Are numerical weather model outputs helpful to reduce tropospheric delay signals in InSAR data?
Interferometric synthetic aperture radar phase data include not only signals due to crustal movements, but also those associated with microwave propagation delay through the atmosphere. In particular, the effect of water vapor can generate apparent signals in the order of a few centimeters or more, and prevent us from detecting such geophysical signals as those due to secular crustal deformation. To examine if and to what extent numerical weather model (NWM) outputs are helpful to reduce the tropospheric delay signals at spatial scales of 5–50 km wavelengths, we compared three approaches of tropospheric signal reduction, using 54 interferograms in central Hokkaido, Japan. The first approach is the conventional topography-correlated delay correction that is based on the regional digital elevation model (DEM). The second approach is based on the Japan Meteorological Agency’s operational meso-scale analysis model (MSM) data, where we compute tropospheric delays and subtract them from the interferogram. However, the MSM data are available at predefined epochs and their spatial resolution is about 10 km; therefore, we need to interpolate both temporally and spatially to match with interferograms. Expecting to obtain a more physically plausible reduction of the tropospheric effects, we ran a 1-km mesh high-resolution numerical weather model WRF (Weather Research and Forecasting model) by ourselves, using the MSM data as the initial and boundary conditions. The third approach is similar to the second approach, except that we make use of the WRF-based tropospheric data. Results show that if the topography-correlated phases are significant, both the conventional DEM-based approach and the MSM-based approach reveal comparable performances. However, when the topography-correlated phases are insignificant, none of the approaches can efficiently reduce the tropospheric phases. Although it could reduce the tropospheric signals in a local area, in none of the case studies did the WRF model produce the “best” performance. Whereas the global atmospheric model outputs are shown to be effective in reducing long-wavelength tropospheric signals, we consider that further improvements are needed for the initial and boundary condition data for high-resolution NWM, so that the NWM-based approach will become more reliable even in the case of a non-stratified troposphere.