Allan R. Robinson
Harvard University
Division of Engineering and Applied Sciences
Department of Earth and Planetary Sciences

•
•
•
•
–
Episodic – Mass. Bay/GOM
–
Sustained – Monterey Bay
•
Harvard University
Patrick J. Haley, Jr., Pierre F.J. Lermusiaux,
Wayne G. Leslie, X. San Liang,
Oleg Logutov, Patricia Moreno
Avijit Gangopadhyay (Umass.-Dartmouth)

Interdisciplinary Ocean Science Today
•
Research underway on coupled physical, biological, chemical, sedimentological, acoustical, optical processes
• Ocean prediction for science and operational applications has now been initiated on basin and regional scales
•
Interdisciplinary processes are now known to occur on multiple interactive scales in space and time with bi-directional feedbacks

•
The concept of Ocean Observing and Prediction
Systems for field and parameter estimations has crystallized with three major components

An observational network: a suite of platforms and sensors for specific tasks

A suite of interdisciplinary dynamical models

Data assimilation schemes
•
Systems are modular, based on distributed information providing shareable, scalable, flexible and efficient workflow and management

•
Data assimilation can contribute powerfully to understanding and modeling physical-acoustical-biological processes and is essential for ocean field prediction and parameter estimation
•
Model-model, data-data and data-model compatibilities are essential and dedicated interdisciplinary research is needed

Interdisciplinary Processes - Biological-Physical-Acoustical Interactions
Physics - Density
Almeira-Oran front in Mediterranean Sea
Fielding et al , JMS, 2001
Biology –
Fluorescence
(Phytoplankton)
Acoustics –
Backscatter
(Zooplankton)
Griffiths et al ,
Vol 12, THE SEA

Biological-Physical-Acoustical Interactions
•
Distribution of zooplankton is influenced by both animal behavior
(diel vertical migration) and the physical environment.
•
Fluorescence coincident with subducted surface waters indicates that phytoplankton were drawn down and along isopycnals, by cross-front ageostrophic motion, to depths of 200 m.
• Sound-scattering layers (SSL) show a layer of zooplankton coincident with the drawn-down phytoplankton. Layer persists during and despite diel vertical migration.
• Periodic vertical velocities of ~20 m/day, associated with the propagation of wave-like meanders along the front, have a significant effect on the vertical distribution of zooplankton across the front despite their ability to migrate at greater speeds.

x = [ x
A x
O x
B
] Unified interdisciplinary state vector
Physics: x
O
= [T, S, U, V, W]
Biology: x
B
= [N i
, P i
, Z i
, B i
, D i
, C i
]
Acoustics: x
A
= [Pressure (p), Phase (

)]
P = e
{
( x ˆ – x t ) (
ˆ t ) T
}
Coupled error covariance with off-diagonal terms
P
AA
P
AO
P = P
OA
P
OO
P
BA
P
BO
P
AB
P
OB
P
BB

THREE PHASES OF (ITERATIVE)
COUPLED MODELS SKILL ASSESSMENT
VALIDATION
•
Structures of models generally capable of representation of relevant dynamical processes
• Adequacy and compatibility of physical/biological data for process identification
• Dynamical models with data assimilation robustly represent events and identify processes
CALIBRATION
• Tuning of models’ parameters: dynamical rates, subgridscale processes, computational protocols
•
Dynamical models without data assimilation represent events that are validated by data previously assimilated
VERIFICATION
• Real-time prediction verified by skill assessment quantities within
a priori specified bounds
• Hindcasts and simulations with quantitative statistical dynamical behavior

To develop, validate, and demonstrate an advanced relocatable regional ocean prediction system for real-time ensemble forecasting and simulation of interdisciplinary multiscale oceanic fields and their associated errors and uncertainties, which incorporates both autonomous adaptive modeling and autonomous adaptive optimal sampling

HOPS/ESSE System
Harvard Ocean Prediction System Error Subspace Statistical Estimation

Harvard Generalized Adaptable Biological Model
(R.C. Tian, P.F.J. Lermusiaux, J.J. McCarthy and A.R. Robinson, HU, 2004)

To achieve regional field estimates as realistic and valid as possible:
• every effort is made to acquire and assimilate both remotely sensed and in situ synoptic multiscale data from a variety of sensors and platforms in real time or for the simulation period , and a combination of historical synoptic data and feature models are used for system initialization
• “ fine-tune
” the model to the region , processes and variabilities : examine model output, modify set-up (e.g. grids, etc.) and alter structure and values of parameters (e.g.
SGS, boundary conditions, etc.)
• continuously evaluate and iterate tuning as necessary

To extend the HOPS-ESSE assimilation, real-time forecast and simulation capabilities to a single interdisciplinary state vector of ocean physicalacoustical-biological fields.
To continue to develop and to demonstrate the capability of multiscale simulations and forecasts for shorter space and time scales via multiple space-time nests (Mini-HOPS), and for longer scales via the nesting of HOPS into other basin scale models.
To achieve a multi-model ensemble forecast capability.

Mini-HOPS – Corsican Channel 2003
• Designed to locally solve the problem of accurate representation of sub-mesoscale synopticity , including inertial motions
• Involves rapid real-time assimilation of high-resolution data in a high-resolution model domain nested in a regional model
• Produces locally more accurate oceanographic field estimates and short-term forecasts and improves the impact of local field high-resolution data assimilation
• Dynamically interpolated and extrapolated high-resolution fields are assimilated through
2-way nesting into large domain models
Modeling Domains
In collaboration with Dr. Emanuel Coelho (NATO Undersea Research Centre)

Mini-HOPS for MREA-03
Prior to experiment, several configurations were tested leading to selection of 2-way nesting with super-mini at Harvard
• During experiment:
– Daily runs of regional and super mini at Harvard
– Daily transmission of updated IC/BC fields for mini-HOPS domains
– Mini-HOPS successfully run aboard NRV Alliance
Mini-HOPS simulation run aboard NRV Alliance in Central mini-HOPS domain (surface temperature and velocity)

Results of MREA03 Re-analysis and Model Tuning
Real-time Model/Data Comparison Re-analysis Model/Data Comparison
Model
Temp.
Observed
Temp.
Bias residue
< .25
o C
• Tuned parameters for stability and agreement with profiles (especially vertical mixing)
• Improved vertical resolution in surface and thermocline
• Corrected input net heat flux
• Improved initialization and synoptic assimilation in dynamically tuned model

Error Analyses and Optimal (Multi) Model Estimates
Real-Time Forecast Training via Maximum-Likelihood Correction
Model
Temp.
Observed
Temp.
Uncorrected Training: Full Data Set
Training: 1 Profile Training: 3 Profiles

Nesting HOPS in a Coarse-Resolution Climate PCM
A.R. Robinson, P.J. Haley, Jr., W.G. Leslie
Concept
N est a high-resolution hydrodynamic model within a coarse-resolution
Global Circulation Model (GCM) to provide high-resolution forecasts of mesoscale features in the Gulf of Maine
• Results applicable to cod and lobster temperaturedependent behavior change, including recruitment
Setup
• Parallel Climate Model (PCM) outputs for 2000 and 2085 for coarse circulation (1 degree)
• Gulf of Maine Feature Model provides higher-resolution synoptic circulation (5km)
• Harvard Ocean Prediction System (HOPS) dynamically adjusts
Feature Model output

Gulf of Maine Feature Model
• Prevalent circulation features are identified
•
Synoptic water-mass structures are characterized and parameterized to develop T-S feature models
•
Temperature and Salinity feature model profiles are placed on a regional circulation template
Gangopadhyay, A., A.R. Robinson, et al ., 2002. Feature-oriented regional modeling and simulations (FORMS) in the Gulf of Maine and
Georges Bank, Continental Shelf Research , 23 (3-4), 317-353
Parallel Climate Model (PCM)
Ocean Fields

Synoptic Estimate of Surface Salinity
OA “synoptic” climatology
Add Maine Coastal Current
Dynamically adjusted fields for September 2000
OA slope water climatology Resulting synoptic estimate

Nesting High Resolution HOPS in a Climate Model
A.R. Robinson, P.J. Haley, Jr., W.G. Leslie
• In order to arrive at future state, calculate the difference between the global model fields in 2000 and 2085 on the shelf and in deep water
• Perturb present-day observed fields by the “climate change” profile to achieve the 2085 fields
Shelf
Deep

Compare September 2000 and September 2085
Dynamically adjusted fields for September 2000
• Temperature increases from
2000-2085
• Salinity increases from 2000-2085
• Details under study
Dynamically adjusted fields for September 2085
Temperature Salinity http://oceans.deas.harvard.edu/UCS
• Fields provided to Union of
Concerned
Scientists
(UCS) for studies on regional climate change effects

Wind-Induced Upwelling
Massachusetts Bay
Episodic upwelling
Monterey Bay
Sustained Upwelling
Red = Wind, Blue = Upwelling

Dominant circulation and biophysical dynamics for trophic enrichment and accumulation
• Patterns are not present at all times
• Most common patterns (solid), less common (dashed)
• Patterns drawn correspond to main currents in the upper layers of the pycnocline where the buoyancy driven component of the horizontal flow is often the largest

ASCOT-01 (6-26 June 2001):
Positions of data collected and fed into models

ASCOT-01: Sample Real-Time Forecast Products
Massachusetts Bay Gulf of Maine
2m Temp.
10m Temp.
3m Temp.
5m Chlorophyll 15m Nitrate
25m Temp.

• Moderate southerly winds lead to upwelling on the western side of Cape Cod Bay
• Near the surface temperature decreases from
17 o C to 12 o C
• Near the surface chlorophyll increases from 1.4 mg Chl/m 3 to 2.3
• One-half day later, chlorophyll
– continues to increase near the surface
– decreases between 5-10m
• Between 3-10m there is maximum primary production
• Advective effects are stronger, bringing the newly produced chlorophyll closer to the surface
• Primary production during the upwelling event is mainly due to ammonium uptake
• Nitrate acts as a passive tracer
P.A. Moreno
Upwelling signature in T
(top) and chlorophyll
(bottom)

Red - data
Black - model
• 20 sets of nutrient uptake parameters tried followed by 3 model runs.
• Nutrient uptake data is used to determine the photosynthesis and nutrient limitation parameters.
• Guided by MWRA primary productivity data and literature survey.

Data: Nutrient uptake rate in the euphotic zone for June 12-13, 2001
20
15
10
5
0
43 44 45 46 47 48 49 53 54 55 56 57 58 59 50 52
Station ID
Phytoplankton equation: dP dt
 P max

1  e   E / P max

 e   E / P max 

 k
NO
3
NO
3
 NO
3
 e  NH
4
 k
NH
4
NH
4
 NH
4


  P  R m

1  e  P

 Z
Photosynthesis Nutrient limitation
Grazing
Choose zooplankton grazing parameters such that the nutrient uptake is approximately balanced by grazing.

Episodic upwelling events in Massachusetts Bay
• Historical wind data (May-June 1985-2005) indicates strongest wind events of 1-2 days and maximum wind stress 2-4 dyn/cm 2
• Design two feature wind events: duration 2 days, 2 and 4 dyn/cm 2
Winds Observed at
Mass. Bay Buoy
Feature Wind Event
Replacing
Observations
Simulation begins 6 June 2001 and lasts 20 days. Feature event starts four days into simulation.

Prediction: For strong winds the chlorophyll is upwelled and advected away from the coast; work in progress
June 2001 Wind
June 2001 Wind
Chlorophyll

Platforms, sensors and integrative models: HOPS-
ROMS real-time forecasting and re-analyses

Coastal upwelling system: sustained upwelling – relaxation – re-establishment
30m Temperature: 6 August – 3 September (4 day intervals)
6 Aug 10 Aug 14 Aug 18 Aug
22 Aug 26 Aug 30 Aug 3 Sep
Descriptive oceanography of re-analysis fields and and real-time error fields initiated at the mesoscale.
Description includes: Upwelling and relaxation stages and transitions, Cyclonic circulation in
Monterey Bay, Diurnal scales, Topography-induced small scales, etc.

Ano Nuevo
Monterey
Bay
HOPS AOSN-II Re-Analysis
18 August 22 August
Point Sur

Which sampling on Aug 26 optimally reduces uncertainties on Aug 27?
4 candidate tracks, overlaid on surface T fcst for Aug 26
•
Based on nonlinear error covariance evolution
•
For every choice of adaptive strategy, an ensemble is computed
IC(nowcast) DA
ESSE fcsts after DA of each track
Aug 24 Aug 26 Aug 27
Best predicted relative error reduction: track 1
DA 1
DA 2
ESSE for Track 1
ESSE for Track 2
2-day ESSE fct
DA 3
ESSE for Track 3
DA 4
ESSE for Track 4

Error Analyses and Optimal (Multi) Model Estimates
Strategies For Multi-Model Adaptive Forecasting
• Error Analyses:
Learn individual model forecast errors in an on-line fashion through developed formalism of multi-model error parameter estimation
• Model Fusion: Combine models via Maximum-Likelihood based on the current estimates of their forecast errors
3-steps strategy, using model-data misfits and error parameter estimation
1. Select forecast error covariance and bias parameterization
2. Adaptively determine forecast error parameters from model-data misfits based on the Maximum-Likelihood principle:
Where is the observational data
3. Combine model forecasts via Maximum-Likelihood based on the current estimates of error parameters (Bayesian Model Fusion)
O. Logutov

Bayesian Adaptive Multi-Model Forecasting
ROMS and HOPS SST forecasts for August 28, 2003 with track of validating NPS aircraft SST data taken on August 29, 2003
Model-data misfits is the source of information that is utilized to estimate the uncertainty parameters in models via Maximum-Likelihood. The models are then combined based on the uncertainty parameters, as

Bayesian Adaptive Multi-Model Forecasting
ROMS and HOPS individual SST forecasts and the NPS aircraft SST data are combined based on their estimated uncertainties to form the central forecast
• A new batch of model-data misfits and priors on uncertainty parameters determine via the Bayesian principle uncertainty parameter values that are employed to combine the forecasts.
• The Bayesian model fusion technique that we advocate treats forecast errors from different models as uncorrelated in order to gain its capability to work with a small sample of past validating events, however, accounts for spatial structure in forecast error covariances.

Multi-Scale Energy and Vorticity Analysis

Multi-Scale Energy and Vorticity Analysis
MS-EVA is a new methodology utilizing multiple scale window decomposition in space and time for the investigation of processes which are:
• multi-scale interactive
• nonlinear
• intermittent in space
• episodic in time
Through exploring:
• pattern generation and
• energy and enstrophy
- transfers
- transports, and
- conversions
MS-EVA helps unravel the intricate relationships between events on different scales and locations in phase and physical space.
Dr. X. San Liang

Multi-Scale Energy and Vorticity Analysis
Window-Window Interactions:
MS-EVA-based Localized Instability Theory
Perfect transfer :
A process that exchanges energy among distinct scale windows which does not create nor destroy energy as a whole.
In the MS-EVA framework, the perfect transfers are represented as field-like variables. They are of particular use for real ocean processes which in nature are non-linear and intermittent in space and time.
Localized instability theory :
BC : Total perfect transfer of APE from large-scale window to meso-scale window.
BT : Total perfect transfer of KE from large-scale window to meso-scale window.
BT + BC > 0 => system locally unstable; otherwise stable
If BT + BC > 0, and
•
BC

0 => barotropic instability;
• BT

0 => baroclinic instability;
•
BT > 0 and BC > 0 => mixed instability

Wavelet Spectra
Monterey Bay
Pt. Sur
Surface Temperature
Pt. AN
Surface Velocity

Multi-Scale Energy and Vorticity Analysis
Multi-Scale Window Decomposition in AOSN-II Reanalysis
The reconstructed largescale and meso-scale fields are filtered in the horizontal with features
< 5km removed.
Time windows
Large scale: > 8 days
Meso-scale: 0.5-8 days
Sub-mesoscale: < 0.5 day
Question : How does the large-scale flow lose stability to generate the meso-scale structures?

Multi-Scale Energy and Vorticity Analysis
• Decomposition in space and time (wavelet-based) of energy/vorticity eqns.
Large-scale Available Potential Energy (APE)
Large-scale Kinetic Energy (KE)
• Both APE and KE decrease during the relaxation period
• Transfer from large-scale window to mesoscale window occurs to account for decrease in large-scale energies (as confirmed by transfer and mesoscale terms)
Windows: Large-scale (>= 8days; > 30km), mesoscale (0.5-8 days), and sub-mesoscale (< 0.5 days)
Dr. X. San Liang

Multi-Scale Energy and Vorticity Analysis
MS-EVA Analysis: 11-27 August 2003
Transfer of APE from large-scale to meso-scale
Transfer of KE from large-scale to meso-scale

Multi-Scale Energy and Vorticity Analysis
Process Schematic

Multi-Scale Energy and Vorticity Analysis
Multi-Scale Dynamics
•
Two distinct centers of instability: both of mixed type but different in cause.
•
Center west of Pt. Sur: winds destabilize the ocean directly during upwelling.
•
Center near the Bay: winds enter the balance on the large-scale window and release energy to the mesoscale window during relaxation.
•
Monterey Bay is source region of perturbation and when the wind is relaxed, the generated mesoscale structures propagate northward along the coastline in a surface-intensified free mode of coastal trapped waves.
•
Sub-mesoscale processes and their role in the overall large, mesoscale, submesoscale dynamics are under study.
Energy transfer from meso-scale window to sub-mesoscale window.

Adaptive Sampling and Prediction (ASAP)
Assessing the Effects of Submesoscale Ocean
Parameterizations (AESOP)
Undersea Persistent Surveillance (UPS)
Layered Organization in the Coastal Ocean (LOCO)

• Entering a new era of fully interdisciplinary ocean science: physical-biological-acousticalbiogeochemical
• Advanced ocean prediction systems for science, operations and management: interdisciplinary, multiscale, multi-model ensembles
• Interdisciplinary estimation of state variables and error fields via multivariate physical-biologicalacoustical data assimilation http://www.deas.harvard.edu/~robinson