1. Introduction
The global oceans cover roughly 70% of the Earth’s surface with a heat capacity over the upper 2.5-m equivalent to that of the entire atmosphere. The oceans exhibit important variability on a range of time scales, from hourly variations of sea state and sea surface temperature (SST) to longer-term changes (e.g., related to the meridional overturning circulation). The ocean plays an active role in numerous phenomena associated with strongly coupled air–sea interactions, such as El Niño–Southern Oscillation and Pacific decadal variability. As a result, coupled atmosphere–ocean general circulation models have been used extensively for seasonal predictions and climate projections. However, the inclusion of coupled interactions across the air–sea interface has generally been considered less important for numerical weather prediction (NWP; Bauer et al. 2015). Indeed, all currently operational global deterministic medium-range NWP systems employ persistent surface marine conditions with no coupled ice–ocean model (Brassington et al. 2015). This is mainly because the dominant time scales of ocean variability are longer than those of medium-range weather forecasting, thereby allowing the parameterization of coupled air–sea interactions. Additionally, mature operational oceanographic forecasting systems have not been available in the past for coupling with NWP.
The importance of coupling between the atmosphere and the ocean for forecasting on time scales of hours to weeks has been demonstrated for a range of physical processes (Belcher et al. 2015) associated with variability of both the sea surface temperature and sea ice cover. Of these, perhaps the most studied has been the strong role of atmosphere–ocean coupling for the development of tropical cyclones (TCs). A well-known feature associated with tropical cyclones is the presence of a cold SST anomaly that usually develops in the right-rear quadrant, known as the “cold wake.” This cold wake can exhibit SST anomalies of 1°–6°C (Price 1981), which may develop over several hours to a few days. Idealized numerical studies have shown that the presence of a cold wake can affect the intensification of TCs by as much as 50% (Schade and Emanuel 1999). Using a regional coupled atmosphere–ocean model of the South Pacific Ocean run over a 20-yr period, Jullien et al. (2014) estimate this reduced intensification to be about 15 hPa °C−1. Interestingly, Jullien et al. (2014) also note an impact of coupling, more generally, on cyclogenesis patterns in the coupled model that are closer to observations.
Numerous case studies have been made demonstrating the strong impact of atmosphere–ocean coupling on TC development (e.g., Bender and Ginis 2000). Case studies using high-resolution coupled models have shown that coupling with an ocean model can reduce the TC minimum sea level pressure (SLP) by up to 10 hPa under certain conditions (Bender and Ginis 2000; Sandery et al. 2010; Ito et al. 2015). Indeed, Bender and Ginis (2000) show a 26% improvement in mean absolute error of TC central SLP. The presence of mesoscale eddies has been shown to affect the intensification (Hong et al. 2000; Jacob and Shay 2003; Zheng et al. 2010; Ma et al. 2017), highlighting the need for high-resolution eddy-permitting ocean models to accurately assess the role of coupling.
Using a case study of Hurricane Katrina (2005), Chen et al. (2010) quantify the separate contributions of ocean vertical mixing, vertical advection, and horizontal advection in the formation of the cold wake. Moreover, they show how the cold wake in turn affects not only the intensification, but also the atmospheric structure of the storm. In particular, they find an increase in the maximum in eye size by 10 km and a reduction of 10% in hurricane-force wind radius, with impacts in the mid- and upper troposphere. Lee and Chen (2012) extend this work through coupling to a wave model and demonstrate the complex three-dimensional asymmetric structures that develop in both thermodynamic and dynamic properties of the hurricane boundary layer.
Studies have also highlighted the potential for coastal baroclinic processes to generate cold anomalies ahead of the hurricane eye (Glenn et al. 2016; Seroka et al. 2016). Moreover, the presence of this upstream cold anomaly is found to be the dominant feature responsible for rapid deintensification of Hurricane Irene (2011) prior to landfall in a series of more than 140 sensitivity tests (Seroka et al. 2016).
In addition to TCs, the role of coupling has been identified for a number of other phenomena, such as monsoons (Fu et al. 2007), the Madden–Julian oscillation (Bernie et al. 2008; Vitart et al. 2008; Shelly et al. 2014; Seo et al. 2014), ocean fronts and eddies (Small et al. 2008; Putrasahan et al. 2013), and coastal features (Seo et al. 2007; Small et al. 2011; Renault et al. 2012). Chelton and Xie (2010) provide a review, with notable impact from coupling due to coastal upwelling and oceanic fronts. Additionally, Pellerin et al. (2004) show that coupling in ice-infested seas can also have strong impacts on subdaily time scales due to rapid changes in coastal sea ice cover (i.e., the formation of coastal polynyas). As the sea ice acts as a barrier between a relatively warm–wet ocean and cold–dry atmosphere, changes in the sea ice cover can have dramatic effects on heat and moisture fluxes, resulting in changes of more than 6°C in less than 24 h. The importance of coupling in polar regions, more generally, has also been recognized (Jung et al. 2016).
A common challenge noted in many of these studies is with regards to how to properly initialize the ocean component of coupled forecasts (Winterbottom et al. 2012; Mulholland et al. 2015). Indeed, several studies found that the impact of coupling sensitively depends on how the system is initialized (Bender and Ginis 2000; Sandery et al. 2010). As a result, focused efforts have been made to develop ocean analysis methods specifically for coupled initialization (Saha et al. 2010; Sandery and O’Kane 2014; Laloyaux et al. 2016; Lea et al. 2015).
While positive impacts of coupling have been shown in the abovementioned studies, a comprehensive evaluation in an operational global NWP framework for forecasts of up to 10 days has yet to show positive results. Jung and Vitart (2006) find that coupling with a 1° resolution ocean model over Northern Hemisphere winter has no significant impact on northern extratropical SLP forecasts. Similarly, Nicholls and Decker (2015) investigate the impact of coupling on nor’easters and find a limited impact of coupling. The lack of positive impact of these studies may be related to the particular details of the coupled model, coupling methodology, and initialization method, as these have all been shown to affect the impact of coupling (e.g., Scoccimarro et al. 2016). Moreover, Jung and Vitart (2006) focused their evaluation on the Northern Hemisphere in winter and note that a more positive impact may be found in the tropics. Indeed, a related study using a similar model configuration found coupling to improve 850-hPa temperatures in the tropics for forecast of 10–15 days (Vitart et al. 2008).
Here, we provide an evaluation of the impact of an interactive air–sea coupling between an operational global deterministic medium-range weather forecasting system and a global ice–ocean prediction system. The ice–ocean prediction system was designed specifically for coupling with the atmosphere, with particular attention paid to the assimilation of surface fields (SST and ice) in order to minimize initialization shock in the coupled forecasts. This coupled forecasting system is now running operationally at the Canadian Centre for Meteorological and Environmental Prediction (CCMEP). We show that while the largest impact is indeed found to be associated with reduced cyclone intensification, the impact of this effect is felt over large spatial scales with positive global implications for forecast skill in both summer and winter seasons. To the authors’ knowledge, this is the first time a peer-reviewed evaluation of coupling has been made in a global medium-range deterministic NWP framework showing a statistically significant positive impact.
Section 2 provides a description of the atmosphere, ocean, and sea ice models together with a description of the coupling methodology, initialization method, forecast experiment design, and verification metrics used. Section 3 presents an evaluation of the sea surface temperature forecasts to demonstrate the impact of an evolving sea surface in the coupled model. As the dominant impact of coupling is found to be on TC intensification, a case study illustrating the effect of coupling for a typhoon in the western Pacific Ocean is presented in section 4. Statistics from a TC-tracking algorithm are provided in section 5. Section 6 presents standard verification scores of coupled and forced (uncoupled) forecasts against independent analyses, as well as a more detailed examination highlighting the impact on error growth in atmospheric forecasts. The overall impact of coupling is also assessed, as compared to the global radiosonde network (section 7). Changes in precipitation fields, as compared to satellite estimates, are then shown in section 8. Finally, conclusions and a discussion of results are presented in section 9.
2. System description and evaluation methodology
The coupled atmosphere–ice–ocean forecasting system used here follows as closely as possible the existing (uncoupled) operational weather and ice–ocean prediction systems in place at CCMEP: namely, the Global Deterministic Prediction System (GDPS; v5.0.0) and the Global Ice–Ocean Prediction System (GIOPS; v2.1.0). A description of these systems, together with the coupling and initialization approaches, is provided below.
a. Numerical models
The numerical model developed and maintained at CCMEP for operational NWP is the Global Environmental Multiscale (GEM) atmospheric model (Côté et al. 1998a,b). The most recent implementation of the GDPS (v5.0.0; CMC 2015) uses GEM in hydrostatic mode with a horizontal resolution of 25 km on a Yin–Yang grid (Qaddouri and Lee 2011). A staggered hybrid log-pressure vertical coordinate is used (Girard et al. 2014), with 80 vertical levels and a model top at 0.1 hPa. A description of the physical parameterizations used by the GDPS is provided in Table 1. A full description of model settings and a detailed evaluation can be found in Bélair et al. (2005), Charron et al. (2012), Zadra et al. (2014), and CMC (2015).
Physical parameterizations used in the GDPS.
The ocean model used in GIOPS is based on version 3.1 of Nucleus for European Modelling of the Ocean (NEMO; Madec et al. 1998; Madec 2008), with additional contributions from the DRAKKAR consortium (Barnier et al. 2007) and Mercator Océan (Lellouche et al. 2013). NEMO is a primitive equation z-level model applying the hydrostatic and Boussinesq approximations, a linear free surface (Roullet and Madec 2000), and partial cell topography (Adcroft et al. 1997). The version used by GIOPS has a tripolar ORCA grid with 50 levels in the vertical, with vertical spacing increasing from 1 m at the surface to 500 m at the ocean bottom. The model configuration has an eddy-permitting global 1/4° resolution (referred to as ORCA025). The sea ice model is based on version 4.0 of the Los Alamos multicategory Community Ice Model (CICE; Hunke 2001; Lipscomb et al. 2007; Hunke and Lipscomb 2010), which uses an elasto–viscous–plastic (EVP) solution (Hunke and Dukowicz 1997, 2002; Bouillon et al. 2009) and 10 ice thickness categories. The Met Office NEMO–CICE4 interface is used (Megann et al. 2014).
The NEMO–CICE configuration used here is as described in Smith et al. (2016a) and Roy et al. (2015). The surface roughness of sea ice is set at the same value used by GEM (1.6 × 10−4 m). As demonstrated in Roy et al. (2015), this improves the consistency in momentum fluxes between the atmosphere and ice, resulting in improved sea ice thicknesses. A parameterization commonly used in NEMO to allow for the penetration of turbulent kinetic energy due to internal and inertial waves was deactivated, as this was found to degrade forecasts of sea surface temperature and, most notably, sea ice cover (Smith et al. 2013). Care has also been taken to ensure numerical convergence of the EVP solution, as this has been shown to have an effect on the lead fraction in the pack ice (Lemieux et al. 2012), and thus, on heat and moisture fluxes in coupled forecasts. Additionally, a parameterization for landfast ice, based on the effect of grounded ice keels, is used (Lemieux et al. 2015), as this can affect the formation of coastal polynyas (areas of open water along the coast).
b. Coupling of GEM and NEMO–CICE
The approach employed here to couple the atmospheric and ice–ocean models differs somewhat from methods used in other Earth system models (ESMs; e.g., Valcke et al. 2012). In most ESMs, an external “coupler” is used that manages data transfer, interpolates model fields between the different grids, and coordinates execution. Here, the strategy is to implement the coupling such that it has a minimal impact on the operational production suites for the respective forecasting systems, while still allowing for a bidirectional transfer of momentum, heat, and moisture at the snow–ice and ocean surface between the models. To maximally exploit the information content of the models, fluxes are calculated using the higher resolution of the two grids (i.e., the NEMO grid in this case), and fields are exchanged at every common time step (i.e., every 720 s). The coupler is used only to exchange data between the models, with regriding made in the respective models (NEMO regrids the atmospheric fields, and GEM regrids the fluxes calculated by NEMO). Calculating the surface layer turbulent and upward radiative fluxes within the ice–ocean model also allows for a more detailed budget by taking into account the different ice and snow thickness categories used by the multicategory ice model.
The coupler was developed specifically for use at CCMEP and is called the Globally Organized System for Simulation Information Passing (GOSSIP). To exchange information, each model sends fields via the GOSSIP TCP/IP server after gathering MPI tiles on its own global grid. Global fields are then exchanged back through GOSSIP by the master process of each model, before being broadcast and remapped to the MPI tiles of the receiving grid. The regriding is then done in parallel by applying precomputed interpolation–aggregation weights. The choice between interpolation and aggregation depends on the local differences in grid resolution. If the destination grid is lower resolution than the source grid, aggregation is triggered (when a destination cell covers at least three source grid points). In other regions, a bilinear interpolation is applied.
The purpose of these exchanges is to simulate interactive and consistent transfers of momentum, heat, and moisture between the two models. Therefore, a flux coupling approach is used based on common shared atmospheric, oceanic, and ice state variables. In a first step, the ice–ocean model receives downward radiative fluxes and state variables from the atmospheric model (valid at its bottom prognostic level). These are used in a second step by the ice–ocean model to compute surface layer (SL) atmospheric fluxes, both over open water and for each of the ice thickness categories. The aggregated ocean and ice SL fluxes are then sent to the atmospheric model, where they are subsequently regridded and further aggregated as part of the atmospheric model’s tiling scheme. The surface upward longwave radiative flux is also computed and aggregated by the ice–ocean model based on the different surface temperatures (snow, ice, or ocean). The time-stepping approach is implemented such that the atmospheric model moves forward one step ahead of the ice–ocean model prior to sending its variables (after having received initial state variables from the ice–ocean model in a separate exchange step at the beginning of the simulation). As a result, there is a 1-time-step phase lag in the fields used by each model. The number of processors used by each model is determined such that both models take the same amount of time to complete 1 time step, allowing for an efficient use of computing resources.
We compute the SL turbulent fluxes using the formulae described in section 2.2 of Roy et al. (2015). The formulae are applied here in the same way, except that instead of being read from a forcing dataset, the prognostic surface wind, air temperature, and specific humidity are provided by the atmospheric model through the GOSSIP server. Stability functions over the ocean and ice are applied following the Benoit–Delage–Girard (BDG; Benoit et al. 1989; Delage 1997; Delage and Girard 1992) and Jordan (JN; Jordan et al. 1999) schemes, respectively, described by Roy et al. (2015). The BDG scheme is reproduced in the ice–ocean model using the atmospheric model code as an external library.
c. Initialization method
Four separate analysis systems are used to initialize the coupled atmosphere–ice–ocean forecasts evaluated here (atmosphere, SST, ice concentration, and ocean). The approach taken here has been to develop an ocean data assimilation system (required to initialize coupled forecasts) that provides the most consistent marine surface conditions, as compared to that used in the atmospheric data assimilation system, in order to reduce initialization shock. Moreover, no changes are made to the atmosphere, ice, or SST analyses, thereby simplifying the implementation of the coupled model in an operational context.
The connections among the four analysis systems are illustrated schematically in Fig. 1. A four-dimensional ensemble–variational (4DEnVar) data assimilation scheme is used to initialize GEM (Buehner et al. 2013, 2015), including a 4D incremental analysis update approach. Analysis increments are produced on a 0.45° × 0.45° horizontal Gaussian grid every 6 h. Surface marine conditions are provided by a daily SST analysis (updated at 1200 UTC) produced using an optimal interpolation (OI) method (Brasnett 2008; Martin et al. 2012) on a global 0.2° grid and a three-dimensional variational (3DVar) ice concentration analysis (Buehner et al. 2016) that produces four analyses a day. The two surface marine analyses both use the previous analysis as background (i.e., a model is not used to produce trial fields). The SST and ice concentration analyses both use the other in some part of their processing. The SST analysis system assimilates pseudo-observations of SST at the freezing point based on the ice concentration analyses. Similarly, the 3DVar ice analysis system uses the SST analysis as part of the quality control procedure to reject erroneous ice concentration observations in areas of relatively warm waters. As noted above, no modification is made to these analyses, as compared to how they are used operationally in the GDPS and for the uncoupled forecasts presented here.
Schematic illustration of the four analysis systems used to initialize the coupled model: a 4DEnVar system for atmosphere (green), a 3DVar ice concentration analysis (blue), an OI SST analysis (red), and a SEEK filter providing an ocean analysis (gray). The exchange of analyses is illustrated with arrows: black arrows show the recycling of fields within a particular analysis system (i.e., background state), red arrows show how SST fields are used, and blue arrows represent the exchange of ice analyses. Note that both the atmosphere and ocean analysis systems use incremental updating procedures that are not represented here.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
To provide skillful coupled forecasts, it is important that the ice–ocean component can be initialized in such a way as to minimize inconsistencies with the atmospheric and surface marine analyses, which may result in sudden changes in surface fluxes (initialization shock). The manner in which the analysis system is implemented for GIOPS was conceived specifically with this in mind. The ocean analysis is produced using the System d’Assimilation Mercator version 2 (SAM2) developed by Mercator Océan (Tranchant et al. 2008; Cummings et al. 2009; Lellouche et al. 2013). The analysis method is based on a reduced order Kalman filter using a singular evolutive extended Kalman (SEEK) formulation (Pham et al. 1998). Background error covariances are modeled by an ensemble of multivariate three-dimensional anomalies derived from a multiyear hindcast simulation (Lellouche et al. 2013). SAM2 assimilates observations of sea level anomaly (SLA) and SST, as well as in situ temperature and salinity profiles. To provide consistency with the atmospheric model, the same SST analysis (described above) as used in the 4DEnVar is assimilated in SAM2, with no additional in situ SST observations. A detailed description of the SAM2 system design and performance is provided in Lellouche et al. (2013).
An important difference in GIOPS from the system described in Lellouche et al. (2013) is the addition of a blending procedure of the SAM2 ocean analysis with the 3DVar total ice concentration analysis. The rescale forecast tendencies method described in Smith et al. (2016a) is used to project the total ice concentration increments across the 10 ice thickness categories used in CICE. This approach permits a detailed constraint on the ice cover, consistent with that used by the 4DEnVar, while avoiding spurious effects on the ice thickness field.
The system is implemented such that two successive assimilation cycles are produced every Wednesday with 7-day assimilation windows (i.e., going back 14 days), together with a daily update cycle with a 1-day assimilation window. The delayed mode cycle is run from time T − 14 days to T − 7 days (where T is the analysis date) in order to include SLA and in situ observations that may only be available after several days’ delay. This cycle provides the continuity in time. To bring the cycle up to real time, a second analysis cycle is produced from time T − 7 days to T. The ice–ocean forecasts in GIOPS are then initialized from the daily update cycle, whereby only SST is assimilated in SAM2 prior to the blending with the 3DVar ice analysis. Additional details on the GIOPSv2.1 system used here are available in Smith et al. (2016b).
An evaluation of forecast skill of the previous version of GIOPS (v1.1) is provided in Smith et al. (2016a), Ryan et al. (2015), and Divakaran et al. (2015). These studies demonstrate a consistent picture of skillful medium-range ice and SST forecasts in both the Northern and Southern Hemispheres, as compared to persistence. Moreover, Ryan et al. (2015) show that as compared to surface drifters, GIOPS SST forecasts have the second-lowest root-mean-square error (RMSE) of all participating operational global ocean forecasting systems.
The assimilation approach used in GIOPS allows for both a robust constraint on the subsurface water masses through use of the two 7-day analysis cycles, while also permitting a close correspondence between the surface conditions in GIOPS analyses and those specified in the 4DEnVar analysis cycles. While this approach will nonetheless result in some inconsistencies in surface fields, it allows for an evaluation of the impact of coupling in an operational medium-range NWP framework without any modification to the atmospheric assimilation methodology [e.g., through use of a weakly coupled assimilation method as described by Lea et al. (2015)].
The main inconsistencies in the initial conditions used for the coupled forecasts are associated with the sea ice cover (ice surface temperature and lead fraction). An earlier version of this system found a large impact of the lead fraction on surface biases in the Arctic (Smith et al. 2013), resulting in an overall cold bias in the coupled model. This issue was caused by a lack of leads in the ice analysis. By allowing leads in the pack ice to remain in the GIOPS analyses [see Smith et al. (2016a) for details], this bias is avoided. The remaining inconsistencies in sea ice conditions due to ice thickness, snow cover, and the calculation of ice surface temperature are not found to have a strong impact on forecast skill (not shown), although they may still have important impacts locally. For example, differences in ice thickness between GIOPS analyses and the climatology used in the 4DEnVar can lead to surface temperature differences of several degrees Celsius. This surface temperature shock dissipates quickly (in less than a day) as the atmospheric boundary layer and snow/ice surface temperature adjust to a common equilibrium.
d. Forecast experiment design
To evaluate the impact of coupling on forecast skill, a series of forecasts were produced for both a summer and winter period. This approach follows the standard practice at CCMEP for delivering updates to the GDPS. Forecasts are produced for two reference periods: 15 June to 31 August 2014 (referred to hereafter as JulAug) and 15 December 2014 to 1 March 2015 (referred to hereafter as JanFeb). The forecasts are initialized at 0000 UTC and run for 240 h. For the JulAug and JanFeb periods, a total of 78 and 76 forecasts were produced for each period, respectively. The atmospheric and ice–ocean models are initialized from their respective (uncoupled) analyses (i.e., GDPSv5.0.0 and GIOPSv2.1.0). Herein, the two sets of summer and winter forecasts will be referred to as the “coupled” and “forced” forecasts.
e. Verification metrics
Following standard practice for verification of NWP systems, the main metrics used here are the mean and standard deviation differences between forecasts (either coupled or forced) and some measure of “truth” (either gridded analysis fields or observations). These metrics are assessed both spatially and temporally, as well as in terms of forecast lead time. When verifying against analyses, it is important to consider the error (or uncertainty) associated with the analysis itself. As a result, verification against analyses is restricted mainly to larger lead times, when forecast error will dominate over analysis uncertainty. Results presented here were compared to two different sets of analyses, GDPSv5.0.0 and ERA-Interim (Dee et al. 2011), and the conclusions are not sensitive to the choice of analysis. As a result, only verification scores compared to ERA-Interim will be presented, as this is a more commonly used reference and permits comparison with previously published studies (e.g., Buehner et al. 2015).
3. SST forecast evaluation
We start by assessing SST fields, as these have a direct impact on surface fluxes and thus, near-surface atmospheric fields. A first-order impact of introducing time-varying SSTs, as compared to persisted SSTs (i.e., as in the forced forecasts), is the inclusion of seasonal variations in SST. Vitart et al. (2008) showed that use of persisted SST anomalies, rather than persisted SST fields, resulted in positive impacts in probabilistic scores beyond day 10 in the ECMWF monthly forecasting system. An additional difference here is that the coupled model includes both rapid (e.g., diurnal) variations and the longer time scale evolution of SST anomalies over 10-day forecasts. Figure 2 shows bias and standard deviation errors [Eqs. (1) and (2)], as compared to the OI SST analyses (described in section 2c), over the JulAug period. Differences for the JanFeb period are roughly equivalent for summer/winter hemispheres as those for JulAug, but with a smaller intensity (not shown). As the OI SST analyses are assimilated by the SAM2 ocean analysis system, differences between the SSTs used to initialize the coupled and forced forecasts are quite small (Figs. 2a,b). Bias and standard deviation errors in SST are generally below 0.6°C, with some larger differences in isolated regions (e.g., areas of strong eddy activity). Note that the OI SST analyses have a global RMSE of about 0.5°C [as compared to in situ observations; Brasnett and Surcel Colan (2016)], and thus the differences seen in Figs. 2a and 2b fall mostly within analysis error. The largest differences are present in coastal areas and in the marginal ice zone, where values of up to 2.5°C can be found. The GIOPS analyses are constrained less strongly to the OI SST analyses in these areas, as larger errors in the OI SST analyses in these locations have been previously identified (Brasnett and Surcel Colan 2016). As such, the bias and standard deviation errors shown in Figs. 2a and 2b should be considered more as “differences” than as actual errors.
SST forecast verification for JulAug, as compared to OI SST analyses. (a) Bias and (b) standard deviation error (STD) in the initial condition between coupled (CPL) and forced (FRC) forecasts. Positive values indicate warmer SSTs in coupled initial conditions in (a). (left) Bias and (right) standard deviation error at 240-h lead time for the (c),(d) forced and (e),(f) coupled forecasts. Global mean values are shown in the bottom left-hand corner of each panel.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
In summer (winter), SSTs will tend to warm (cool) over the 10-day forecast period due to their seasonal evolution. As a result, forced forecasts show negative (positive) biases at 240-h lead time in the summer (winter) hemisphere due to the persistence of SST analyses (Fig. 2c). On the other hand, coupled forecasts are able to capture the seasonal evolution of SSTs, but they develop biases (Fig. 2e), due possibly to errors associated with the surface radiative balance or surface evaporation (Hourdin et al. 2015).
While coupled and forced forecasts show SST biases of similar amplitude, the standard deviation error is smaller for coupled forecasts (Figs. 2d,f). Notably smaller values for coupled forecasts are present over much of the northern Pacific and Atlantic Oceans, in particular over the western half of each basin. Larger standard deviation errors in SST for forced forecasts are also found in the marginal ice zone and in the eastern equatorial Pacific and Atlantic Oceans. Larger standard deviation error in the coupled forecasts can be seen in isolated areas of strong eddy activity (e.g., Brazil–Malvinas confluence region, Agulhas Current) due to the presence of unfiltered mesoscale structures in the model fields.
These results highlight the ability of the ocean model to provide skill in evolving the SST, as compared to persistence of the initial conditions, as is used in forced (operational) forecasts of the GDPS. An additional effect, not well captured in the comparison with OI SST analyses presented in Fig. 2, is the extent to which the forecasts adequately represent the formation of cold wakes due to the passage of tropical cyclones. Individual case study evaluations using surface drifters suggest that the coupled model is able to form well these cold wakes [consistent with previous studies using ocean models of a similar complexity noted above; e.g., Chen et al. (2010)], while they are almost entirely absent from the OI SST analyses [and thus, forced forecasts; Brasnett and Surcel Colan (2016)]. An example is provided in the following section.
4. Impact of coupling on forecasts of Typhoon Neoguri
The two-way interaction with the ice–ocean model is expected to have a range of impacts on weather forecasts, as described in section 1. The most significant of these is the strong air–sea interaction associated with intense cyclones. As an illustration, a short case study of category-5 Supertyphoon Neoguri is provided below.
Typhoon Neoguri was a powerful typhoon that struck Japan in 2014. It originated in the western Pacific Ocean and developed into a tropical storm on 3 July 2014. Neoguri reached peak intensity as a category-5 supertyphoon on 7 July 2014 and made landfall on 9 July 2014. The minimum central pressure for Neoguri over the period 4–10 July 2014, as represented in both the forced and coupled forecasts, is shown in Fig. 3. For each day and model version, the minimum central pressure is shown for lead times of 24, 48, 72, 96, and 120 h to illustrate the consistency of forecasts in time. The forced forecasts demonstrate a significant increase in the peak minimum central pressure as a function of lead time, whereas the coupled forecasts produce very similar values for different lead times. These results are consistent with numerous studies that show that interactive coupling tends to reduce the intensity of cyclones and may correct a tendency for higher-resolution models to produce overly intense storms.
Time series of minimum central pressure for Typhoon Neoguri over the period 4–10 Jul 2014, as represented in both the forced (cool colors) and coupled (warm colors) forecasts. For each day and model version, the minimum central pressure is shown for lead times of 24, 48, 72, 96, and 120 h. The GDPS analysis is shown as a thick black line.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
The stronger intensification of Typhoon Neoguri in forced forecasts is related to the availability of heat from the ocean. As noted in section 1, an important feature of tropical cyclones is the development of a cold wake in the right-rear quadrant, associated mainly with entrainment of cooler subsurface waters through ocean mixing and Ekman pumping in response to strong winds. The resulting reduction in SST leads to a reduction in sensible and latent heat fluxes, thereby limiting the intensification of the cyclone. This process is illustrated for Typhoon Neoguri in Fig. 4, which shows the differences in SST and latent heat fluxes for forced and coupled forecasts. The presence of an important cold wake in the coupled forecasts can be seen in Fig. 4, with SSTs roughly 4°C cooler than the SST analysis used by the forced forecasts. This cooling is consistent with observations from a nearby surface drifter that measured an SST of 26°C—only slightly cooler than the coupled forecasts. This cooling resulted in a significant reduction of latent heat fluxes with differences greater than 500 W m−2. As a result, the coupled forecasts provide a reduced intensification of Typhoon Neoguri with storm intensity forecasts that are more consistent in time.
The 96-h SST forecasts initialized at 0000 UTC 6 Jul 2014 for Typhoon Neoguri from the (a) forced (FRC) and (b) coupled (CPL) systems. Contemporary SST drifter observations are shown as black-outlined circles. (c) The CPL–FRC differences in SST, with CPL-observed differences from drifter observations shown as black-outlined circles. (d) CPL − FRC differences in latent heat fluxes for 48-h forecasts. The observed track and maximum sustained wind speed for Typhoon Neoguri are shown as black-outlined circles in (d).
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
5. Impact of coupling on TC statistics
In this section, we assess the impact of coupling on the development and evolution of cyclones using a TC detection and tracking algorithm. The algorithm used here was originally developed by Caron et al. (2011) and is used following the implementation strategy for the GDPS described by Zadra et al. (2014). While the period of evaluation employed here is relatively short, it nonetheless provides an indication of the overall impact of coupling on TC statistics.
The automated TC detection and tracking algorithm produces a time series of TC positions and properties (e.g., central pressure and maximum wind speed) from a set of model forecasts, which can then be compared to observed values. TC detection is based on the following criteria: minimum in SLP below 1012 hPa, cyclonic relative vorticity greater than 10−5 s−1 at 850 hPa, maximum wind speed above 22 kt (1 kt = 0.5144 m s−1), and 850-hPa wind speed or vorticity larger than 250-hPa wind speed or vorticity. For tracking, a maximum displacement speed of 36 kt is used to distinguish separate TCs. A minimum separation distance of 400 km between TCs is also applied. These criteria follow those defined in Zadra et al. (2014), where additional details regarding the criteria and the implementation of the algorithm can be found.
Following the application of the detection and tracking algorithm, the forecast TCs are paired with observed TCs, and values are tabulated for the number of hits (paired TCs), misses (nonpaired TCs), and false alarms (nonpaired TCs in forecasts). Standard derived scores (bias, false alarm rate, etc.) are then generated from the tabulated counts. Observed track data for the Atlantic and eastern Pacific Oceans are obtained from U.S. National Hurricane Center advisories, whereas for the west Pacific Ocean, advisories from the Joint Typhoon Warning Center are used. Only cyclones that reached tropical cyclone intensity (35 kt) are considered here.
TC activity for the JulAug period considered here was fairly average for the Atlantic and western Pacific basins, with abnormally high activity in the eastern Pacific Ocean. For the Atlantic Ocean, there were two category-1 hurricanes and one category-2 hurricane, according to the Saffir–Simpson hurricane wind scale. The western Pacific Ocean was notable over the JulAug period for the number of large storms, as there were four typhoons, including three category-5 supertyphoons. The eastern Pacific Ocean was particularly active during this period, with six tropical storms and a number of hurricanes (three category 1, two category 3, one category 4, and one category 5).
TC statistics for coupled and forced forecasts over the JulAug period are shown in Fig. 5. Both sets of forecasts show a similar response, with a growing number of false alarms as a function of lead time until about day 9. Zadra et al. (2014) show, with an earlier version of the GDPS, that the increase in false alarms can be partially addressed by lowering the value of the trigger velocity used in the Kain and Fritsch deep convection scheme over the tropical oceans (from 0.05 to 0.01 m s−1). In particular, they obtain an 18% reduction in the false alarm ratio at day 5 over the northern extratropics. The impact of coupling with the ocean found here has a similar response over the west Pacific region, with a 21% reduction in the false alarm ratio at day 5. Smaller, but nonetheless significant, reductions in the false alarm ratio of about 10% over the east Pacific and Atlantic regions are also found. The lower values for false alarm ratio are due to a systematic reduction in the intensity of cyclones in coupled forecasts, consistent with previous studies (e.g., Ito et al. 2015). As such, coupling with the ocean not only affects the intensity of severe hurricanes and typhoons (as shown in previous studies), but also results in fewer TCs overall. Moreover, this reduction occurs with no significant impact on either the TC hit (detection) rate or track errors (not shown). However, the lack of significance for the hit rate and track errors may be affected by the limited number of cases assessed here.
TC false alarm ratio for the JulAug period for the (a) west Pacific, (b) east Pacific, and (c) Atlantic basins. Results for the forced model are shown in blue, and those for the coupled model are in red. Shaded areas indicate the 95% confidence intervals. The bottom row shows differences between the coupled and forced false alarm ratios.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
6. Verification of forecasts compared to ERA-Interim
a. Spatial differences compared to ERA-Interim
We will now assess the global atmospheric impacts of coupling, as compared to independent analyses. First, we examine the spatial representation of differences in near-surface fields. Figures 6 and 7 show the impact of coupling [Eq. (3)] on standard deviation error of 120-h forecasts for the JulAug and JanFeb periods, as compared to ERA-Interim (Dee et al. 2011). The impact of coupling is computed as the forced-minus-coupled differences in standard deviation error normalized by the standard deviation error of the forced forecasts, providing fields in units of percentage change in error. Areas with smaller standard deviation error in coupled (forced) forecasts are shown in warm (cool) colors. An arbitrary value of 5% is used as a lower threshold to remove spurious values associated with small standard deviation errors in the denominator.
Percent difference in standard deviation errors of 120-h coupled and forced forecasts, as compared to ERA-Interim for (left) JulAug and (right) JanFeb for (a),(b) 850-hPa geopotential height, (c),(d) wind speed, (e),(f) specific humidity, and (g),(h) temperature fields. Warm (cool) colors represent smaller standard deviation errors of coupled (forced) forecasts, as compared to ERA-Interim.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
As in Fig. 6, but for 100-hPa geopotential height.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
The 850-hPa geopotential height fields show significant improvements in coupled forecasts, with decreases in standard deviation error in excess of 15% over large areas of the western Pacific Ocean in JulAug (Fig. 6a), as well as for the Indian and southwest Pacific Oceans in JanFeb (Fig. 6b). Improvements of 10% and even up to 30% can also be seen over smaller areas of the eastern Pacific and North Atlantic Oceans in JulAug and the South Pacific Ocean in JanFeb. The improvements in geopotential height fields are related mainly to the decrease in cyclone intensity discussed in the previous section and consistent with previous studies assessing the impact of coupling on tropical cyclones (e.g., Ito et al. 2015). As a result, the reduced cyclone intensity produces both a mean increase in geopotential height fields in the western Pacific Ocean in JulAug (Fig. 8a) and south Indian Ocean in JanFeb (not shown), as well as the reduced errors in standard deviation seen in Fig. 6. The reduced intensity of major cyclones in coupled forecasts also produces differences in standard deviation error of winds in the areas of strong cyclone activity (west Pacific and south Indian Oceans in JulAug and JanFeb, respectively; Figs. 6c,d). A positive impact on humidity in the western Pacific Ocean in JulAug can also be noted, with a smaller effect in JanFeb in the Indian Ocean. Only a small effect on temperature is found.
Coupled − forced differences in (a) mean 850-hPa geopotential height and (b) 200-hPa specific humidity fields for the JulAug period.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
Interactive coupling with the ocean has been shown to result in a reduction in storm intensity due mainly to the effect of the surface cooling on surface fluxes (primarily by reducing latent heat fluxes). Based on results from case studies, Chen et al. (2010) suggest that this change in surface fluxes may even affect upper-tropospheric fields. We now assess this here in terms of the impact of coupling [Eq. (3)] on standard deviation errors for 120-h forecasts at 100 hPa (Fig. 7). Consistent with Chen et al. (2010), Fig. 7 shows that the reduction in cyclone intensity does indeed have an important impact on the upper troposphere. Geopotential height, wind, and temperature fields all show significant reductions in standard deviation with improvements of greater than 15% in geopotential height over a large area of the western Pacific Ocean. Indeed, mean differences in coupled-minus-forced specific humidity at 200 hPa (Fig. 8b) show a net reduction centered in the western Pacific Ocean consistent with reduced vertical fluxes of humidity to the upper troposphere.
While the reduction in standard deviation in coupled forecasts in the lower troposphere is localized around the storm position, at upper levels, the impact covers a much broader area. This suggests that the reduced vertical flux of heat and moisture in active cyclone areas in the coupled forecasts is having an important nonlocal impact over an extended region.
b. Temporal variability compared to ERA-Interim
To illustrate the temporal variability of coupling impacts, Hovmöller plots of forced-minus-coupled differences in 120-h standard deviation error [Eq. (2)] from ERA-Interim for upper- (100 hPa) and lower- (850 hPa) tropospheric geopotential heights for JulAug and JanFeb are shown in Fig. 9 for the northern extratropics (20°–70°N) and tropics (20°S–20°N), respectively. Clearly, the most important differences in geopotential heights are associated with a number of events in areas of dominant cyclone activity. In particular, reductions in differences in geopotential heights in JulAug of greater than 20 m at lower levels (Fig. 9c) and 10 m in the upper troposphere (Fig. 9a) are found. These changes are associated mainly with two supertyphoons, Neoguri and Halong, active over the periods 3–10 July and 28 July to 10 August, respectively. Similarly, in JanFeb, the impact of TC Bansi in the Indian Ocean (10–18 January) can be seen in Figs. 9b and 9d, as well as the impact of coupling on several smaller cyclone events.
Hovmöller plots showing forced − coupled differences in standard deviation error of 120-h geopotential height forecasts, as compared to ERA-Interim at (a),(b) 100 and (c),(d) 850 hPa. Differences for JulAug are shown over the (a),(c) northern extratropics and for (b),(d) JanFeb over the tropics in units of dam. (e),(f) Maps showing the northern extratropics (20°–70°N) and tropics (20°S–20°N) regions used. Warm (cool) colors represent smaller standard deviation errors of coupled (forced) forecasts, as compared to ERA-Interim.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
During periods without active cyclone activity, coupled and forced forecasts provide similar standard deviation errors (Figs. 10, 11). As such, the impact of coupling appears to occur intermittently based on the presence of active cyclones. This intermittent impact is sufficiently important, however, in that it results in a net reduction in standard deviation error over the JulAug and JanFeb periods (e.g., see trend lines in Figs. 10, 11).
Time series of coupled (red) and forced (blue) standard deviation errors, as compared to ERA-Interim of 850-hPa geopotential height forecasts for JulAug at lead times of (a) 24, (b) 72, and (c) 120 h. Values are calculated over the northern extratropical region shown in Fig. 6e and shown in units of dam. Linear trend lines are shown for coupled and forced forecasts as red and blue dashed lines, respectively, to illustrate differences over the full period.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
As in Fig. 10, but for JanFeb. Values are calculated over the tropics region shown in Fig. 6f.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
It is also interesting to note that the impact of coupling is felt even at quite short lead times. While much larger differences can be seen at 120 h (Figs. 10c, 11c), differences are present nonetheless at 24-h lead times during periods of peak cyclone activity. This fast atmospheric response is due mainly to the rapid cooling of the sea surface in response to the strong winds associated with passing cyclones. Diurnal variations in SST and small differences between the initial SST of coupled and forced forecasts also make a small contribution to this effect.
As discussed in the previous section, the reduction in cyclone intensity due to coupling also leads to a reduction in moisture fluxes into the upper troposphere (Fig. 8). Interestingly, this occurs most strongly downstream from main cyclone activity, with a maximum over the North Pacific Ocean (Fig. 12). The reduction in vertical moisture fluxes, and their impact aloft, accumulates over the 10-day forecast period such that differences between standard deviation errors of coupled and forced forecasts grow significantly with increasing lead time. In particular, a systematic reduction in large standard deviation error events (forecast “busts”) occurs, providing a more temporally consistent level of standard deviation error in coupled forecasts.
As in Fig. 10, but for 200-hPa specific humidity forecasts at lead times of (a) 24, (b) 72, (c) 120, (d) 144, (e) 168, and (f) 240 h. Values are calculated over the North Pacific Ocean and are shown in units of g kg−1.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
The impact of coupling can also be assessed in terms of the effect on forced-minus-coupled differences in standard deviation errors [Eq. (2)] in geopotential height fields as a function of pressure level and lead time (Fig. 13). As noted previously, the primary impact of coupling is felt near the surface, but with reduced standard deviation error growth aloft as well. The overall result is a reduction in standard deviation error of geopotential height of about 0.2 dam at the surface and at 100 hPa in JulAug over the northern extratropics. For JanFeb, the impact is somewhat smaller, with a reduction in standard deviation error of geopotential height, as compared to ERA-Interim, of about 0.1 dam near-surface and 0.2 dam centered at 100 hPa. Improvements in standard deviation errors at 500 hPa also occur, but with a greater delay (after about 120 h) and of smaller magnitude (roughly 0.05 dam).
Forced − coupled differences in standard deviation error of geopotential height forecasts as a function of pressure level and lead time, as compared to ERA-Interim for (a) JulAug over the northern extratropics and (b) JanFeb over the tropics. Warm (cool) colors represent smaller standard deviation errors of coupled (forced) forecasts as compared to ERA-Interim. Regions are shown in Figs. 6e and 6f. The confidence level for each lead time and pressure level (in percent) is also indicated.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
The error growth rate can be assessed by examining the increase in standard deviation error of geopotential heights over successive 24-h periods [Eq. (4)]. Figure 14 shows the standard deviation error growth rate for four pressure levels (850, 500, 250, and 100 hPa) for coupled and forced forecasts. Over the northern extratropics in JulAug, both sets of forecasts show a maximum in error growth occurring between days 6 and 8, with a decline thereafter as errors associated with midlatitude synoptic weather patterns begin to saturate. In the tropics in JanFeb, a different behavior is found, with a fairly constant level of error growth over days 2–10.
Impact of coupling on error growth in geopotential height, as compared to ERA-Interim for (a),(c) JulAug over the northern extratropics and (b),(d) JanFeb over the tropics. Regions are shown in Figs. 6e and 6f. Standard deviation error growth is estimated as the daily rate of change of standard deviation errors between coupled–forced forecasts and ERA-Interim (a),(b). (c),(d) The difference in standard deviation error growth in coupled (red) and forced (blue) forecasts is expressed as a % change. Positive values represent a smaller increase in standard deviation error growth in coupled forecasts.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
The impact of coupling [Eq. (5)] can be estimated as the forced-minus-coupled differences in standard deviation error growth rate expressed as the percentage change, as compared to the standard deviation error growth rate for forced forecasts (Figs. 14c,d). When assessed in this manner, a similar response to coupling is found for both the northern extratropics in JulAug and the tropics in JanFeb, with a relatively constant decrease in standard deviation error growth for days 1–7 (i.e., positive values in Figs. 14c,d). This decrease corresponds to an improvement at 850 hPa of about 5% over the northern extratropics in JulAug and 15% over the tropics in JanFeb. While coupling likely has an impact on standard deviation error growth beyond day 8, the differences in standard deviation error growth rate for days 9 and 10, seen here in a deterministic context, are obscured due to chaotic variability.
7. Verification of forecasts against radiosonde observations
We now present an evaluation against global radiosonde observations to provide a reference for the impact of coupling, as compared to previous forecast model improvements (Fig. 15). The statistical significance for the bias and standard deviation errors are shown as colored boxes on the left and right sides, respectively. Significance tests are computed using the Student’s t test for mean and the F test for standard deviation. Statistically significant improvements in the coupled (forced) model are shown in red (blue).
Verification of global 120-h coupled (red) and forced (blue) forecasts, as compared to radiosonde observations for (top) JulAug and (bottom) JanFeb. The bias (dashed) and standard deviation (solid) scores are shown for (a),(d) temperature, (b),(e) zonal wind, and (c),(f) geopotential height. The confidence levels for (left) bias and (right) standard deviation are shown as colored boxes, with the color indicating which experiment has the better score.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
In the previous section, we showed how the primary impacts of coupling are localized at the surface in areas of cyclone activity, with broader impacts spreading downstream at higher levels. Here, the global impacts of coupling are assessed against observations. Given that the radiosonde network is distributed predominantly over land, the verification against radiosondes is expected to emphasize the nonlocal impacts of coupling. As the direct effects of coupling are felt over the oceans, this verification is expected to underestimate the overall impact. Moreover, it will neglect impacts of coupling over less-well-observed areas, such as the polar regions.
Globally, the impact of coupling is a statistically significant reduction in standard deviation error, as compared to radiosondes observations for temperature, zonal winds, and geopotential heights for both the JulAug and JanFeb evaluation periods. Dewpoint depression also shows statistically significant reductions in standard deviation errors (not shown). As found for the verification against analyses (section 6), the largest improvements, as verified against radiosondes in standard deviation error, occur near the surface and in the upper troposphere (indicated by the presence of the confidence-level boxes on the right in red). Both positive and negative changes are found for biases. The change in bias is associated with a systematic impact on surface fluxes due to coupling (e.g., reduced latent heat fluxes over TCs). As a result, a small shift in the mean state occurs in certain regions, such as over the tropics. Moreover, the impact on biases is quite sensitive to the particular choice of model physics. Here, a comparison is made with the operational version of the model with no additional adjustments to take into account the effects of coupling. As such, some degradation in biases of temperature and geopotential height is found.
8. Comparison of precipitation forecasts
Finally, here, we compare coupled and forced forecasts with satellite estimates of precipitation (Fig. 16) for JulAug. For this, daily mean estimates from version 7 of the Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis are used for comparison with day 5 (from 96 to 120 h) precipitation accumulation forecasts averaged over the whole JulAug period. A systematic decrease in precipitation in coupled forecasts is found in areas of significant cyclone activity, with the largest impact seen in the western Pacific Ocean between 10° and 40°N. The smaller precipitation values are consistent with the reduced intensification of cyclones noted above and improve the agreement of coupled forecasts with estimates from TRMM (Figs. 16b,c). A slight increase in precipitation can also be noted between 0° and 10°N in the central and eastern Pacific Ocean. This increase may be associated with the inclusion of diurnal SST variations in coupled forecasts or a slight increase in the Hadley circulation intensity.
Spatial distribution of accumulated daily precipitation for the JulAug period (mm). (a) TRMM satellite estimates, with differences for 96–120 h for (b) coupled and (c) forced forecasts. (d) Corresponding coupled − forced differences in accumulated precipitation forecasts over 96–120-h lead times. Warm (cool) colors in (d) indicate increased (decreased) precipitation in coupled forecasts.
Citation: Monthly Weather Review 146, 4; 10.1175/MWR-D-17-0157.1
9. Conclusions and discussion
Here, we provide an evaluation of the impact of an interactive air–sea coupling between an operational global deterministic medium-range forecasting system and an ice–ocean forecasting system. This coupled system was developed in the context of an experimental forecasting system that has been running in operations at the CCMEP since July 2016 and became fully operational on 1 November 2017. To the authors’ knowledge, this is the first time coupling with an ice–ocean model has been shown to produce statistically significant global improvements to an operational medium-range NWP system.
An evaluation against both independent analyses and radiosonde observations shows important reductions in standard deviation errors in geopotential height, wind, temperature, and humidity fields. We show that the most significant impact is found to be associated with a reduced intensification of cyclones, with a 21% reduction in the false alarm ratio in the western Pacific Ocean. The improved representation of cyclones leads to smaller standard deviation errors in geopotential height fields of more than 20 m at the surface and 10 m in the mid-to-upper troposphere for 120-h forecasts, with commensurate benefits for wind, temperature, and humidity fields. Whereas impacts on surface fields are found locally in the vicinity of cyclone activity, large-scale improvements in the mid-to-upper troposphere are found with positive global implications for forecast skill. Moreover, coupling is found to produce fairly constant reductions in error growth for forecast days 1–7 of about 5% over the northern extratropics in JulAug and 15% over the tropics in JanFeb.
Here, we focus on evaluating the impact of coupling using the traditional summer and winter periods used for evaluating system updates at CCMEP, although larger impacts may actually be found in spring and fall. A significant impact of air–sea interaction found here is related to the formation of cold wakes in tropical cyclones. As the annual peak SSTs and significant cyclone activity occurs during fall, larger impacts may be expected. Indeed, preliminary results from the experimental coupled system running during the fall of 2016 showed significantly reduced forecast errors, as compared to the operational (forced) system. Moreover, the particular periods of study used here have below-average cyclone frequency overall, and thus, greater impacts could be expected for other years.
The coupling approach employed here was to develop and evaluate the coupling of a global NWP system with an ice–ocean model with the minimum impact to the operational atmospheric forecasting suite. As such, the only modification to the atmospheric configuration was the exchange of fluxes with the ice–ocean model with the initialization and atmospheric model physics and dynamics left unchanged. However, the strong impact of coupling on atmospheric fields shown here suggests an additional potential benefit may be possible by recalibrating certain parameter choices and model physics to exploit the coupled interactions. For instance, the trigger velocity used in the deep convection parameterization could be reassessed in the manner of Zadra et al. (2014), given the impact of coupling on cyclone intensity found here. Moreover, such a retuning exercise may resolve the increase in certain biases noted, as compared to radiosonde observations. In any case, a dedicated treatment of model biases in a coupled context is required in order to extend the potential benefits of coupling through coupled data assimilation (e.g., Lea et al. 2015).
The impact of coupling found here is due predominantly to the improved representation of air–sea interactions in tropical cyclones. Indeed, the impact associated with cyclones is so large, it eclipses to some extent the role of other processes, such as upwelling and interactions with sea ice. This may be due in part to the use of summer and winter evaluation periods, as sea ice and many other marine effects would be expected to have maximum impact during spring and fall. Future work is needed to isolate these coupled processes and better understand how they can be used to benefit medium-range NWP
Acknowledgments
This work was made possible by the support of staff at the Canadian Centre for Meteorological and Environmental Prediction, including people within the research, development, and operations teams, as well as by support from upper management. In particular, we would like to acknowledge the contributions of E. Lapalme for his assistance with the Maestro suite and Y. Chartier for his assistance in running the forecasts. The authors would also like to acknowledge support from the Canadian Operational Network of Coupled Environmental Prediction Systems (CONCEPTS) and, in particular, F. Davidson for his tireless efforts to advance CONCEPTS objectives. We would also like to acknowledge ongoing support from Mercator Océan in developing the ocean forecasting component and, in particular, G. Garric and C. E. Testut. The authors also thank the three reviewers for their helpful comments.
REFERENCES
Adcroft, A., C. Hill, and J. Marshall, 1997: Representation of topography by shaved cells in a height coordinate ocean model. Mon. Wea. Rev., 125, 2293–2315, https://doi.org/10.1175/1520-0493(1997)125<2293:ROTBSC>2.0.CO;2.
Barnier, B., and Coauthors, 2007: Eddy-permitting ocean circulation hindcasts of past decades. CLIVAR Exchanges, No. 12, International CLIVAR Project Office, Southampton, United Kingdom, 8–10.
Bauer, P., A. Thorpe, and G. Brunet, 2015: The quiet revolution of numerical weather prediction. Nature, 525, 47–55, https://doi.org/10.1038/nature14956.
Bélair, S., J. Mailhot, J. W. Strapp, and J. I. MacPherson, 1999: An examination of local versus nonlocal aspects of a TKE-based boundary layer scheme in clear convective conditions. J. Appl. Meteor., 38, 1499–1518, https://doi.org/10.1175/1520-0450(1999)038<1499:AEOLVN>2.0.CO;2.
Bélair, S., L.-P. Crevier, J. Mailhot, B. Bilodeau, and Y. Delage, 2003a: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part I: Warm season results. J. Hydrometeor., 4, 352–370, https://doi.org/10.1175/1525-7541(2003)4<352:OIOTIL>2.0.CO;2.
Bélair, S., R. Brown, J. Mailhot, B. Bilodeau, and L.-P. Crevier, 2003b: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part II: Cold season results. J. Hydrometeor., 4, 371–386, https://doi.org/10.1175/1525-7541(2003)4<371:OIOTIL>2.0.CO;2.
Bélair, S., J. Mailhot, C. Girard, and P. Vaillancourt, 2005: Boundary layer and shallow cumulus clouds in a medium-range forecast of a large-scale weather system. Mon. Wea. Rev., 133, 1938–1960, https://doi.org/10.1175/MWR2958.1.
Belcher, S. E., H. T. Hewitt, A. Beljaars, E. Brun, B. Fox-Kemper, J.-F. Lemieux, G. Smith, and S. Valcke, 2015: Ocean–waves–sea ice–atmosphere interactions. Seamless Prediction of the Earth System: From Minutes to Months, G. Brunet, S. Jones, and P. Ruti, Eds., World Meteorological Organization, 155–169.
Bender, M. A., and I. Ginis, 2000: Real-case simulations of hurricane–ocean interaction using a high-resolution coupled model: Effects on hurricane intensity. Mon. Wea. Rev., 128, 917–946, https://doi.org/10.1175/1520-0493(2000)128<0917:RCSOHO>2.0.CO;2.
Benoit, R., J. Côté, and J. Mailhot, 1989: Inclusion of a TKE boundary layer parameterization in the Canadian regional finite-element model. Mon. Wea. Rev., 117, 1726–1750, https://doi.org/10.1175/1520-0493(1989)117<1726:IOATBL>2.0.CO;2.
Bernie, D., E. Guilyardi, G. Madec, J. Slingo, S. Woolnough, and J. Cole, 2008: Impact of resolving the diurnal cycle in an ocean–atmosphere GCM. Part 2: A diurnally coupled CGCM. Climate Dyn., 31, 909–925, https://doi.org/10.1007/s00382-008-0429-z.
Bougeault, P., and P. Lacarrère, 1989: Parameterization of orography-induced turbulence in a mesobeta-scale model. Mon. Wea. Rev., 117, 1872–1890, https://doi.org/10.1175/1520-0493(1989)117<1872:POOITI>2.0.CO;2.
Bouillon, S., M. A. Morales Maqueda, V. Legat, and T. Fichefet, 2009: An elastic–viscous–plastic sea ice model formulated on Arakawa B and C grids. Ocean Modell., 27, 174–184, https://doi.org/10.1016/j.ocemod.2009.01.004.
Brasnett, B., 2008: The impact of satellite retrievals in a global sea‐surface‐temperature analysis. Quart. J. Roy. Meteor. Soc., 134, 1745–1760, https://doi.org/10.1002/qj.319.
Brasnett, B., and D. Surcel Colan, 2016: Assimilating retrievals of sea surface temperature from VIIRS and AMSR2. J. Atmos. Oceanic Technol., 33, 361–375, https://doi.org/10.1175/JTECH-D-15-0093.1.
Brassington, G. B., and Coauthors, 2015: Progress and challenges in short- to medium-range coupled prediction. J. Oper. Oceanogr., 8, s239–s258, https://doi.org/10.1080/1755876X.2015.1049875.
Buehner, M., J. Morneau, and C. Charette, 2013: Four-dimensional ensemble–variational data assimilation for global deterministic weather prediction. Nonlinear Processes Geophys., 20, 669–682, https://doi.org/10.5194/npg-20-669-2013.
Buehner, M., and Coauthors, 2015: Implementation of deterministic weather forecasting systems based on ensemble–variational data assimilation at Environment Canada. Part I: The global system. Mon. Wea. Rev., 143, 2532–2559, https://doi.org/10.1175/MWR-D-14-00354.1.
Buehner, M., A. Caya, T. Carrieres, and L. Pogson, 2016: Assimilation of SSMIS and ASCAT data and the replacement of highly uncertain estimates in the Environment Canada Regional Ice Prediction System. Quart. J. Roy. Meteor. Soc., 142, 562–573, https://doi.org/10.1002/qj.2408.
Caron, L.-P., C. G. Jones, and K. Winger, 2011: Impact of resolution and downscaling technique in simulating recent Atlantic tropical cyclone activity. Climate Dyn., 37, 869–892, https://doi.org/10.1007/s00382-010-0846-7.
Charron, M., and Coauthors, 2012: The stratospheric extension of the Canadian global deterministic medium-range weather forecasting system and its impact on tropospheric forecasts. Mon. Wea. Rev., 140, 1924–1944, https://doi.org/10.1175/MWR-D-11-00097.1.
Chelton, D. B., and S.-P. Xie, 2010: Coupled ocean–atmosphere interaction at oceanic mesoscales. Oceanography, 23, 52–69, https://doi.org/10.5670/oceanog.2010.05.
Chen, S., T. J. Campbell, H. Jin, S. Gaberšek, R. M. Hodur, and P. Martin, 2010: Effect of two-way air–sea coupling in high and low wind speed regimes. Mon. Wea. Rev., 138, 3579–3602, https://doi.org/10.1175/2009MWR3119.1.
CMC, 2015: Global Deterministic Prediction System (GDPS): Update from version 4.0.1 to version 5.0.0. Canadian Meteorological Centre Tech. Note, 59 pp., http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/tech_notes/technote_gdps-500_20151215_e.pdf.
Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998a: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373–1395, https://doi.org/10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.
Côté, J., J.-G. Desmarais, S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998b: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 1397–1418, https://doi.org/10.1175/1520-0493(1998)126<1397:TOCMGE>2.0.CO;2.
Cummings, J., and Coauthors, 2009: Ocean data assimilation systems for GODAE. Oceanography, 22, 96–109, https://doi.org/10.5670/oceanog.2009.69.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Delage, Y., 1997: Parameterising sub-grid scale vertical transport in atmospheric models under statically stable conditions. Bound.-Layer Meteor., 82, 23–48, https://doi.org/10.1023/A:1000132524077.
Delage, Y., and C. Girard, 1992: Stability functions correct at the free convection limit and consistent for both the surface and Ekman layers. Bound.-Layer Meteor., 58, 19–31, https://doi.org/10.1007/BF00120749.
Divakaran, P., and Coauthors, 2015: GODAE OceanView inter-comparison for the Australian region, J. Oper. Oceanogr., 8, s112–s126, https://doi.org/10.1080/1755876X.2015.1022333.
Fu, X., B. Wang, D. E. Waliser, and L. Tao, 2007: Impact of atmosphere–ocean coupling on the predictability of monsoon intraseasonal oscillations. J. Atmos. Sci., 64, 157–174, https://doi.org/10.1175/JAS3830.1.
Girard, C., and Coauthors, 2014: Staggered vertical discretization of the Canadian Environmental Multiscale (GEM) model using a coordinate of the log-hydrostatic-pressure type. Mon. Wea. Rev., 142, 1183–1196, https://doi.org/10.1175/MWR-D-13-00255.1.
Glenn, S. M., and Coauthors, 2016: Stratified coastal ocean interactions with tropical cyclones. Nat. Commun., 7, 10887, https://doi.org/10.1038/ncomms10887.
Hines, C. O., 1997a: Doppler-spread parameterization of gravity-wave momentum deposition in the middle atmosphere. Part 1: Basic formulation. J. Atmos. Sol.-Terr. Phys., 59, 371–386, https://doi.org/10.1016/S1364-6826(96)00079-X.
Hines, C. O., 1997b: Doppler-spread parameterization of gravity-wave momentum deposition in the middle atmosphere. Part 2: Broad and quasi monochromatic spectra, and implementation. J. Atmos. Sol.-Terr. Phys., 59, 387–400, https://doi.org/10.1016/S1364-6826(96)00080-6.
Hong, X., S. W. Chang, S. Raman, L. K. Shay, and R. Hodur, 2000: The interaction between Hurricane Opal (1995) and a warm core ring in the Gulf of Mexico. Mon. Wea. Rev., 128, 1347–1365, https://doi.org/10.1175/1520-0493(2000)128<1347:TIBHOA>2.0.CO;2.
Hourdin, F., A. Găinusă‐Bogdan, P. Braconnot, J.-L. Dufresne, A.-K. Traore, and C. Rio, 2015: Air moisture control on ocean surface temperature, hidden key to the warm bias enigma. Geophys. Res. Lett., 42, 10 885–10 893, https://doi.org/10.1002/2015GL066764.
Hunke, E. C., 2001: Viscous–plastic sea ice dynamics with the EVP model: Linearization issues. J. Comput. Phys., 170, 18–38, https://doi.org/10.1006/jcph.2001.6710.
Hunke, E. C., and J. K. Dukowicz, 1997: An elastic–viscous–plastic model for sea ice dynamics. J. Phys. Oceanogr., 27, 1849–1867, https://doi.org/10.1175/1520-0485(1997)027<1849:AEVPMF>2.0.CO;2.
Hunke, E. C., and J. K. Dukowicz, 2002: The elastic–viscous–plastic sea ice dynamics model in general orthogonal curvilinear coordinates on a sphere—Incorporation of metric terms. Mon. Wea. Rev., 130, 1848–1865, https://doi.org/10.1175/1520-0493(2002)130<1848:TEVPSI>2.0.CO;2.
Hunke, E. C., and W. H. Lipscomb, 2010: CICE: The Los Alamos Sea Ice Model documentation and software user’s manual. Los Alamos National Laboratory Tech Rep. LA-CC-06-012, 76 pp.
Ito, K., T. Kuroda, K. Saito, and A. Wada, 2015: Forecasting a large number of tropical cyclone intensities around Japan using a high-resolution atmosphere–ocean coupled model. Wea. Forecasting, 30, 793–808, https://doi.org/10.1175/WAF-D-14-00034.1.
Jacob, S. D., and L. K. Shay, 2003: The role of oceanic mesoscale features on the tropical cyclone–induced mixed layer response: A case study. J. Phys. Oceanogr., 33, 649–676, https://doi.org/10.1175/1520-0485(2003)33<649:TROOMF>2.0.CO;2.
Jordan, R. E., E. L. Andreas, and A. P. Makshtas, 1999: Heat budget of snow-covered sea ice at North Pole 4. J. Geophys. Res., 104, 7785–7806, https://doi.org/10.1029/1999JC900011.
Jullien, S., P. Marchesiello, C. E. Menkes, J. Lefèvre, N. C. Jourdain, G. Samson, and M. Lengaigne, 2014: Ocean feedback to tropical cyclones: Climatology and processes. Climate Dyn., 43, 2831–2854, https://doi.org/10.1007/s00382-014-2096-6.
Jung, T., and F. Vitart, 2006: Short-range and medium-range weather forecasting in the extratropics during wintertime with and without an interactive ocean. Mon. Wea. Rev., 134, 1972–1986, https://doi.org/10.1175/MWR3206.1.
Jung, T., and Coauthors, 2016: Advancing polar prediction capabilities on daily to seasonal time scales. Bull. Amer. Meteor. Soc., 97, 1631–1647, https://doi.org/10.1175/BAMS-D-14-00246.1.
Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47, 2784–2802, https://doi.org/10.1175/1520-0469(1990)047<2784:AODEPM>2.0.CO;2.
Kain, J. S., and J. M. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.
Laloyaux, P., M. Balmaseda, D. Dee, K. Mogensen, and P. Janssen, 2016: A coupled data assimilation system for climate reanalysis. Quart. J. Roy. Meteor. Soc., 142, 65–78, https://doi.org/10.1002/qj.2629.
Lea, D., I. Mirouze, M. Martin, R. King, A. Hines, D. Walters, and M. Thurlow, 2015: Assessing a new coupled data assimilation system based on the Met Office coupled atmosphere–land–ocean–sea ice model. Mon. Wea. Rev., 143, 4678–4694, https://doi.org/10.1175/MWR-D-15-0174.1.
Lee, C. Y., and S. S. Chen, 2012: Symmetric and asymmetric structures of hurricane boundary layer in coupled atmosphere–wave–ocean models and observations. J. Atmos. Sci., 69, 3576–3594, https://doi.org/10.1175/JAS-D-12-046.1.
Lellouche, J.-M., and Coauthors, 2013: Evaluation of global monitoring and forecasting systems at Mercator Océan. Ocean Sci., 9, 57–81, https://doi.org/10.5194/os-9-57-2013.
Lemieux, J.-F., D. A. Knoll, B. Tremblay, D. M. Holland, and M. Losch, 2012: A comparison of the Jacobian-free Newton–Krylov method and the EVP model for solving the sea ice momentum equation with a viscous-plastic formulation: A serial algorithm study. J. Comput. Phys., 231, 5926–5944, https://doi.org/10.1016/j.jcp.2012.05.024.
Lemieux, J.-F., B. L. Tremblay, F. Dupont, M. Plante, G. C. Smith, and D. Dumont, 2015: A basal stress parameterization for modeling landfast ice. J. Geophys. Res. Oceans, 120, 3157–3173, https://doi.org/10.1002/2014JC010678.
Li, J., and H. W. Barker, 2005: A radiation algorithm with correlated-k distribution. Part I: Local thermal equilibrium. J. Atmos. Sci., 62, 286–309, https://doi.org/10.1175/JAS-3396.1.
Lipscomb, W. H., E. C. Hunke, W. Maslowski, and J. Jakacki, 2007: Ridging, strength, and stability in high-resolution sea ice models. J. Geophys. Res., 112, C03S91, https://doi.org/10.1029/2005JC003355.
Lott, F., and M. J. Miller, 1997: A new subgrid-scale orographic drag parametrization: Its formulation and testing. Quart. J. Roy. Meteor. Soc., 123, 101–127, https://doi.org/10.1002/qj.49712353704.
Ma, Z., J. Fei, L. Liu, X. Huang, and Y. Li, 2017: An investigation of the influences of mesoscale ocean eddies on tropical cyclone intensities. Mon. Wea. Rev., 145, 1181–1201, https://doi.org/10.1175/MWR-D-16-0253.1.
Madec, G., 2008: NEMO reference manual, ocean dynamics component: NEMO-OPA. Preliminary version. IPSL Note du Pole de modélisation 27, 91 pp.
Madec, G., P. Delécluse, M. Imbard, C. Lévy, 1998: OPA 8.1 ocean general circulation model reference manual. IPSL Note du Pole de modélisation 11, 91 pp.
Martin, M., and Coauthors, 2012: Group for High Resolution Sea Surface Temperature (GHRSST) analysis fields inter-comparisons. Part 1: A GHRSST multi-product ensemble (GMPS). Deep-Sea Res. II, 77–80, 21–30, https://doi.org/10.1016/j.dsr2.2012.04.013.
McFarlane, N. A., 1987: The effect of orographically excited gravity wave drag on the general circulation of the lower stratosphere and troposphere. J. Atmos. Sci., 44, 1775–1800, https://doi.org/10.1175/1520-0469(1987)044<1775:TEOOEG>2.0.CO;2.
McTaggart-Cowan, R., and A. Zadra, 2015: Representing Richardson number hysteresis in the NWP boundary layer. Mon. Wea. Rev., 143, 1232–1258, https://doi.org/10.1175/MWR-D-14-00179.1.
Megann, A., and Coauthors, 2014: GO5.0: The joint NERC–Met Office NEMO global ocean model for use in coupled and forced applications. Geosci. Model Dev., 7, 1069–1092, https://doi.org/10.5194/gmd-7-1069-2014.
Mulholland, D. P., P. Laloyaux, K. Haines, and M. A. Balmaseda, 2015: Origin and impact of initialization shocks in coupled atmosphere–ocean forecasts. Mon. Wea. Rev., 143, 4631–4644, https://doi.org/10.1175/MWR-D-15-0076.1.
Nicholls, S. D., and S. G. Decker, 2015: Impact of coupling an ocean model to WRF nor’easter simulations. Mon. Wea. Rev., 143, 4997–5016, https://doi.org/10.1175/MWR-D-15-0017.1.
Noilhan, J., and S. Planton, 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, 536–549, https://doi.org/10.1175/1520-0493(1989)117<0536:ASPOLS>2.0.CO;2.
Pellerin, P., H. Ritchie, F. J. Saucier, F. Roy, S. Desjardins, M. Valin, and V. Lee, 2004: Impact of a two-way coupling between an atmospheric and an ocean-ice model over the Gulf of St. Lawrence. Mon. Wea. Rev., 132, 1379–1398, https://doi.org/10.1175/1520-0493(2004)132<1379:IOATCB>2.0.CO;2.
Pham, D. T., J. Verron, and M. C. Roubaud, 1998: A singular evolutive extended Kalman filter for data assimilation in oceanography. J. Mar. Syst., 16, 323–340, https://doi.org/10.1016/S0924-7963(97)00109-7.
Price, J. F., 1981: Upper ocean response to a hurricane. J. Phys. Oceanogr., 11, 153–175, https://doi.org/10.1175/1520-0485(1981)011<0153:UORTAH>2.0.CO;2.
Pudykiewicz, J., R. Benoit, and J. Mailhot, 1992: Inclusion and verification of a predictive cloud-water scheme in a regional numerical weather prediction model. Mon. Wea. Rev., 120, 612–626, https://doi.org/10.1175/1520-0493(1992)120<0612:IAVOAP>2.0.CO;2.
Putrasahan, D. A., A. J. Miller, and H. Seo, 2013: Regional coupled ocean–atmosphere downscaling in the southeast Pacific: Impacts on upwelling, mesoscale air–sea fluxes, and ocean eddies. Ocean Dyn., 63, 463–488, https://doi.org/10.1007/s10236-013-0608-2.
Qaddouri, A., and V. Lee, 2011: The Canadian Global Environmental Multiscale model on the Yin-Yang grid system. Quart. J. Roy. Meteor. Soc., 137, 1913–1926, https://doi.org/10.1002/qj.873.
Renault, L., J. Chiggiato, J. C. Warner, M. Gomez, G. Vizoso, and J. Tintoré, 2012: Coupled atmosphere–ocean–wave simulations of a storm event over the Gulf of Lion and Balearic Sea. J. Geophys. Res., 117, C09019, https://doi.org/10.1029/2012JC007924.
Roullet, G., and G. Madec, 2000: Salt conservation, free surface, and varying levels: A new formulation for ocean general circulation models. J. Geophys. Res., 105, 23 927–23 942, https://doi.org/10.1029/2000JC900089.
Roy, F., M. Chevallier, G. C. Smith, F. Dupont, G. Garric, J.-F. Lemieux, Y. Lu, and F. Davidson, 2015: Arctic sea ice and freshwater sensitivity to the treatment of the atmosphere–ice–ocean surface layer. J. Geophys. Res., 120, 4392–4417, https://doi.org/10.1002/2014JC010677.
Ryan, A. G., and Coauthors, 2015: GODAE OceanView Class 4 forecast verification framework: Global ocean inter-comparison. J. Oper. Oceanogr., 8, s98–s111, https://doi.org/10.1080/1755876X.2015.1022330.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1058, https://doi.org/10.1175/2010BAMS3001.1.
Sandery, P. A., and T. J. O’Kane, 2014: Coupled initialization in an ocean–atmosphere tropical cyclone prediction system. Quart. J. Roy. Meteor. Soc., 140, 82–95, https://doi.org/10.1002/qj.2117.
Sandery, P. A., G. B. Brassington, A. Craig, and T. Pugh, 2010: Impacts of ocean–atmosphere coupling on tropical cyclone intensity change and ocean prediction in the Australian region. Mon. Wea. Rev., 138, 2074–2091, https://doi.org/10.1175/2010MWR3101.1.
Schade, L. R., and K. A. Emanuel, 1999: The ocean’s effect on the intensity of tropical cyclones: Results from a simple coupled atmosphere–ocean model. J. Atmos. Sci., 56, 642–651, https://doi.org/10.1175/1520-0469(1999)056<0642:TOSEOT>2.0.CO;2.
Scoccimarro, E., P. G. Fogli, K. A. Reed, S. Gualdi, S. Masina, and A. Navarra, 2017: Tropical cyclone interaction with the ocean: The role of high-frequency (subdaily) coupled processes. J. Climate, 30, 145–162, https://doi.org/10.1175/JCLI-D-16-0292.1.
Seo, H., A. J. Miller, and J. O. Roads, 2007: The Scripps Coupled Ocean–Atmosphere Regional (SCOAR) model, with applications in the eastern Pacific sector. J. Climate, 20, 381–402, https://doi.org/10.1175/JCLI4016.1.
Seo, H., A. C. Subramanian, A. J. Miller, and N. R. Cavanaugh, 2014: Coupled impacts of the diurnal cycle of sea surface temperature on the Madden–Julian oscillation. J. Climate, 27, 8422–8443, https://doi.org/10.1175/JCLI-D-14-00141.1.
Seroka, G., T. Miles, Y. Xu, J. Kohut, O. Schofield, and S. Glenn, 2016: Hurricane Irene sensitivity to stratified coastal ocean cooling. Mon. Wea. Rev., 144, 3507–3530, https://doi.org/10.1175/MWR-D-15-0452.1.
Shelly, A., P. Xavier, D. Copsey, T. Johns, J. M. Rodríguez, S. Milton, and N. Klingaman, 2014: Coupled versus uncoupled hindcast simulations of the Madden–Julian Oscillation in the Year of Tropical Convection. Geophys. Res. Lett., 41, 5670–5677, https://doi.org/10.1002/2013GL059062.
Small, R. J., and Coauthors, 2008: Air–sea interaction over ocean fronts and eddies. Dyn. Atmos. Oceans, 45, 274–319, https://doi.org/10.1016/j.dynatmoce.2008.01.001.
Small, R. J., T. Campbell, J. Teixeira, S. Carniel, T. A. Smith, J. Dykes, S. Chen, and R. Allard, 2011: Air–sea interaction in the Ligurian Sea: Assessment of a coupled ocean–atmosphere model using in situ data from LASIE07. Mon. Wea. Rev., 139, 1785–1808, https://doi.org/10.1175/2010MWR3431.1.
Smith, G. C., and Coauthors, 2013: Small-scale ice-ocean-wave processes and their impact on coupled environmental polar prediction. Proc. ECMWF-WWRP/THORPEX Polar Prediction Workshop, Reading, United Kingdom, ECMWF, 24 pp.
Smith, G. C., and Coauthors, 2016a: Sea ice forecast verification in the Canadian Global Ice Ocean Prediction System. Quart. J. Roy. Meteor. Soc., 142, 659–671, https://doi.org/10.1002/qj.2555.
Smith, G. C., and Coauthors, 2016b: The CMC Global Ice Ocean Prediction System v2.1. Canadian Centre for Meteorological and Environmental Prediction Tech. Note, 33 pp., http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/tech_notes/technote_giops-210_e.pdf.
Sundqvist, H., 1978: A parametrization scheme for non-convective condensation including prediction of cloud water content. Quart. J. Roy. Meteor. Soc., 104, 677–690, https://doi.org/10.1002/qj.49710444110.
Tranchant, B., C. E. Testut, L. Renault, N. Ferry, F. Birol, and P. Brasseur, 2008: Expected impact of the future SMOS and Aquarius Ocean surface salinity missions in the Mercator Océan operational systems: New perspectives to monitor ocean circulation. Remote Sens. Environ., 112, 1476–1487, https://doi.org/10.1016/j.rse.2007.06.023.
Valcke, S., and Coauthors, 2012: Coupling technologies for Earth system modelling. Geosci. Model Dev., 5, 1589–1596, https://doi.org/10.5194/gmd-5-1589-2012.
Vitart, F., and Coauthors, 2008: The new VAREPS-monthly forecasting system: A first step towards seamless prediction. Quart. J. Roy. Meteor. Soc., 134, 1789–1799, https://doi.org/10.1002/qj.322.
Vosper, S. B., H. Wells, and A. R. Brown, 2009: Accounting for non-uniform static stability in orographic drag parametrization. Quart. J. Roy. Meteor. Soc., 135, 815–822, https://doi.org/10.1002/qj.407.
Wells, H., S. B. Vosper, A. N. Ross, A. R. Brown, and S. Webster, 2008: Wind direction effects on orographic drag. Quart. J. Roy. Meteor. Soc., 134, 689–701, https://doi.org/10.1002/qj.247.
Winterbottom, H. R., E. W. Uhlhorn, and E. P. Chassignet, 2012: A design and an application of a regional coupled atmosphere-ocean model for tropical cyclone prediction. J. Adv. Model. Earth Syst., 4, M10002, https://doi.org/10.1029/2012MS000172.
Zadra, A., M. Roch, S. Laroche, and M. Charron, 2003: The subgrid-scale orographic blocking parametrization of the GEM model. Atmos.–Ocean, 41, 155–170, https://doi.org/10.3137/ao.410204.
Zadra, A., R. McTaggart-Cowan, P. A. Vaillancourt, M. Roch, S. Bélair, and A.-M. Leduc, 2014: Evaluation of tropical cyclones in the Canadian Global Modeling System: Sensitivity to moist process parameterization. Mon. Wea. Rev., 142, 1197–1220, https://doi.org/10.1175/MWR-D-13-00124.1.
Zheng, Z.-W., C.-R. Ho, Q. Zheng, Y.-T. Lo, N.-J. Kuo, and G. Gopalakrishnan, 2010: Effects of preexisting cyclonic eddies on upper ocean responses to category 5 typhoons in the western North Pacific. J. Geophys. Res., 115, C09013, https://doi.org/10.1029/2009JC005562.
