and amplitude were similar to the observed signals in the interferogram. In addition, the location of the strong signals was improved slightly in comparison with that without the assimilation.
At the presentation, we will show results from numerical simulations with and without data assimilation, and discuss them.