impact OF satellite OBSERVATIONS
on Sea Surface Temperature forecasts via variational data assimilation AND HEAT
flux calibration
Charlie
N. Barron(1), Clark
Rowley(1), Scott R. Smith(1), Jackie May(1),
Jan M. Dastugue(1), Peter L. Spence(2), and Silvia Gremes-Cordero(3)
(1) Naval Research Laboratory, Code 7321,
Stennis Space Center, MS, 39529, (USA)
(2) Vencore, Stennis Space Center, MS, 39529,
(USA)
(3) University of New Orleans, Stennis Space
Center, MS, 39529, (USA)
ABSTRACT
Satellite observations are used
to guide forecasts of sea surface temperature (SST) through variational data
assimilation and heat flux calibration. In the experiments considered,
assimilation is conducted using the Navy Coupled Ocean Data Assimilation
(NCODA) in either a standard 3D variational (3DVAR) or an alternative 4DVAR
formulation. Heat flux for the forecasts follows the original operational
highest-quality time series or modifies the flux-determining fields using the
Naval Research Laboratory ocean flux (NFLUX) capability. These alternatives are
evaluated relative to independent, unassimilated in situ sea surface
temperature (SST) observations in two sets of year-long experiments, sets based
on atmospheric fields from either the global or the regional operational
atmospheric model. Each set begins with a control run with standard forcing and
standard 3DVAR assimilation, and the experimental variants employ the various
combinations of 4DVAR assimilation and NFLUX-modified forcing. Results in the
California Current region demonstrate that the combination of 4DVAR
assimilation with NFLUX-modified forcing tends to produce forecasts in best
overall agreement with independent in situ observations.
1.
Introduction
Satellite observations support a
variety of avenues to improve sea surface temperature (SST) forecasts. The
upwelling visible, infrared and microwave radiation intensities integrated
across various wavelength bands or channels are sensitive to various physical
properties of the atmosphere and ocean. In this way satellite instruments can
offer information not only on SST itself but also on other ocean and atmospheric
properties that influence heat flux and therefore the evolution of SST over the
forecast. Among these properties are atmospheric temperature, relative
humidity, temperature and humidity from the upper atmosphere to the ocean
surface, and cloud cover. Use of these satellite observations is categorized
either within systems to modify heat fluxes or as assimilation to refine the
ocean and possibly boundary layer states.
Brief overviews of the Naval Research
Laboratory Ocean Surface Flux (NFLUX) and 3D and 4D variational (3DVAR/4DVAR)
data assimilation in the Navy Coupled Ocean Data Assimilation (NCODA; Cummings,
2005) are in Section 2, with greater detail available in the references and in the
GHRSST XVI proceedings (Barron et al., 2016). Section 3 reports on experiments
in the California Current and northern Arabian Sea regions. Section 4
summarizes our conclusions to date and projects future developments relating to
NFLUX and the COFFEE project.
2.
Methods
The experiments examining the
effectiveness of heat flux correction and variational ocean data assimilation
for reducing forecast errors of SST rely on two capabilities recently developed
and introduced for use in U.S. Navy ocean forecast systems: NFLUX (May et al.,
2016, 2014; Van de Voorde et al., 2015) and NCODA 4DVAR (Smith et al., 2015).
NFLUX combines satellite observations or retrievals related to wind speed, air
and sea surface temperature, atmospheric temperature and moisture profiles, and
cloud conditions with other operational products or databases of aerosols,
trace gases, and other properties to provide more accurate estimates of the
ocean and atmospheric properties related to various components of heat flux.
Heat flux is partitioned into its constituent components of shortwave (or
solar), longwave, sensible, and latent heat flux. Flux estimates are expressed
in terms of COARE 3.0 bulk flux algorithm (Fairall et al., 2003; Wallcraft et
al., 2008) for coupling with ocean models. Radiant heat flux components are
estimated using the Rapid Radiative Transfer Model for Global circulation
models (RRTM-G; Iacono et al., 2000). NFLUX can produce estimates of flux
fields using swath-level observations from satellites; these are interpolated
to produce full-field estimates using 2DVAR assimilation with background fields
from regional Navy Coupled Ocean Atmosphere Prediction System (COAMPS) or
global Navy Global Environmental Model (NAVGEM) forecasts.
The second capability evaluated
in these experiments is data assimilation of satellite altimeter or SST
observations and subsurface observations of temperature and salinity using
NCODA 3DVAR (Smith et al., 2012) or 4DVAR capabilities. NCODA 3DVAR has been
the standard assimilation in Navy ocean prediction systems, and 4DVAR is a
recently-introduced capability that is anticipated to provide greater forecast
skill in priority regions. A primary difference between these capabilities is
that 3DVAR only modifies the initial model state at nowcast time, while 4DVAR modifies
the model trajectory over a recent hindcast period, typically 3
days, to adjust not only the
nowcast state but also the trajectory and dynamic balance of the model in the
window leading up to the nowcast. It is anticipated that the increased computational
cost of 4DVAR assimilation will reduce forecast error by not only correcting
the nowcast but also the dynamic balance leading into and during the forecast.
3.
Experiments
Experiments are conducted using
the Navy Coupled Ocean Model (NCOM; Barron et al., 2006; Rowley and Mask, 2015)
from May 2013-April 2014 in two domains (Fig. 1): a California Current region
and a northern Arabian Sea region. In each region, a set of four
experimental cases is run for each of the atmospheric forcing cases, the
regional COAMPS and global NAVGEM. In each set of experiments, the control run
uses the original atmospheric forcing and standard 3DVAR assimilation, while
the experimental variants use standard or NFLUX-modified heat fluxes combined
with 3DVAR or 4DVAR assimilation. The forecasts cycle daily with assimilation
of satellite SST (GOES, AVHRR, VIIRS), altimeter (Jason, Altika), and in situ
temperature and salinity profile observations. Surface-only in situ data are
not assimilated; these are a means of independent validation. Other
observations are independent when used to evaluate the forecast period, as the
daily assimilation includes no data measured after the 00:00 UTC analysis. Each
experimental case starts in April 2013 as initialized from the operational
global run, allowing a one-month spin-up before the 12-month evaluation.
Table 1 shows evaluation
of forecasts out to 96-hours from the California Current experiments using
NAVGEM or COAMPS forcing as the background. Matchups are interpolated
horizontally and in time from the 3-hourly forecast fields on the ~3.5 km model
grids. Temporal interpolation treats the forecast fields as a continuous time
series sampled at every three hours; for example, the matchups labeled 24 h are
sampled from the time series beginning with the 29 April 2016 24 UTC forecast
to the 03-24 hour forecasts from 30 April 2016, 03-24 hour forecasts from 1 May
2013, and continuing in that manner to the 24 UTC forecast field from 29 April
2014. In that way the 24 h forecast time series has three-hourly forecast
fields valid at the times from 00 UTC 1 May 2013 through 24 UTC 30 April 2014. In
general, the cases using NFLUX modification of the heat fluxes outperform those
with the standard unmodified fluxes, and the cases with 4DVAR assimilation have
smaller errors than those using 3DVAR assimilation.
Similar statistics are shown in
Table 2 for the 3DVAR assimilation cases in the northern Arabian Sea
forecasts over the same time period. While the 57M matchups indicate that the
NFLUX cases have accuracy in general similar or superior to the accuracy of
cases with standard fluxes, spurious forecasts with matchups errors as large as
12°C cold have been identified very nearshore along the northern coast of
Qatar. Such large negative biases appear to be due to errors in the longwave
terms where the NFLUX estimates are contaminated surface temperature values appropriate
for land regions rather than water cells. This may be a consequence of
imprecision in aligning coarser land/sea masks appropriate for atmospheric
products with higher-resolution land/sea masks corresponding to the ocean
model. Work continues to resolve these discrepancies and complete a corrected
set of northern Arabian Sea cases. Additional development is underway to
extend NFLUX corrections into the forecast period and extend the 4DVAR
assimilation into the atmospheric boundary layer. Examples of using NFLUX in
other applications are reported in Rowley et al., 2015.
4.
Conclusion
COFFEE uses satellite-based heat
flux corrections and 3D/4D variational assimilation capabilities to enable more
accurate SST forecasts. Year-long results (May 2013-April 2014) in the
California Current indicate that forecast skill is generally improved through
the use of NFLUX corrections combined with 4DVAR assimilation. Preliminary
results in the northern Arabian Sea similarly support the use of NFLUX
corrections; issues in longwave flux corrections will be resolved before
completing the Arabian Sea cases. Work is proceeding on extending corrections
in a forecast mode in short term forecasts, providing a capability that is
responsive to environmental and forecast system changes. Demonstration of these
capabilities in these regional cases is a first step in establishing their
applicability in other regions and globally. Such a capability is envisioned to
play a role in mediating imbalances between components of regional and global
coupled modeling systems.
5.
Acknowledgements
Work under the Calibration of
Ocean Forcing with satellite Flux Estimates (COFFEE) project was supported by
the Naval Research Laboratory and the Office of Naval Research, which further
supported participation in GHRSST XVI and preparation of these through the
Multisensor Improved Sea Surface Temperature for Integrated Ocean Observing
System (MISST-IOOS) project.
6.
References
Barron, C.N., J.M.
Dastugue, J. May, C. Rowley, S.R. Smith, P.L. Spence, and S. Gremes-Cordero,
2016: Forecast of SST: calibration of ocean forcing with satellite flux
estimates (COFFEE). Proceedings of GHRSST XVI, 20-24 July 2015, ESTEC,
Netherlands, 196-201.
Barron, C.N., A.B. Kara, P.J. Martin, R.C. Rhodes, and L.F. Smedstad,
2006: Formulation, implementation and examination of vertical coordinate
choices in the global Navy Coastal Ocean Model (NCOM). Ocean Modelling 11(3-4),
347-375, doi:10.1016/j.ocemod.2005.01.004.
Cummings, J.A., 2005: Operational multivariate ocean data
assimilation. Quart. J. Roy. Met. Soc. 131, 3583-3604.
Fairall, C.W., E.F. Bradley, J.E. Hare, A.A. Grachev, and J.B. Edson,
2003: Bulk parameterization of air sea fluxes: updates and verification for the
COARE algorithm. J. Climate, 16, 571-591.
Iacono, M.J., E.J. Mlawer, S.A. Clough, and J.J. Morcette, 2000:
Impact of an improved longwave radiation model, RRTM, on the energy budget and
thermodynamic properties of the NCAR community climate model, CCM3. J. Phys.
Oceanography, 105(D11), 14,873-14,890.
May, J.C., C. Rowley, and N. Van de Voorde, 2016: The Naval Research
Laboratory ocean surface flux (NFLUX) system: Satellite-based turbulent heat
flux products. Journal of Applied Meteorology and Climatology, 55,
1221-1237, DOI 10.1175/JAMC-D-15-0187.1.
May, J., N. Van de Voorde, and C. Rowley, 2014: Validation test report
for the NRL ocean flux (NFLUX) quality control and 3d variational analysis
system. NRL Memorandum Report, NRL/MR/7320--14-9524.
Rowley, C., C. Barron, G. Jacobs, and J. May, 2015: Ocean surface
structure assimilation at NRL. Joint Center for Satellite Data Assimilation
Quarterly, 51, 1-4.
Rowley, C., and A. Mask, 2014: Regional and coastal prediction with
the relocatable ocean nowcast/forecast system. Oceanography, 27(3),
44-55, DOI 10.5670/oceanog.2014.67.
Smith, S.R., M.J. Carrier, H.E. Ngodock, J. Shriver, and P.
Muscarella, 2015: Validation Testing Report for the Navy Coastal Ocean Model
Four-Dimensional Variational Assimilation (NCOM 4DVAR) System, Version 1.0.
NRL Memorandum Report NRL/MR/7320-15-9574.
Smith, S.R., J.A. Cummings, P. Spence, S.N. Carroll, C. Rowley, O.M.
Smedstad, P. Chu, B. Lunde, J. Shriver and R. Helber, 2012. Validation Test
Report for the Navy Coupled Ocean Data Assimilation 3D Variational Analysis
(NCODA-VAR) System, Version 3.43. NRL Memorandum Report, NRL/MR/7320-12-9363.
Van de Voorde, N.E., J. May, and C. Rowley, 2015: NFLUX PRE:
Validation of new specific humidity, surface air temperature, and wind speed
algorithms for ascending/descending directions and clear or cloudy conditions.
NRL Memorandum Report, NRL/MR/7320-15-9611.
Wallcraft, A.J., A.B. Kara, H.E. Hurlburt, E.P. Chassignet, and G.H.
Halliwell, 2008: Value of bulk heat flux parameterizations for ocean SST
prediction. J. Mar. Sys., 74, doi:10.1016/j.marsys.2008.01.009.