Variational Data Assimilation for the Global Ocean - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 324

Fig. 13.10 Time of RMS and mean bias error statistics for temperature observations in HYCOM
Pacific basin. Upper panel reports RMSE, middle panel reports mean bias, and bottom gives
temperature data counts. Tick marks along time axis indicate 24-h update cycle periods

observation space and represent averages across all data assimilated for a particular
analysis variable. Innovation RMS errors for temperature (Fig. 13.10 ) and salinity
(Fig. 13.11 ) show increased errors for the first few update cycles while the free
running model adjusts to the data. After this initial adjustment time, RMS errors
are very stable, with temperature errors

PSU.
The model innovations are remarkably unbiased in both temperature and salinity.
The 3DVAR analysis produces a reduction in error from the innovations to the
residuals of about 60 %, which is clearly seen in both temperature and salinity.
However, the time series of the layer pressure error statistics (Fig. 13.12 )arethe
most interesting. When cycling with HYCOM, the 3DVAR includes a sixth analysis
variable, layer pressure. Layer pressure innovations are computed as differences in
the depths of density layers in the observations and the model forecast. The layer
pressure correction fields are then used to correct isopycnal layer depths in the
model. Unlike the fairly rapid response of the free-running model to the assimilation
of temperature and salinity observations, bias in the layer structure of the model
spin-up takes about a month to adjust to the data. Layer pressure RMS errors remain

0:4 ı C and salinity errors 0:1