Quantitative Impact Assessment of Ocean Observations on Tropical Cyclone Prediction and Ocean Monitoring: Results

from a New, Rigorously-Validated OSE-OSSE System

• George Halliwell, NOAA/AOML/PhOD

• Michael Mehari, CIMAS/Univ. of Miami & NOAA/AOML/PhOD

• Villy Kourafalou, RSMAS/Univ. of Miami

• Robert Atlas, NOAA/AOML

• HeeSook Kang, RSMAS/Univ. of Miami

• Matthieu Le Hénaff, CIMAS/Univ. of Miami & NOAA/AOML/PhOD

• Ioannis Androulidakis, RSMAS/Univ. of Miami

Joint OSEval-TT and DA-TT Meeting

La Spezia, Italy

October 11-13, 2017

Outline

• Brief description of the new ocean OSE-OSSE system developed by the Joint AOML/CIMAS/RSMAS Ocean Modeling and OSSE Center (OMOC)

• Initial applications of the OSE-OSSE system

– Oil spill dispersion in the open Gulf of Mexico

– Improve tropical cyclone intensity (TC) prediction (primary focus of this presentation)

• Impact of ocean observing systems on improving ocean model initialization in coupled hurricane prediction systems

• Summarize key results with respect to improving TC prediction

– OSSEs to evaluate impact of existing ocean observing system components on reducing ocean model initialization errors

– OSE to demonstrate the impact of assimilating the existing ocean observing system on improving intensity prediction (Hurricane Gonzalo, 2014)

– OSSEs to evaluate the impact of rapid-response airborne pre-storm ocean surveys on reducing ocean model initialization errors

• Briefly summarize initial development of a global OSE-OSSE capability

Components of the OMOC OSE-OSSE System

• Nature Run (NR)– Multi-year validated free run by an advanced ocean model (the “truth”)• HYbrid Coordinate Ocean Model (HYCOM) run at 0.04⁰

• Ocean Nowcast-Forecast System• Forecast Model (FM)

• HYCOM with substantially different configuration and lower resolution compared to the NR (“fraternal twin” system)

• Ocean Data Assimilation (DA) procedure• In-house statistical interpolation system designed specifically for HYCOM

Lagrangian vertical coordinates

• Synthetic Observation Simulation Toolbox• Addition of realistic errors

OSSE System Validation Steps (Atlantic hurricane domain)

• 1. Nature Run Evaluation:• The NR accurately reproduces the climatology and variability of

ocean phenomena of interest• Kourafalou et al., 2016, Prog. in Oceanogr.

• Androulidakis et al., 2016, Ocean Dyn.

• 2. Forecast Model Evaluation:

• Errors between the FM and NR have approximately the same magnitude and properties as errors between state-of-the-art ocean models and the real ocean• Halliwell et al., 2017, J. Oper. Oceanogr.

• 3. OSSE System Evaluation

• Compare OSSEs to reference OSEs that assimilate the identical sets of synthetic and real observations, respectively• Halliwell et al., 2017, J. Oper. Oceanogr.

• OSSEs demonstrate slightly higher observing system impacts (average ~10% based on RMSE) compared to OSEs (calibration needed)

• OSSE results were calibrated to correct for this overestimate

5

OSSE to Evaluate Existing Ocean Observing System Components During the 2014 Hurricane Season (June through October)

Argo floats

XBT

Argo floatsXBT transectsIn-situ SSTSatellite SST (not shown)Altimetry (not shown)

OSSE Analysis Domain – Skill Score Analysis

Will show impact on reducing errors in Tropical Cyclone Heat Potential (TCHP) over the enclosed analysis domain.

26

0TCHP 26 ,

H

pc T z dz

expt=experiment evaluatedref = unconstrained FM

MSE is calculated daily versus truth provided by the NR

7

Control (assim. all obs.)deny altimetrydeny Argo floatsdeny XBTdeny all SST

Control (assim. all obs.)Altimetry onlyArgo floats onlyXBT onlySST (sat. + in-situ) only

Initialize from unconstrained FM

Initialize from Control experiment

Data denial experiments

Individual observing system experiments

Time series of daily skill score over analysis domain

OSE to demonstrate observing system impact on hurricane intensity prediction

• An ocean OSE run during the 2014 hurricane season is used to evaluate the impact of the existing operational ocean observing system on coupled model intensity forecasts for Hurricane Gonzalo

– CTRL – Control experiment that assimilates all observations– NODA – Experiment that denies all observations

• Ocean analyses from these two experiments are used to initialize the HYCOM-HWRF regional coupled TC prediction system to performs the forecasts

Min. p

Max. v

Hurricane Gonzalo ForecastsCompared to Best Track

Ocean obs. assimilation improves intensity

TCHP

SST

Hurricane Gonzalo Initial Ocean Fields

OBS (from AOML/PhOD)

OBS (from RSS)

OSSE Quantitative Assessment of Synthetic Airborne Surveys Conducted Prior to Hurricane Gonzalo (2014)

Two types of surveys deploying synthetic profilers conducted for each storm:

1. Deploy AXBTs to a depth of 400 m2. Deploy AXCTDs to a depth of 1000 m

Surveys conducted about two days before storm arrival.

Profile assimilation was added to two experiments:

CONTROL – assimilates synthetic version of existing ocean observing system components (altimetry, Argo, ship XBT, satellite and in-situ SST)

NOALT – same as CONTROL but denies all 4 available altimeters

RMS error (RMSE) is analyzed within the parallelogram containing the profiles.

RMSE primarily represents errors in mesoscale structure

Correction Resulting from Synthetic AXCTD Profiles

-19% -37%-28%

• Additional corrections resulting from adding rapid-response profiles to CONTROL are confined to the immediate survey region– Correlation scales are short [O(100 km)]– Little time for impacts to spread

Additional percentage RMSE reduction over CONTROL resulting from profile assimilation is shown in each panel.

For maximum benefit, rapid-response pre-storm profile surveys should cover as large an area as possible.

Percentage RMSE Reduction With Respect to NR

Without altimetry assimilation

NOALTadd AXBTadd AXCTD

Large reduction in mesoscale errors in both dynamical (SSH) and thermodynamical (TCHP, δT0-100) fields

Additional reduction in mesoscale errors for AXCTDs vs. AXBTs in SSH only.

Error reduction is referenced to the error of the unconstrained FM simulation

NOALT AXBT AXCTD

Percentage RMSE Reduction With Respect to NR

With altimetry assimilation

CONTROLadd AXBTadd AXCTD

Substantial reduction in mesoscale errors in thermodynamical fields only.

Minimal additional reduction in mesoscale errors for AXCTDs vs. AXBTs in SSH.

Error reduction is referenced to the error of the unconstrained FM simulation

CONTROL AXBT AXCTD

Impact of Existing Ocean Observing Systems

• Large mesoscale error reduction in SSH –> primarily from altimetry assimilation• Smaller mesoscale error reduction in TCHP and other thermodynamical fields• Large regions of positive and negative bias remain in TCHP• Rapid-response surveys have the greatest impact on correcting model

thermodynamical fields for these reasons

Unconstrained FM Error (vs. NR) CONTROL Error (vs. NR)

Snapshots, 14 September 2014

TCHP Bias Reduction

Storm Experiment TCHP bias

(kJ cm-2)

TCHP bias

magnitude

percent reduction

from CONTROL

Gonzalo

CONTROL 2.42

AXBT2D 0.35 85%

AXCTD2D 0.07 97%

Although the initial bias was small, it was effectively reduced by assimilation of airborne profilers.

Impacts of Global Ocean Observing Systems on the Capability to Monitor Indicators of Seasonal-to-Interannual Variability using Ocean Analysis-Forecast Systems

Goals:1. Demonstrate the capability of ocean analysis-

forecast systems to represent observed indicators of seasonal-to-interannual ocean variability

2. Perform OSEs to determine the impact of individual components of the global ocean observing system on the accurate representation of these indicators versus observations

Meridional heat flux across MOCHA array

ObservedUnconstrained model

Indicators:Time series of observation-based indicators will be extracted from the model to determine impacts.These indicators include:1. AMOC transport and MHF at RAPID-MOCHA

array (figure at right)2. AMOC and MHF at other latitudes3. Transport at choke points4. WHWP5. ENSO indices

Future Plans

• Perform OSEs and OSSEs for the 2017 Atlantic hurricanes• Extensive ocean observations were collected in major storms

• Extend OSSEs to evaluate impacts on coupled hurricane intensity prediction

• Continue to develop global OSE, and eventually OSSE, capabilities– Quantitatively assess global ocean observing system components with respect to

monitoring seasonal-to-interannual ocean variability

– Unconstrained global simulations by the NR and FM can provide initial and boundary conditions for higher-resolution regional OSSE systems that evaluate observations with respect to short-term ocean forecast applications such as hurricane prediction.

• Longer-term goals– Extend OSSE system capabilities to the nearshore ocean and shallow seas

• Present system designed for the deep ocean only

• Collaboration with COSS-TT (oral session in Ocean Sciences 2018)

– Perform OSEs and OSSEs that focus on other short-term forecast applications

– Embed new data assimilation methods in the OSSE system

• Validated and calibrated OSSE systems can be used to evaluate and compare different DA procedures