Ocean Ensemble Forecasting and Adaptive Sampling - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 384

variables defining the model state, (c) the actual correlation between errors of the
observed variable and the model state variables, and (d) the growth and movement
of the change in the estimated state imparted by the supplemental observations. In
many applications, there is a special region called a verification region and a special
time called a verification time . One often wishes to collect and use supplemental
observations at an earlier observation time to minimize the forecast error variance
within the verification region at the verification time. The problem of identifying the
best location for deploying mobile observation platforms is often called the adaptive
sampling or targeting problem. The importance of this problem has been heightened
in oceanic applications by the advent of Autonomous Underwater Vehicles (AUVs)
and underwater gliders. These observing platforms need to be told where to go and
when. Since one must decide where to take the supplemental observations well
before the targeting time, it is critical to solve the adaptive sampling problem in
an accurate and timely manner. The ETKF based technique is used to provide the
guidance of the ocean adaptive sampling for the supplemental ocean observations.
The ETKF uses an ensemble forecast initialized at an initialization time to
quickly obtain the prediction error covariance matrix associated with a particular
deployment of observation by solving a low rank Kalman filter equation. The
technique can quickly assess the ability of a large number of future feasible
sequences of observations to reduce the forecast error variance. The ETKF was
developed by Bishop et al. ( 2001 ) and first used to provide the optimal flight
tracks, where Global Positioning System (GPS) dropwindsondes were released
during the Winter Storm Reconnaissance (WSR) program ( Szunyogh et al. 2000 ),
for improving the 24-72 h forecasts over the continental United States ( Majumdar
et al. 2002 ). It was also used for the medium range forecasts through a single
model ensemble ( Buizza et al. 2003 ; Sellwood et al. 2008 ), and a multi-model
ensemble ( Majumdar et al. 2010 ), as well as for tropical cyclone predictions
( Majumdar et al. 2011 ). While the ETKF technique is increasingly used in the area
of atmospheric adaptive sampling, there are relatively few applications in the area
of ocean adaptive sampling.
In this study, the ETKF ocean adaptive sampling technique is applied to the
glider data collected during the AOSN II field campaign that took place in the
Monterey Bay in August 2003. The goal for the month-long field experiment was
to build a fundamental understanding for upwelling and relaxation processes as
well as their impact on the other biological (ecosystem productivity) and chemical
(nutrient fertilization) counterparts in the Monterey Bay. To achieve the goal, it was
important to develop strategies to command sophisticated robotic vehicles to the
locations where the observations collected by them could be the most useful ones
( AOSN 2003 ). Multiple AUVs and underwater gliders were deployed during the
field campaign to collect data so that the data could be integrated into ocean forecast
models for improving the model performance.
The ocean ensemble and adaptive sampling technique presented here is a
continued effort of the verification of ocean modeling project ( Hong et al. 2009a ).
The deterministic run in Hong et al. ( 2009a ) is used as the control run of the
ensemble simulation in this study. Consequently, the model, model configuration,