GPS derived ZTDs and those obtained from ray-tracing (Fig. 4). It
can be seen that the meso-scale model (MANAL) as well as the
fine-mesh model (CReSS) agree well with the GPS estimates
until around September 2nd, 12 h UT. Thereafter the JMA model
seems to drift away from what GPS measures, whereas the finemesh
model continues to predict similar troposphere delays as
GPS does. On September 3rd 0 h UT, the fine-mesh model is
initialized newly, using input from the JMA model, and thus
forced to follow the behavior of the meso-scale model, leading
to a larger bias between the weather models and the reality as
measured by GPS. Both models take about 12 h until they agree
again well with the GPS ZTDs. Since the JMA model is not too
much offset at the epoch of the next initialization, i.e.
September 4th, 0 UT, both weather models closely follow the
results obtained from GPS. On September 5th, the day before the
typhoon makes landfall, both weather models predict similar
absolute values of zenith total troposphere delay, whereas the
GPS results show a different pattern.
Fig. 4. Zenith totals delays (in meter) obtained from GPS (green) and from ray-tracing using the MANAL (red) and CReSS model (blue) for station 3017 during the typhoon
passage. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)