Rapid Assessment ofthe Coastal Ocean Environment
217
remotely sensed data, acquired via adaptive sampling designed to have optimal
impact and control predictability is assimilated.
Data. assimilation methods used by HOPS include a robust Optimal Interpolation
(OI) scheme with weights set by simple engineering-type assumptions and a quasioptimal scheme, Error Subspace Statistical Estimation (ESSE). The advanced
ESSE method determines the nonlinear evolution ofthe oceanic state and its uncertainties by minimizing the most energetic errors under the constraints of the
dynamical and measurement models and their errors (Lermusiaux and Robinson,
1999; Lermusiaux, 1999). Measurement models relate state variables to sensor
data. Substantial real time efficiency is achieved by reducing the error covariance
to its dominant eigendecomposition. Error propagation is estimated via an ensembIe forecast using the full nonlinear model. The evolving error subspace is characterized by singular error vectors and values, i.e., time evolving three dimensional
error empirical orthogonal functions (EOFs). Melding weights for assimilation are
determined using a minimum error variance criterion. Importantly, melding occurs
in the error subspace and is thus much less costly than a classical analysis with the
full error covariances. The error subspace is updated at the melding step by combining the forecast principal errors, i.e., errors arising from the dynamical model
and the loss of predictability, with the error covariances of the measurements.
ESSE was utilized in real time for the first time during Rapid Response 96. The
influence of historical synoptic data is reduced as data is gathered and eliminated
when and if sufficient data is acquired.
During RR96, dynamically adjusted (heated while in dynamical balance) historical synoptic data from 1995 was used in combination with the Mediterranean
Ocean DataBase (MODB) climatology. In RR97, quasi-synoptic data from the
Physical Oceanography of the Eastem Mediterranean (POEM) cmise of 1986,
along with MODB data, was used to formulate initial conditions. In the Gulf of
Cadiz region (RR98), the MODB MED4 database, supplemented with data from
the UK Hydrographic Oftice, the NODC CD-ROM and the SACLANTCEN web
site, provided the basis of the initial conditions.
The Rapid Response exercises have generally provided extensive sampling during their duration and provide extremely rich data sets as a result. Fig. 11.7 shows
the sampling pattems during RR97. Three panels (top left and right and bottom
left) represent the data collected during the exercise; 5 ships and 17 aircraft flights
acquired 416 CTDs and XCTDs and 975 XBTs and AXBTs during the period 18
August - 4 October 1997. The sampling pattems were designed for maximum spatial coverage with constraints imposed by platform capabilities and mission needs.
Some pattems were pre-planned while others were adaptively planned as forecasts
and conditions dictated. The bottom right panel of Fig. 11.7 shows the data positions for the POEM 1987 data which formed the initial conditions for the RR97
forecasting.
Fig. 11.8 illustrates the effect of, and need for adaptive sampling. The left panel
shows the expected error of an objectively analyzed data field after an AXBT
flight. The path of the aircraft is obvious and was adaptively designed to reduce
217
remotely sensed data, acquired via adaptive sampling designed to have optimal
impact and control predictability is assimilated.
Data. assimilation methods used by HOPS include a robust Optimal Interpolation
(OI) scheme with weights set by simple engineering-type assumptions and a quasioptimal scheme, Error Subspace Statistical Estimation (ESSE). The advanced
ESSE method determines the nonlinear evolution ofthe oceanic state and its uncertainties by minimizing the most energetic errors under the constraints of the
dynamical and measurement models and their errors (Lermusiaux and Robinson,
1999; Lermusiaux, 1999). Measurement models relate state variables to sensor
data. Substantial real time efficiency is achieved by reducing the error covariance
to its dominant eigendecomposition. Error propagation is estimated via an ensembIe forecast using the full nonlinear model. The evolving error subspace is characterized by singular error vectors and values, i.e., time evolving three dimensional
error empirical orthogonal functions (EOFs). Melding weights for assimilation are
determined using a minimum error variance criterion. Importantly, melding occurs
in the error subspace and is thus much less costly than a classical analysis with the
full error covariances. The error subspace is updated at the melding step by combining the forecast principal errors, i.e., errors arising from the dynamical model
and the loss of predictability, with the error covariances of the measurements.
ESSE was utilized in real time for the first time during Rapid Response 96. The
influence of historical synoptic data is reduced as data is gathered and eliminated
when and if sufficient data is acquired.
During RR96, dynamically adjusted (heated while in dynamical balance) historical synoptic data from 1995 was used in combination with the Mediterranean
Ocean DataBase (MODB) climatology. In RR97, quasi-synoptic data from the
Physical Oceanography of the Eastem Mediterranean (POEM) cmise of 1986,
along with MODB data, was used to formulate initial conditions. In the Gulf of
Cadiz region (RR98), the MODB MED4 database, supplemented with data from
the UK Hydrographic Oftice, the NODC CD-ROM and the SACLANTCEN web
site, provided the basis of the initial conditions.
The Rapid Response exercises have generally provided extensive sampling during their duration and provide extremely rich data sets as a result. Fig. 11.7 shows
the sampling pattems during RR97. Three panels (top left and right and bottom
left) represent the data collected during the exercise; 5 ships and 17 aircraft flights
acquired 416 CTDs and XCTDs and 975 XBTs and AXBTs during the period 18
August - 4 October 1997. The sampling pattems were designed for maximum spatial coverage with constraints imposed by platform capabilities and mission needs.
Some pattems were pre-planned while others were adaptively planned as forecasts
and conditions dictated. The bottom right panel of Fig. 11.7 shows the data positions for the POEM 1987 data which formed the initial conditions for the RR97
forecasting.
Fig. 11.8 illustrates the effect of, and need for adaptive sampling. The left panel
shows the expected error of an objectively analyzed data field after an AXBT
flight. The path of the aircraft is obvious and was adaptively designed to reduce