A Next-generation Operational Global Ocean Nowcast/Forecast System
at the Naval Oceanographic Office
E. Joseph Metzger1, Alan J. Wallcraft1, Prasad G. Thoppil1, James A. Cummings1, Ole Martin Smedstad2, Deborah S. Franklin2, and Harley E. Hurlburt3
1Naval Research Laboratory, Stennis Space Center, MS
2QinetiQ North America, Stennis Space Center, MS
3Florida State University, Tallahassee, FL
Abstract: The new Global Ocean Forecast System (GOFS) version 3.0 was declared an operational product at the Naval Oceanographic Office (NAVOCEANO) on 20 March 2013. It can accurately depict (nowcast) and predict (forecast) the oceanic "weather," some components of which include the 3-D temperature (T), salinity (S) and current structure, the surface mixed layer, and the location of mesoscale features such as eddies, meandering currents and fronts, thus providing improved environmental awareness to the warfighter. The system has undergone significant validation testing in both hindcast and real-time mode, and samples of these will be presented.
Introduction: The development of an advanced global ocean prediction system has been a long-term Navy interest. The space scale of ocean eddies and meandering currents is typically ~100 km and current speeds routinely exceed 1 m/s in the Gulf Stream (Atlantic Ocean) and Kuroshio (Pacific Ocean). Numerical ocean models with sufficiently high horizontal and vertical resolution are needed to depict the 3-D structure with an accuracy superior to climatology and/or persistence (i.e., a forecast of no change). The accelerated development of these prediction systems would not have been possible without the computational resources provided by the DoD HPCMP. Throughout the research and development stages of numerical ocean models and data assimilation techniques, HPC has played a key role. This is especially true with regard to Challenge projects that allowed development of high horizontal resolution global systems long before it became feasible to run them in an operational environment. In addition, non-Challenge and Capability Application Projects have also provided considerable resources toward advancement of these systems.
The new Global Ocean Forecast System (GOFS) version 3.0 was declared an operational product at the Naval Oceanographic Office (NAVOCEANO) on 20 March 2013. This next generation system is based on the HYbrid Coordinate Ocean Model (HYCOM) (Bleck, 2002). It was developed as part of a multi-institutional consortium between academia, government and private industry, funded initially by the National Ocean Partnership Program and in subsequent years by the Office of Naval Research. It replaced the two-model global ocean prediction system based on the Navy Coastal Ocean Model (NCOM) and the Navy Layered Ocean Model. The 1/12° (9 km equatorial resolution) eddy-resolving global HYCOM system has approximately 2.2 times the horizontal resolution as 1/8° (~15 km mid-latitude resolution near 40°N) eddy-permitting NCOM. Eddy-resolving versus eddy-permitting is an important distinction associated with dynamical implications for both ocean model dynamical interpolation skill in the assimilation of ocean data and for ocean model forecast skill (Hurlburt et al., 2008). HYCOM is also uniquely designed to allow an accurate transition between deep and shallow water, historically a challenging problem for ocean models. Its generalized hybrid vertical coordinate is a substantial advance over the vertical coordinate system used in NCOM. The HYCOM-based system represents the world's first eddy-resolving global ocean prediction system with both high horizontal and vertical resolution and has been validated against observational data in hindcast mode by Metzger et al. (2008, 2010).
GOFS 3.0 Description: This version of HYCOM is on a 1/12° global tri-pole grid with 32 hybrid vertical coordinate layers. By replacing the North Pole singularity on a standard Mercator grid with two poles at 47°N we have all three poles over land where they do no harm. Atmospheric forcing is from the 0.5° Navy Operational Global Atmospheric Prediction System (NOGAPS). The truly generalized vertical coordinate can be isopycnal (density tracking — often best in the deep stratified ocean), levels of equal pressure (nearly fixed depths — best used in the mixed layer and unstratified ocean) or terrain-following (often the best choice in shallow water). HYCOM combines all three approaches by choosing the optimal distribution at every grid point and time step. The hybrid coordinate extends the geographic range of applicability of traditional isopycnic coordinate models toward shallow coastal seas and unstratified parts of the world ocean. It maintains the significant advantages of an isopycnal model in stratified regions while allowing more vertical resolution near the surface and in shallow coastal areas, hence providing a better representation of the upper ocean physics.
HYCOM employs the Navy Coupled Ocean Data Assimilation (NCODA) (Cummings, 2005; 2013), which is a fully 3-D variational analysis scheme, to assimilate observational data. The data include surface observations from satellites, including altimeter sea surface height (SSH) anomalies, sea surface temperature (SST), and sea ice concentration, plus in-situ SST observations from ships and buoys as well as T & S profile data from eXpendable BathyThermographs, Conductivity Temperature and Depth sensors and Argo profiling floats. The 3-D ocean environment can be more accurately nowcast and forecast by combining these diverse observational data types via data assimilation and using the dynamical interpolation skill of the model.
Variants of GOFS 3.0 have been running daily in pre-operational mode at the Navy DoD Supercomputing Resource Center (DSRC) since 22 December 2006. Originally running on the IBM machines (Romulus — Power4+ and then Babbage — Power5+), the system was moved to the Cray XT5 (Einstein) and now runs on the IBM iDataPlex machines (Kilrain and Cernan). It is presently configured to use 44 nodes (720 processors) to run HYCOM and produce the NCODA analyses with an additional two nodes set aside for pre- and post-processing. HYCOM efficiently scales to large processor counts and can easily be configured to fit within the operational resource window. Each day, the system starts four days in arrears of the nowcast time (to assimilate all available late arriving observational data) and then runs forward to create a 7-day forecast. It generates 3-D whole domain instantaneous archive files at 3-hourly temporal frequency. These are horizontally interpolated to a constant 1/12° latitude/longitude grid and vertically remapped to 40 z-levels. These output files are then used as boundary conditions for nesting regional ocean models run at NAVOCEANO. Graphical output can be found at http://www7320.nrlssc.navy.mil/GLBhycom1-12/skill.html and numerical output is served to the public at http://www.hycom.org.
Results: The evaluation of GOFS 3.0 nowcast/forecast skill was performed on a year-long hindcast spanning June 2007 - May 2008. Two 48-member sets of 14-day forecasts (using initial conditions from the 1-year hindcast) were also evaluated to examine medium-range forecast skill using different atmospheric forcing. The first used "operational quality" forcing, i.e., the first five days used forecast NOGAPS forcing which was blended toward climatological forcing over a 5-day period and then reverted fully to climatology. The second used "analysis quality" forcing, i.e., analysis NOGAPS forcing for the entire 14-day "forecast" period. No oceanic data were assimilated during the forecast period.
The eddy-resolving capability of GOFS 3.0 is highlighted in Figure 1, using the Kuroshio Extension region as an example. The figure demonstrates how the system can be validated in real time by using drifting buoy trajectories (generally in the form of animations, but here in a sequence of snapshots). Drifting buoy temperature (but not velocity) is assimilated via NCODA, allowing the trajectory to be an independent validation source. The white box in Figure 1a focuses on a warm core eddy about to detach from the Kuroshio and a pair of drifting buoys is noted on the western and eastern sides (the white and black buoy markers (respectively) near 165°E, ~32-34°N in Figure 1b). These two drifting buoys pass within a half degree of each other while traveling in opposite directions. Close examination indicates the two buoys are on opposite sides of a saddle point that still connects the main current with the detaching eddy. This provides an example of HYCOM/NCODA being able to accurately assimilate the satellite altimeter data and act as an accurate dynamical interpolator of the surface information.
To assess medium-range (14-day) forecast skill for the oceanic mesoscale, we compare GOFS 3.0 SSH forecasts against the verifying hindcast analysis and use root mean square error (RMSE) and anomaly correlation (AC) as the metrics. Figure 2 shows forecast skill for the global ocean, the Gulf Stream and Kuroshio Extension (western boundary current regions with strong mesoscale flow instabilities), the South China Sea (a region with both non-deterministic eddy generation west of Luzon Strait and seasonally varying offshore flow), and the relatively shallow Yellow/Bohai Sea (a region in which the ocean responds rapidly and mostly deterministically to the atmospheric forcing). The AC plots (Figure 2a-e) show three curves: forecasts of persistence (cyan), forecasts using operational quality forcing, and .forecasts. that use analysis quality NOGAPS forcing (green). Murphy and Epstein (1989) indicate AC > 0.6 represents useful model forecast skill. Additionally, the black curves on the RMSE plots (Figure 2f-j) represent verification of climatology (i.e., the system's annual mean). Anomaly correlation (RMSE) decreases (increases) with forecast length and forecasts using both operational and analysis quality forcing beat persistence for all regions. The spread between these curves is smallest for the Gulf Stream and Kuroshio Extension regions because of the non-deterministic nature of the mesoscale flow instabilities. Here the predictive skill depends more on the quality of the initial state, the accuracy of the model dynamics, and the time scale of the flow instabilities than on the atmospheric forcing. However, even in these highly variable regions, the system shows forecast skill for the oceanic mesoscale out to 10+ days for the Gulf Stream and 14+ days for the Kuroshio, depending upon the forcing. The Yellow/Bohai Sea is more sensitive to the atmospheric forecast forcing than the initial state; hence the persistence forecast skill diminishes precipitously. The atmospheric forcing has skill out to about two days in this region based on the ocean model response, but then rapidly diverges from the analysis quality forcing curve that remains high for AC (low for RMSE) throughout the forecast period. The next largest spread between the forecasts using operational and analysis quality forcing is for the South China Sea region. This is due to the relatively broad and shallow shelf areas in the southwest part of the domain and the rapidly transitioning nature of the monsoon winds.
A temperature/salinity vs. depth error analysis is shown in Figure 3 using unassimilated profile observations. The top two rows are for the western Pacific Ocean (120-170°E, 20-50°N), which includes the Kuroshio, and the bottom two rows are for the tropical latitudes (±20°). Overall, GOFS 3.0 has cold and fresh biases that exhibit seasonal variations (especially in the more northern region), but the mean error of temperature (salinity) is generally less than 0.25°C (0.05 psu). The system assimilates satellite SST and the bias is very small at the surface in the tropical region (also see section 3.5). Bias is somewhat larger in the western Pacific (although still < 0.2°C) and this is likely due to the mesoscale variability associated with the Kuroshio not being exactly hindcast. The salinity bias is often quite uniform with depth. The RMSE of temperature often peaks between ~50-200 m, i.e., the depth range of high variability associated with the mixed layer and thermocline, whereas the RMSE of salinity is typically highest at the surface and slowly decreases with depth. Overall, GOFS 3.0 has the ability to reproduce the vertical distributions of temperature and salinity with reasonable accuracy.
Future capabilities: New capabilities are also under development to improve ocean and ice forecast skill. In the near future (FY14), the 1/12° global system will implement a new ice model, the Community Ice CodE (CICE) (Hunke and Lipscomb, 2008), which has improved ice dynamics and thermodynamics. This will lead to a more realistic ice environment near the poles in both hemispheres. Most of our development efforts are now focused on increasing the horizontal resolution to 1/25° (~4.5 km at the equator), with a scheduled transition to NAVOCEANO by the end of FY15. This version will also incorporate tidal forcing, which is important because of their interaction with currents, the potential for enhanced vertical mixing, their effect on acoustic propagation and their impact on submarine safety.
Impacts: A next generation ocean nowcasting/forecasting system based on 1/12° global HYCOM is operational at the Navy DSRC. It can accurately depict and forecast such features as western boundary currents and sharp ocean fronts, thus providing improved environmental awareness to the Fleet. Other naval applications include optimum track ship routing, search and rescue, anti-submarine warfare and surveillance, tactical planning, and providing boundary conditions for regional and coastal nested model. HPC resources have played a major role in making this state-of-the-art system feasible, beginning with the preliminary development of HYCOM and continuing all the way through its transition to an operational product.
Bleck, R., 2002: An oceanic general circulation model framed in hybrid isopycnic-Cartesian coordinates. Ocean Modeling, 4, 55-88.
Cummings, J.A., 2005: Operational multivariate ocean data assimilation. Quart. J. Royal Met. Soc., 131, 3583-3604.
Cummings, J.A. and O.M. Smedstad, 2013: Variational data analysis for the global ocean. In: S.K. Park and L. Xu (Eds.), Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications Vol. II., DOI 10.1007/978-3-642-35088-7_13, Springer-Verlag Berlin Heidelberg.
Hunke, E.C. and W. Lipscomb, 2008: CICE: The Los Alamos sea ice model, documentation and software user's manual, version 4.0. Tech. Rep. LA-CC-06-012, Los Alamos National Laboratory, Los Alamos, NM. (http://climate.lanl.gov/models/cice/index.htm).
Hurlburt, H.E., E.P. Chassignet, J.A. Cummings, A.B. Kara, E.J. Metzger, J.F. Shriver, O.M. Smedstad, A.J. Wallcraft, and C.N. Barron, 2008: Eddy-resolving global ocean prediction. In: M. Hecht, H. Hasumi, (Eds.), Ocean Modeling in an Eddying Regime, Geophysical Monograph 177. American Geophysical Union, Washington, DC.
Metzger, E.J., O.M. Smedstad, P.G. Thoppil, H.E. Hurlburt, A.J. Wallcraft, D.S. Franklin, J.F. Shriver and L.F. Smedstad, 2008: Validation Test Report for Global Ocean Prediction System V3.0 - 1/12° HYCOM/NCODA: Phase I, NRL Memo. Report, NRL/MR/7320—08-9148. (http://www7320.nrlssc.navy.mil/pubs.php.)
Metzger, E.J., O.M. Smedstad, P.G. Thoppil, H.E. Hurlburt, D.S. Franklin, G. Peggion, J.F. Shriver and A.J. Wallcraft, 2010: Validation Test Report for Global Ocean Forecast System V3.0 - 1/12° HYCOM/NCODA: Phase II, NRL Memo. Report, NRL/MR/7320—10-9236. (http://www7320.nrlssc.navy.mil/pubs.php.)
Murphy, A.H., Epstein, E.S., 1989. Skill scores and correlation coefficients in model verification. Mon. Weather Rev. 117, 572-581.
Figure 1: Sea surface height (cm) from a) 1/12° global HYCOM/NCODA for the Kuroshio on 1 July 2007. Drifting buoy tracks over a 1 day time period are overlain on each panel. The white box defines the focus area for panels b-g that span the time frame 1-6 July 2007, respectively. A warm core eddy is about to detach from the Kuroshio and two drifting buoys (highlighted in white and black) are traversing its western and eastern sides.
Figure 2: Verification of 14-day ocean forecasts, (a-e) median SSH anomaly correlation and (f-j) median SSH RMSE (cm) versus forecast length (days) in comparison with the verifying analysis for (a,f) the global domain (45°S-45°N), (b,g) the Gulf Stream region (76-40°W, 35-45°N), (c,h) the Kuroshio Extension region (120-179°E, 20-55°N), (d,i) the South China Sea (100-122°E, 0-27°N), and (e,j) the Yellow/Bohai Sea (118-127°E, 30-42°N). The curves depict median statistics over forty-eight 14-day forecasts spanning the period June 2007 - May 2008. The cyan curves verify forecasts of persistence (i.e. no change from the initial state), the red curves use operational quality atmospheric forcing that reverts toward climatology after five days (see Sect. 3.0), and the green curves verify "forecasts" with analysis-quality atmospheric forcing for the duration. The black curve on the RMSE plots is climatology (i.e. annual model mean SSH). Note the range of the y-axis is different on the RMSE panels.
Figure 3: Temperature/salinity vs. depth error analysis in the upper 500 m against unassimilated profiles over the hindcast spanning 1 June 2007 - 31 May 2008 for two regions: (a,b) western Pacific (120-170°E, 20-50°N), and (c,d) the near-tropical region (±180°, 20°S-20°N). The observations are 6 to 24 hours in the future of the HYCOM/NCODA analysis at 00Z. The columns represent the four boreal seasons and annual mean (first column = June-July-August (JJA) summer, second column = September-October-November (SON) fall, third column = December-January-February (DJF) winter, fourth column = March-April-May (MAM) spring, and last column = annual mean). Rows a,c are temperature (°C) and rows b,d are salinity (psu). The black curves are mean error (scale at top of each panel) and the red curves are RMSE (scale along bottom of each panel). The number of unassimilated profiles used in each panel is indicated by N = xxxx. This is the total of near-surface observations and the value decreases with depth, since not all profiles extend to 500 m depth.