Ocean Ensemble Forecasting and Adaptive Sampling - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 394

16.4 Ocean Ensemble Forecast

Ocean ensemble generation is based on the ET technique, which has been used
for atmospheric ensemble generation ( Bishop et al. 2009 ) and for coupled atmo-
sphere/ocean ensemble generation ( Holt et al. 2011 ). The ET technique provides
initial perturbations that (1) have an initial variance consistent with the best available
estimates of initial condition error variance, (2) are dynamically conditioned by
a process similar to that used in the breeding technique ( Toth and Kalnay 1993 ,
1997 ), (3) add to zero at the initial time, (4) are quasi-orthogonal and equally
likely, and (5) partially respect mesoscale balance constraints by ensuring that each
initial perturbation is a linear sum of forecast perturbations from the preceding
forecast. The analysis error variance is used to constrain the magnitude of initial
perturbations that represent transformations or linear combinations of ensemble
forecast perturbations, so called ET perturbations ( Bishop and Toth 1999 ; Bishop
et al. 2009 ). The analysis error variance used in this study is scaled from the NCODA
ocean analysis to adjust large untruthful values from the sparse ocean observations.
A complete description of the ET technique and the detailed steps to creating an ET
ensemble can be found in Bishop et al. ( 2009 ).
The ocean ensemble with 20 ensemble members is initialized from a set of
perturbations derived from a control deterministic NCOM run for one month
from August 1-31, 2003. The NCOM monthly run is performed in a sequential
incremental update cycle with an update interval of 24 h and produces 72 h forecast
at each analysis update time ( Hong et al. 2009a ). The differences between every 12 h
forecast (up to 24 h) and monthly mean generate 62 perturbations, which provide a
database for random selection of initial ensemble perturbations.
From August 7-19, the winds are upwelling favorable with north/northwesterly
( Doyle et al. 2008 ) and induce strong upwellings with two upwelling centers
developed off Point Ano Nuevo and Point Sur ( Hong et al. 2009a ). Ensemble means
display stronger upwellings from the two upwelling centers than in the control run
and provide features more comparable with the observation (Fig. 16.3 ). Stronger
horizontal SST gradients occur between the upwelled cold water and the offshore
warm water. The seaward advection is more consistent with the observation from
the ensemble mean on August 12 (upper panel in Fig. 16.3 ). Later in the upwelling
period, a cold tongue of upwelled water off Point Ano Nuevo is advected southward
across the mouth of the Monterey Bay and joins with the upwelled cold water from
Point Sur, resulting in a large, cold-water region located just off the coast both in
ensemble mean and the observation. These results indicate that the ensemble means
are more accurate to the observation MCSST than the control run.
The ensemble spread increases with the forecast lead time as shown for SST
forecast in Fig. 16.4 . Large ensemble spread transports southward with time, reflect-
ing the upwelled cold water movement. It indicates that the forecasted transport of
upwelled cold water across the mouth of the Monterey Bay during the upwelling
period has high uncertainty.