Atmospheric Sensing Using GNSS RO - GNSS Remote Sensing: Theory, Methods and Applications - page 144

Assimilating the RO refractivity is relatively straightforward for its simple
forward model (e.g., Eq. 6.1 ). On the other hand the forward model used for
assimilating the bending becomes more complicated and computationally expen-
sive. However, the bending angle assimilation could be preferable because the use
of bending angle reduces the preprocessing of RO data before being assimilated and
increase the vertical range of useful RO data. More specifically, it circumvents the
complication of refractivity observation errors associated with the upper boundary
condition in the refractivity retrieval, which requires a priori information (e.g.,
climatology) above a certain altitude (above 30 km). In addition, the vertical
correlation of the observation errors in the RO refractivity due to the Abel
inversion also make the error covariance used for refractivity data assimilation more
complicated (Healy and Thepaut 2006 ).
Most numerical weather prediction (NWP) operational centers assimilate either
bending angle or refractivity profiles using one-dimensional (1D) observation oper-
ators, with the assumption of local spherically symmetric atmosphere. In the lower
troposphere, however, specific humidity can vary appreciably on horizontal scales
of a few tens of kilometers, which may result in significant horizontal gradients
in refractivity (e.g., Healy 2001 ). It is likely that the errors of representativeness
due to the violation of the spherical symmetry assumption could dominate over
measurement errors in the moist lower troposphere as they appear to do in the upper
troposphere and lower stratosphere (Kuo et al. 2004 ; Syndergaard et al. 2005 ).
In this regard, the two-dimensional (2D) observation operators could be used to
reduce the representative errors of the RO observation particularly in the presence
of strong horizontal gradients. In addition the 2D operators should also reduce
the forward model error. Two-dimensional (2D) observation operators make use
of apriori knowledge of the horizontal gradients in the atmosphere provided by
an NWP forecast when simulating the observation. Due to the high computational
cost of simulating the bending angel through the conventional 2D geometric optics
ray-tracing (Zou et al. 2000 ), some simplified 2D mapping operators, i.e., the “non-
local” refractivity (Syndergaard et al. 2005 ) and phase operator (Sokolovskiy et al.
2005 ) have been proposed. Also, faster refractivity and bending angle operators have
also been developed to incorporate both the tangent point drifting and horizontal
gradient effects on the RO measurements (Poli 2004 ). Note that the increasing NWP
model resolution will make the benefits of 2D operator more obvious due to the
better-resolved horizontal structure of the atmosphere.

6.4.2 Operational Assimilation of GNSS RO in NWP Models

Since the launch of GPS/MET, several NWP studies using GPS RO observations
have yielded promising positive impact on weather forecasting, which is especially
evident in the upper troposphere and lower stratosphere (UTLS) regions (Healy
and Thepaut 2006 ; Cucurull et al. 2006 ; Aparicio and Deblonde 2008 ). The RO
data (e.g., GPS/MET) have also been used to identify the biases in NWP analysis