1. Introduction
El Niño–Southern Oscillation (ENSO) is the largest interannual climate mode, with its core in the equatorial Pacific but having global influence through atmospheric teleconnections (e.g., Trenberth et al. 1998). Its global impact and potential for making societally useful long-range predictions have been the primary motivation of investment in the Tropical Pacific Observing System (TPOS). The current TPOS was initially populated by the Tropical Atmosphere Ocean (TAO) array in the early 1980s (McPhaden et al. 1998) and was later enhanced by the Triangle Trans-Ocean Buoy Network (TRITON) array in the western tropical Pacific (west of 160°E) after 2000 (Ando et al. 2005). The TAO/TRITON array has greatly advanced our ability to describe, understand, and predict ENSO variability by sampling both the surface meteorology and the subsurface ocean through the thermocline (McPhaden et al. 2010). Currently, the observation capabilities of in situ TPOS are supplemented by Argo floats and surface drifters.
The TAO/TRITON array, however, has been confronted with ongoing challenges. Sustained observations from moored array suffered from the phasing out of the NOAA ship Ka’imimoanoa, which was routinely tasked with cruises for the maintenance of the TAO array. As a result of the decommissioning of Ka’imimoanoa, the data return from the Pacific tropical moored buoy array (TMA, meaning the collective TAO/TRITON moored array) reduced from a regular rate of 80%–90% to around 40% during 2012–14. The sharp decline in the observations from the TMA has been referred to as the “2012–2014 TAO crisis” (Tollefson 2014). The second TMA crisis emerged from the recent phasing out of the TRITON array because of a reduction in investments by the sponsor of TRITON, Japan’s Agency for Marine-Earth Science and Technology (JAMSTEC). Since its decline in 2017, only one near-equatorial TRITON mooring remains in the western Pacific. The near-equatorial TRITON moorings were critical for monitoring ENSO onset because high-frequency winds (e.g., the westerly wind events during an El Niño) are usually observed in the western tropical Pacific at the start of an event (e.g., Eisenman et al. 2005).
Besides ongoing issues in the maintenance of the TMA since the completion of the TAO moored array in the 1990s, there have been advances in new observing capabilities, including the emergence of altimetry, Argo, satellite wind, and ocean color observations, among others. For example, about 4000 Argo floats are now distributed over the global ocean with a nominal spacing of about 3° both zonally and meridionally, and each float returns a profile of temperature and salinity from 2000-m depth to the sea surface every 10 days (Roemmich and Owens 2000; Roemmich et al. 2019).
These advances in ocean observation technologies have complemented measurements from the moored arrays in different ways. Compared to the TAO/TRITON array, for instance, satellite observations have the advantage of global coverage and higher spatial resolution, and therefore capture a larger fraction of the spatial scales of variability, and further, potentially providing more reliable large-scale integrals (e.g., wind fetch) and derivatives (wind curl and divergence). Similarly, Argo, by sampling temperature and salinity more densely in both the zonal and vertical directions, provides geostrophic currents on scales appropriate for diagnoses of low-frequency phenomena, which is not the case for the TAO/TRITON. More recently, new versions of Argo floats have also been developed that are more suitable for measurements along the equator (Gasparin et al. 2015; Roemmich et al. 2019), where divergent surface currents can cause drifting floats to be swept away from the equator. Thus, the emerging observing techniques have the potential to greatly contribute to the future design and evolution of the TPOS.
Motivated by a dramatic decline in the sustained observations from TAO/TRITON, the scientific community organized a TPOS 2020 workshop in January 2014 (http://www.ioc-goos.org/tpos2020). The primary goal of the workshop was to evaluate the needs of the TPOS and its future evolution toward a more resilient observing system. An outcome of the workshop was the initiation of the TPOS 2020 Project (Cravatte et al. 2016) with goals to evaluate and recommend changes in the configuration of the current TPOS in the light of advances in the ocean observing technologies since the implementation of TAO in 1994. By embracing the integration of diverse and new observing technologies, the proposed TPOS 2020 recommendations aim to evolve the TPOS into a more robust and sustainable observing system in support of enhancing our understanding of the tropical Pacific climate variability on different time scales, and to improve its predictions.
In the context of TPOS 2020, in this study a series of ocean observing system simulation experiments (OSSEs) are conducted to evaluate the relative merits of ocean observing platforms in the tropical Pacific. Our focus will be on the in situ component of the present TPOS configuration, particularly the TAO/TRITON array and the Argo network. With this objective in mind, data assimilation (DA) experiments are conducted based on the new ocean data assimilation system that is under development at the Joint Center for Satellite Data Assimilation (JCSDA) and the Environmental Modeling Center (EMC)/National Centers for Environmental Prediction (NCEP). Our DA experiments to understand the influence of the TMA and Argo include a free ocean model run with no observations assimilated, and assimilation runs with the TAO/TRITON and Argo temperature and salinity “observations” assimilated as separate networks or jointly. We note that in the future, similar DA experiments will also be conducted with the upcoming TPOS configuration recommended by the TPOS 2020 Project.
Previously, most observing system experiments (OSEs)/OSSEs about the TPOS (e.g., Balmaseda et al. 2007; Vidard et al. 2007; Yan et al. 2007; Balmaseda and Anderson 2009; Fujii et al. 2015a,b; Xue et al. 2017) have centered on the ENSO time scale. This is not surprising as ENSO drove the implementation of the original TPOS. On the other hand, the variability in the tropical Pacific occurs across a wide range of spatial and temporal scales. In terms of time scale, the variability ranges from diurnal (e.g., Kawai and Wada 2007) to seasonal/interannual and even longer, which is controlled by different basin-dependent processes. On intraseasonal time scales, for example, the sea surface temperatures (SSTs) in the tropical Pacific are modulated by ocean Kelvin waves (e.g., Kessler 2012) and by the MJO, and the eastern equatorial Pacific is also populated by the tropical instability waves (TIWs; Philander et al. 1986). Considering the scales of ocean variability that need to be observed and resolved, the evaluations of our experiments go beyond the ENSO time scale alone, and extend to multiple time scales by additionally examining variations at the high-frequency (<20 days) and intraseasonal (20–90 days) time scales.
The rest of the paper is arranged as follows. Section 2 describes the configuration of the Nature Run, which is used to generate proxy observations, the construction of synthetic observations, and the Joint Effort for Data Assimilation (JEDI)-based ocean data assimilation (ODA) system and experiments. In the study, different ocean models are used in Nature Run and DA runs, which is suggested to be superior to the application of a same model for OSSEs (Halliwell et al. 2014). Section 3 evaluates the relative roles of TAO/TRITON versus Argo on the multiple time scale variations of temperature and salinity, with a short evaluation about ocean currents in section 4. Section 5 briefly examines the sensitivity of experiments to two data assimilation parameters (i.e., correlation scale and observational errors). A summary and discussion are given in section 6.
2. Model and experiments
a. The Nature Run configuration
The Nature Run adopted in this study is the free run with daily mean outputs conducted in Zhu et al. (2020), for which the NCEP Climate Forecast System coupled model with 1-m vertical resolutions in the upper ocean (CFSm501; Zhang et al. 2019) was applied. The free run was integrated for 20 years starting from the CFSR (Saha et al. 2010) initial state at 0000 UTC 2 January 1980, and the 6th–10th years are used as our Nature Run. CFSm501 is a variant of the NCEP Climate Forecast System version 2 (CFSv2; Saha et al. 2014) with two major modifications:
The convection scheme is replaced with the relaxed Arakawa–Schubert (RAS; Moorthi and Suarez 1992) scheme. The update significantly improves the simulations of the MJO (particularly its eastward propagation across the Maritime Continent; Zhu et al. 2017a, 2020);
The vertical ocean resolution was enhanced from 10 to 1 m near the surface (to facilitate the resolution change, the oceanic component of CFSv2 was also replaced by MOM5). The updates improved the oceanic barrier layer distribution, oceanic mixed layer diurnal cycle (Ge et al. 2017; Zhang et al. 2019) and intraseasonal SST/sea surface salinity (SSS) variability (Zhu et al. 2020). For the barrier layer (figures not shown), CFSm501 presents clear improvements over the model version with only the atmospheric convection scheme replaced (i.e., CFSv2RAS; Zhu and Kumar 2019), in both its spatial distributions and thickness values.
All other setups of CFSm501 follow the configurations of CFSv2 (Saha et al. 2014). Its atmospheric component is a lower resolution version of the Global Forecast System (GFS) with a horizontal spectral resolution of T126 (105-km grid spacing) and 64 vertical levels in a hybrid sigma-pressure coordinate. The oceanic horizontal resolution is 0.5° × 0.5° poleward of 30°S/30°N, with the meridional resolution gradually increasing to 0.25° between 10°S and 10°N. The oceanic vertical resolution in CFSm501 is 1 m for the upper 10 m, and 10 m for the layer between the 10- and 200-m depths. The atmospheric and oceanic components exchange surface momentum, heat and freshwater fluxes, SST, and surface information on sea ice, every 30 min. The surface fluxes are parameterized by the GFDL version of bulk formulas based on Beljaars (1995).
We note that although our Nature Run has a resolution of 0.25° between 10°S and 10°N and could be criticized for not having a high enough resolution to qualify as the “nature run” for assessing sampling associated with some variabilities, the horizontal resolution may be adequate to address the sampling issues related to moored array and Argo platforms for purposes of resolving temporal and spatial variability (e.g., the MJO, the barrier layer) that is the focus of OSSE experiments in this paper.
b. Construction of synthetic observations
From the daily output of the Nature Run (see section 2a; Zhu et al. 2020), synthetic observations are constructed following the current TAO/TRITON and Argo configurations (Fig. 1). For TAO/TRITON (58 moorings in total), synthetic temperature and salinity observations (including subsurface salinity at five equatorial sites) are constructed by linearly/bilinearly interpolating from the model grids to all TAO/TRITON sites on the vertical/horizontal direction. Our TAO/TRITON synthetic observations are generated every 24 h. We note that our experiments cannot assess the full value of TAO/TRITON, such as its high-frequency observing capability (as high as minutes) and observations of surface meteorological conditions, but could be considered to address the general impact of these ocean temperature and salinity observations on the large-scale coupled ocean–atmosphere system. Future OSSE experiments can be extended toward assessing sampling strategies at higher temporal scales.
The distributions of TAO/TRITON buoys (green squares; 58 in total) and Argo floats (small orange dots; 436 in total) for OSSE. Argo floats are sampled every 3° × 3° boxes every 10 days corresponding to its current resolution, and TAO/TRITON buoys are sampled according to their current locations every 24 h.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
For Argo (436 floats in total), considering its near-homogeneous distribution, both synthetic temperature and salinity profiles were constructed for 3° × 3° boxes in the equatorial Pacific (orange dots in Fig. 1) and every 10 days, which generally corresponds to its standard spatial and temporal resolutions of sampling (Roemmich and Owens 2000). Vertically, real Argo floats collect temperature and salinity measurements down to 2000 m deep with fine resolutions, usually every 1 m in the top 20 m (most Argo floats have no measurements at depths shallower than 4 m), every 10 m between 20 and 200 m, and every 20 m below 200-m depth. Our synthetic Argo profiles vertically mimic the real Argo’s vertical resolution over the upper ocean (shallower than 300 m) but degenerate to the model vertical resolution at deep oceans where the oceanic vertical resolution in CFSm501 is coarser than a typical Argo profile.
Note that our procedure for constructing synthetic Argo floats does not follow a true Argo distribution, but is a simplified method assuming a homogenous distribution. The procedure is similar to studies assimilating Level 3 satellite data, and it has been applied in other Argo-related OSSE studies (e.g., Gasparin et al. 2020). There are two advantages for this procedure. First, it is simple and can be more easily implemented compared to methods following a true Argo distribution. Second, it avoids uncertainties related to the large temporal/spatial variations of Argo distributions. One example about its distribution variations is along the equator. During the early years, there were not many floats near the equator, because divergent surface currents cause floats to be swept away from the equator. Nowadays, however, by employing new-generation floats that do not drift away from the equator, the Argo sampling along the equator could even be higher than one profile per 3° × 3° box (Gasparin et al. 2015).
One challenge for OSSEs is to construct synthetic observations with realistic error properties (Fujii et al. 2019). In our four main experiments (see section 2d), relatively small observational errors (0.1°C for temperature and 0.03 psu for salinity) are applied in view of several factors: 1) the ocean models used in Nature Run and DA experiments are both from the GFDL model “family” (MOM5 vs MOM6), and the representativeness of errors coming from their different resolutions (2:1 horizontally) is also expected to be small, 2) atmospheric forcing for DA experiments is prescribed from Nature Run with no additional perturbations added, and 3) measurements by TAO/TRITON and Argo have high accuracies (e.g., the original Argo target called for temperature and salinity accuracies of 0.005°C and 0.01 psu, with a pressure accuracy of 2.5 dbar; Riser et al. 2016). In section 5, an additional sensitivity experiment (see section 2d as well) is also conducted by increasing the observational errors by a factor of 5 (i.e., 0.5°C for temperature and 0.15 psu for salinity).
c. A JEDI-based ocean data assimilation system for MOM6
The ocean model used in the assimilation experiments is MOM6 (Adcroft and Hallberg 2006), at a resolution of 1° × 1° and 75 layers in a hybrid vertical coordinate system. We note that the ocean model (MOM6) used for DA experiments is different from the Nature Run (MOM5), and hence they likely have different mean states. This mimics the real situation of assimilating observational data using a model that has biases. Note that the application of different ocean models for OSSEs is suggested to be superior to the application of a same model (Halliwell et al. 2014).
The data assimilation system is based on the JEDI development (Trémolet and Auligné 2020) at the Joint Center for Satellite Data Assimilation (JCSDA). The MOM6 interface to JEDI is also developed at the JCSDA under the Sea ice Ocean Coupled Assimilation Project (SOCA; Holdaway et al. 2020). This study uses a subset of the current capability of the SOCA-developed JEDI interface to MOM6. The assimilation cycle of the JEDI-based ODA system is daily. This section only describes the features used in this work to allow for the assimilation of synthetic potential temperature and practical salinity profiles using an incremental 3DVAR (e.g., Weaver et al. 2005).
The two forward operators used in the assimilation experiments are implemented in a generic way within the Unified Forward Operator (UFO; https://github.com/JCSDA/ufo) repository of the JEDI system for the vertical interpolation and within SOCA for the horizontal interpolation. Both UFO and SOCA provide the adjoint operators needed for the variational minimization.
d. Experiments
Utilizing the JEDI-based ODA system and the synthetic TAO/TRITON and Argo observations, four main experiments are conducted together with two additional sensitivity studies (Table 1).
The first experiment (referred to as noDA; Table 1) is a forced free run with no synthetic observations assimilated. The experiment is conducted with the same 1° resolution MOM6 as in other DA experiments. It will be a benchmark, giving an estimate about how the tropical Pacific state can be reconstructed by an ocean model if only the overlying atmospheric conditions are perfectly known. This simulation also gives an estimate of “biases” in the ocean model used for OSSE compared to the one in the Nature Run. Any improvements relative to noDA in the following DA experiments represent the additional values of ocean observations in reconstructing the ocean states.
The second, third, and fourth experiments (referred to as crtTAO, Argo, and crtTAO+Argo, respectively; Table 1) are designed to explore the relative roles of TAO/TRITON and Argo in their current TPOS configuration. Synthetic TAO/TRITON and Argo observations (see section 2c) are separately assimilated in the crtTAO and Argo experiments, and jointly assimilated in the crtTAO+Argo experiment.
The fifth and sixth experiments (referred to as crtTAO_RScl and crtTAO_obserr; Table 1) are two sensitivity runs designed to test two DA parameters (i.e., correlation scale and observational errors). Other than the tested parameters, all other configurations and parameters in the two experiments are set as the same as in crtTAO. In crtTAO_RScl the correlation scale is increased to 1400 km from the default scale of 700 km that is applied in all other DA experiments. The crtTAO_obserr experiment is conducted to evaluate the sensitivity of results to observational errors that are 5 times as large as in other DA experiments.
List of experiments.
All the above six experiments are driven by the same daily mean atmospheric forcing from the Nature Run (see section 2a; Zhu et al. 2020), a procedure assuming that the atmospheric state is perfectly known. It should be noted that the assumption of “perfectly” known atmospheric conditions clearly overstates the efficiency of a forced ocean model run in reproducing the tropical Pacific states, and correspondingly may underestimate the contribution of ocean observations. The experiments are all initialized from the same ocean state, which is achieved by a 13-yr spinup run driven by the Nature Run forcing. Based on the atmospheric forcing and initial condition, four main experiments are conducted for five years and two sensitivity experiments for one month, with all simulations saving daily outputs (which are compared with the corresponding outputs from the Nature Run).
3. Evaluation of temperature and salinity
a. Comparison of mean biases (
This section evaluates how the mean state
Mean difference of (left) SST (°C) and (right) SSS (psu) with respect to the Nature run in (a),(e) noDA, (b),(f) crtTAO, (c),(g) Argo, and (d),(h) crtTAO+Argo. The black squares in (b), (d), (f), and (h) and the small blue dots in (c), (d), (g), and (h) indicate where TAO/TRITON buoys and Argo floats are located, respectively.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
Mean difference of (left) temperature (°C) and (right) salinity (psu) along the equator with respect to the Nature Run in (a),(e) noDA, (b),(f) crtTAO, (c),(g), Argo, and (d),(h) crtTAO+Argo. In (a), the 20°C isothermal is also included as a solid (dashed) curve for the Nature Run (noDA). The black squares in (b), (d), (f), and (h) and the small gray dots in (c), (d), (g), and (h) indicate where TAO/TRITON and Argo measurements are available, respectively.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
In general, the mean temperature difference between noDA and the Nature Run is not large at the surface (Fig. 2a), partially because the ocean surface states are strongly constrained by surface fluxes, which are set the same in all cases. For SST in noDA (Fig. 2a), in addition to warming bias along the longitudes of 10°–20°N, there are clear warm biases (~1°C) over the equatorial western Pacific, which extend downward to around 50-m depth (Fig. 3a). Additionally, there are cold biases over the off-equatorial northeastern Pacific and the far eastern Pacific basin, with an increased cold bias of approximately −2°C in the east coastal Pacific region. For the subsurface Pacific, the simulated thermocline in noDA (dashed curve in Fig. 3a) is shallower than that in Nature Run (solid curve in Fig. 3a). Correspondingly, noDA is clearly colder than the Nature Run at the subsurface, with a cold bias of about −1.5°C centered along the thermocline.
When the TAO/TRITON synthetic observations are assimilated (i.e., the crtTAO experiment), there is no improvement about the surface warm biases along 10°–20°N (Fig. 2b vs Fig. 2a), which can be expected from the positioning of TAO/TRITON moorings (black squares in Fig. 2b). For the same reason, the cold bias in the eastern coastal region in noDA is also not corrected. In contrast, the warm bias in the equatorial western Pacific and cold bias over the off-equatorial northeastern Pacific are well constrained by the TAO/TRITON observations. At the subsurface, the strong cold bias along the thermocline is effectively constrained as well (Fig. 3b vs Fig. 3a). For the near-surface ocean, assimilation of TAO/TRITON reduces the warm bias in the western basin but enhances it in the central and eastern basin except near the mooring sites (a feature that will be discussed later). The latter degradation is also present at the surface (Fig. 2a vs Fig. 2b).
When the Argo synthetic observations are assimilated (i.e., the Argo experiment), the SST mean bias distribution is generally like that in crtTAO in the central and eastern equatorial Pacific with some warm biases. Due to the better spatial coverage of Argo (small dots in Fig. 2c) than TAO/TRITON, however, some improvements are also seen, such as in correcting the cold biases in the eastern coastal Pacific and part of warm biases along 10°–20°N. For the subsurface ocean, the strong cold bias is also well corrected by Argo, and the correction is even better than for TAO/TRITON experiment due to a better spatial coverage of Argo. When both TAO/TRITON and Argo observations are assimilated, a further improvement in bias correction over Argo is achieved near the TAO/TRITON mooring sites (Fig. 2d vs Fig. 2c; Fig. 3d vs Fig. 3c).
The results also indicate that DA increased biases for temperature; for example, the surface (Figs. 2b–d) and near-surface (Figs. 3b–d) ocean present clear warm biases relative to the Nature Run in the eastern equatorial Pacific in three DA experiments (although they are negligible close to the TAO/TRITON mooring sites in crtTAO and crtTAO+Argo). The warm biases are much smaller or negligible in noDA (Figs. 2a and 4a), but become significant in DA experiments. This deterioration might be associated with the subsurface thermal difference between three DA experiments and noDA (Figs. 3b–d vs Fig. 3a). In noDA (Fig. 3a) a cooler subsurface water is likely to impart cold bias in the near-surface ocean of the eastern equatorial Pacific as a result of climatological upwelling. The cold biases, however, are absent on the equator in noDA (Figs. 2a and 5a), suggesting that they might be offset by some warming effects related to other physics/processes (e.g., ocean boundary layer parameterizations). Thus, when the strong subsurface cold bias is corrected by DA (Figs. 3b–c), its associated cooling effect on the near-surface ocean is absent. However, warming effects remain leading to the emergence of warm biases in the near surface of the equatorial eastern Pacific in three DA experiments (Figs. 2b–d).
Root-mean-square differences (RMSD) of low-frequency (left) SST (°C) and (right) SSS (psu) with respect to the Nature Run in (a),(e) noDA, (b),(f) crtTAO, (c),(g) Argo, and (d),(h) crtTAO+Argo. The green squares in (b), (d), (f), and (h) and the small blue dots in (c), (d), (g), and (h) indicate where TAO/TRITON buoys and Argo floats are located, respectively.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
Root-mean-square differences (RMSD) of low-frequency (left) temperature (°C) and (right) salinity (psu) along the equator with respect to the Nature Run in (a),(e) noDA, (b),(f) crtTAO, (c),(g) Argo, and (d),(h) crtTAO+Argo. The green squares in (b), (d), (f), and (h) and the small gray dots in (c), (d), (g), Nd (h) indicate where TAO/TRITON and Argo measurements are available, respectively.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
For salinity, biases are also small at the surface in noDA (Fig. 2e) compared to the typical SSS biases in contemporary climate models, with some positive biases in the central southern tropical Pacific and small negative biases in the western Pacific. At the subsurface (Fig. 3e), the entire equatorial Pacific features a freshening bias, and the largest biases are present in the upper ocean of the western warm pool region and a layer below the thermocline. In the crtTAO experiment, the surface biases are reduced in both the central southern tropical Pacific and the western Pacific (Fig. 2f). For the subsurface freshening bias, the salinity observations at five TAO/TRITON mooring sites (black squares in Fig. 3f) make effective corrections for the upper 150 m ocean where salinity is sampled by TAO/TRITON, but there are generally no improvements in the far western basin and the ocean deeper than 150 m.
When Argo profiles alone are assimilated (i.e., the Argo experiment), improvements in salinity become even more evident. For the surface salinity, the positive bias in the central southern tropical Pacific in noDA (Fig. 2e) disappears in the Argo experiment (Fig. 2g). For the subsurface salinity, the widespread freshening bias (Fig. 3e) is also well corrected, with some residual freshening remaining in the near-surface ocean (Fig. 3g). When the TAO/TRITON and Argo observations are jointly assimilated, the salinity corrections are similar to the Argo experiment (Fig. 2h vs Fig. 2g and Fig. 3h vs Fig. 3g), suggesting a limited role of TAO/TRITON in correcting salinity biases.
b. Comparison of low-frequency component (
In this section, the low-frequency (>90 days) component is evaluated, which corresponds to the seasonal cycle and interannual variability in this study and is dominated by ENSO in the tropical Pacific. We note that a 5-yr integration is not long enough to get a statistically reliable interannual mode, but it should be able to capture certain interannual variability (the 5-yr period for DA experiments includes a strong La Niña event in the Nature Run). Figure 4 shows the root-mean-square differences (RMSDs) of low-frequency SSTs and SSSs in the four main experiments relative to the Nature Run, and Fig. 5 presents the subsurface comparisons along the equator.
In noDA (Figs. 4a and 7a), the RMSD distribution of low-frequency temperature indicates its linkage with ENSO variability, but with contributions from seasonal cycle as well. In particular, the low-frequency SST features large RMSDs (>1.0°C) in the central and eastern Pacific Ocean with small RMSDs over other regions; at the subsurface, the largest RMSDs (>1.2°C) are present along the equatorial thermocline. All the features are like the variance distribution of ENSO-related temperature variability (e.g., Wang et al. 2016). For low-frequency salinity, its RMSD distribution presents a good resemblance to the mean bias distribution (Fig. 4e vs Fig. 2e; Fig. 5e vs Fig. 3e). For example, at the surface (Fig. 4e), large RMSDs of salinity (>0.6 psu) appear in the central southern tropical Pacific, a region with large mean biases (Fig. 2e); at the equatorial subsurface (Fig. 5e), the largest RMSDs (>0.2 psu) exist in the upper ocean of the western warm pool region, and relatively large RMSDs (~0.1 psu) are also evident in a layer below the equatorial thermocline sloping up from the west to the east with larger RMSDs at its two ends. Compared to the amplitudes of low-frequency variations in the Nature Run (see Figs. S1a,d and S2a,d in the online supplemental material), the RMSDs for both temperature and salinity are substantially smaller in noDA, suggesting the value of atmosphere-driven numerical models in reproducing the low-frequency variabilities.
When the TAO/TRITON synthetic observations are assimilated (i.e., the crtTAO experiment), the low-frequency temperature is clearly better reproduced. At surface (Fig. 4b), the large RMSDs in the central and eastern Pacific in noDA become much smaller in crtTAO, with the RMSDs of low-frequency SST generally below 1.0°C over the entire equatorial Pacific. At the subsurface (Fig. 5b), the RMSDs are also reduced from >1.2°C along the equatorial thermocline in noDA to below 0.8°C in almost the entire upper 300-m ocean in crtTAO. For the low-frequency salinity, however, crtTAO presents fidelity similar to that of noDA (Fig. 4f vs Fig. 4e and Fig. 5f vs Fig. 5e), indicating that the current TAO/TRITON configuration cannot improve the representation of the low-frequency salinity variability.
When the Argo synthetic observations are assimilated (i.e., the Argo experiment), the low-frequency temperature and salinity are better constrained. The RMSDs of SST are smaller than 0.5°C over almost the entire equatorial Pacific (Fig. 4c), and the subsurface temperature RMSDs are also reduced to below 0.4°C over most of the vertical section along the equator (Fig. 5c). The improvement by the Argo observations is more evident in salinity; the RMSDs of low-frequency SSS in the Argo experiment are smaller than 0.2 psu over most of the equatorial Pacific (Fig. 4g), and the subsurface counterparts are also mostly reduced to below 0.07 psu (Fig. 5g). In the crtTAO+Argo experiment (Figs. 4d, 6h, and 7d,h), all RMSD distributions are almost identical to their counterparts in the Argo experiment, but further improvements can also be seen by additionally assimilating the TAO/TRITON observations; for example, the SST RMSDs become slightly smaller in the eastern equatorial Pacific in crtTAO+Argo (Fig. 4d vs Fig.4c).
Time series of RMSD averaged over the Niño-3.4 (170°E–120°W, 5°S–5°N) region with respect to the Nature Run for noDA (black line), crtTAO (red line), Argo (green line), and crtTAO+Argo (blue line) at depths of (a),(f) 0.5, (b),(g) 55, (c),(h) 105, (d),(i) 205, and (e),(j) 303 m. All RMSDs are calculated for the low-frequency component with (left) temperature (°C) and (right) salinity (psu).
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
As in Fig. 4, but for the intraseasonal component.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
Figure 6 presents time series of RMSDs averaged over the Niño-3.4 (170°E–120°W, 5°S–5°N) region at different depths for the fourth model year out of our 5-yr experimental period. In noDA, the low-frequency temperature tends to have larger RMSDs at depths above the thermocline (black curves in Figs. 6a–c) than below the thermocline (black curves in Figs. 6d–e), with a stronger seasonality. For example, at the 55-m depth the temperature RMSDs range from ~0.2°C in October to ~4°C in May/June, while at the depths of 205 and 303 m the RMSDs remain little changed through the year with RMSDs of ~0.15° and ~0.02°C, respectively.
The low-frequency temperature in crtTAO (red curves in Figs. 6a–e) presents clearly smaller RMSDs than those in noDA at depths above the thermocline, suggesting the utility of the present TAO/TRITON moorings in monitoring the ENSO-related subsurface thermal evolution. At the depths of 205 m (303 m), the temperature RMSDs in crtTAO are around 0.3°C (0.1°C), larger than those in noDA, indicating the effects of too coarse vertical resolution below the thermocline for the current configuration of TAO/TRITON moorings. When the Argo observations are assimilated (Argo and crtTAO+Argo; green and blue curves in Figs. 6a–e), the low-frequency temperature is better reproduced with smaller RMSDs than in both noDA and crtTAO at all five depths, indicating an advantage of high vertical resolutions sampling with Argo floats. For low-frequency salinity over the Niño-3.4 region (Figs. 6f–j), it is evident that the TAO/TRITON sampling alone does not correct the errors in noDA, but the Argo measurements effectively correct them with smaller RMSDs in both Argo and crtTAO+Argo than in noDA through the year, demonstrating the efficacy of Argo in sampling the low-frequency subsurface salinity variability.
c. Comparison of intraseasonal component (
The performance reproducing the intraseasonal (20–90 days) temperature and salinity variability is compared in this section. The oceanic intraseasonal variability in tropical Pacific includes phenomena like tropical instability waves (TIWs; Philander et al. 1986; Willett et al. 2006) and wind-forced intraseasonal Kelvin waves (e.g., Kessler 2012). Because the TAO/TRITON mooring zonal separation is essentially one TIW wavelength, the mooring array cannot constrain TIW propagation within the DA system. In noDA, the largest RMSDs of the intraseasonal SST variability (related to TIWs) are in the central and eastern Pacific, as was the case for the low-frequency SST, although the RMSDs are meridionally more confined to the equatorial band (Fig. 7a vs Fig. 4a). The difference is consistent with the different meridional structures of ENSO- and TIW-related SST variabilities, with the former having a larger meridional scale than the latter. Compared to the amplitudes of intraseasonal variations in the Nature Run (see Figs. S1b,e and S2b,e), the RMSDs for both temperature and salinity are generally smaller in noDA, suggesting that atmosphere-driven numerical models can reproduce the intraseasonal signals with high fidelity.
In crtTAO (Fig. 7b), the RMSD spatial distribution of intraseasonal SSTs is similar to that in noDA, with the largest RMSDs in the eastern equatorial Pacific, but assimilating the TAO/TRITON synthetic observations has some improvements only near the mooring sites (green squares in Fig. 7b). Away from the TAO/TRITON mooring sites (including the far eastern coastal region), RMSDs larger than in noDA are evident in crtTAO. When the Argo synthetic observations are assimilated instead (i.e., the Argo experiment; Fig. 7c), the intraseasonal SST variability is clearly better reproduced: while RMSDs larger than 0.4°C are evident in the eastern equatorial Pacific together with few spots larger than 0.6°C in noDA, RMSDs are smaller than 0.4°C over most of the tropical Pacific in Argo. A similar improvement is shown in the crtTAO+Argo experiment. For the intraseasonal SSS variability, however, no matter whether TAO/TRITON or Argo observations are assimilated, comparable RMSDs are achieved with respect to noDA (Figs. 7f–h vs Fig. 7e).
For the depth–longitude section along the equator, as with the low-frequency variability the largest RMSDs of intraseasonal temperature variability (Figs. 8a–d) are also present along the equatorial thermocline, which is expected from uncertainties in reproducing the wind-forced intraseasonal Kelvin waves with maximum signals along the equatorial thermocline. In noDA, the RMSDs are larger than 0.4°C along the thermocline, with two locations with amplitude larger than 0.6°C in the western and eastern Pacific. As the TAO/TRITON synthetic observations are assimilated (e.g., crtTAO; Fig. 8b), improvements of intraseasonal temperature variability (with smaller RMSDs) are near the mooring sites (green squares in Fig. 8b). Between the mooring sites, however, columns with large RMSDs (>0.4°C) extend from surface to depths below the thermocline. In the Argo run (Fig. 8c), on the other hand, improvements of intraseasonal temperature variability relative to noDA are evident over the whole equatorial Pacific, with RMSDs smaller than 0.4°C; small regions with >0.4°C RMSDs in Argo are at the western and eastern ends of the equatorial thermocline with RMSDs larger than 0.6°C in noDA. A similar improvement is also achieved in crtTAO+Argo (Fig. 8d).
As in Fig. 5, but for the intraseasonal component.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
It should be noted that the result about limited improvements of crtTAO in reconstructing the subsurface intraseasonal temperature variability relative to noDA (Fig. 8b vs Fig. 8a) does not mean that TAO/TRITON cannot be used to monitor the subsurface intraseasonal thermal evolutions. In fact, because ocean models can realistically capture the wind-forced intraseasonal Kelvin waves through prescribed atmospheric forcing, additional improvements are hard to achieve from the present TAO/TRITON measurements. However, the TAO/TRITON measurements alone are still able to capture the subsurface intraseasonal signal, and they are being used to routinely monitor the intraseasonal Kelvin waves (e.g., https://www.pmel.noaa.gov/tao/drupal/disdel/) that are usually related to intraseasonal westerly wind events.
For the intraseasonal subsurface salinity variability along the equator, the RMSD distribution in noDA (Fig. 8e) also resembles that for the low-frequency component (Fig. 5e), but with a smaller amplitude. The resemblance might suggest that precipitation plays an important role in the SSS low-frequency and intraseasonal variations, and thus the RMSD distributions. The largest RMSDs (>0.12 psu) are present in the upper ocean of the western warm pool region, and relatively large RMSDs (~0.05 psu) are also evident in a layer below the equatorial thermocline sloping from the deep west to the shallower east with larger RMSDs at its two ends. These errors are corrected only near the mooring sites in crtTAO (Fig. 8f), but over the entire equatorial Pacific in Argo (Fig. 8g) and crtTAO+Argo (Fig. 8h). This comparison suggests that the current Argo coverage with salinity measurements could efficiently improve the intraseasonal salinity variability as is the case for temperature.
d. Comparison of high-frequency component (
This section compares how the four main experiments capture the high-frequency (<20 days) variability, such as short-lived eddies (e.g., Chen and Han 2019). We note that the model resolution (0.5° with an equatorial refinement meridionally; see section 2a) in the Nature Run is clearly too coarse to explicitly represent mesoscale eddies (especially in high-latitude oceans). The resolution, however, is high enough to resolve baroclinic deformation radius within the equatorial ocean (Hallberg 2013) and should be able to represent some large eddies.
For SST in noDA (Fig. 9a) and three assimilation runs (Figs. 9b–d), the RMSD distributions are dominated by large errors in the eastern equatorial Pacific, similar to those for the low-frequency (Figs. 4a–d) and intraseasonal (Figs. 7a–d) components but with an even smaller meridional scale. The RMSD distributions of SSS (Figs. 9e–h) feature large errors along two off-equatorial bands, which are also discernible in those for the intraseasonal component (Figs. 7e–h). The off-equatorial structure symmetric to the equator is reminiscent of the “double-ITCZ” problem with the RAS convection scheme (see Fig. 4c in Zhu et al. 2017b), implying a response uncertainty of models (MOM5 for the Nature Run versus MOM6 for the noDA and DA experiments) to a given freshwater flux forcing. Compared to the amplitudes of high-frequency variations in the Nature Run (see Figs. S1c,f and S2c,f), the RMSDs for both temperature and salinity are even larger in noDA, suggesting that the high-frequency variations are mainly driven internally through various instabilities in the ocean with less contribution from atmospheric forcing than its low-frequency or intraseasonal counterparts.
As in Fig. 4, but for the high-frequency component.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
Compared with noDA (Figs. 9a,e), it is evident that the three DA experiments (Figs. 9b–d and 11f–h) demonstrate no superiority in reproducing the high-frequency SST and SSS variability of the Nature Run; particularly for SSS, assimilations of Argo measurements clearly deteriorate the fidelity of its high-frequency variability (Figs. 9g,h vs Fig. 9e). It should be noted that our way of generating synthetic Argo measurements could underestimate (overestimate) the value of Argo in reproducing high-frequency variability over regions with higher (lower) density of real Argo floats. The overall effect over the entire tropical Pacific and multiple years is assumed to be marginal, which, however, needs to be tested by further sensitivity experiments.
The above degradation is also seen at the subsurface, such as at the depth–longitude section along the equator (Fig. 10). The RMSD distributions of high-frequency components are also like those for the low-frequency (Fig. 5) and intraseasonal (Fig. 8) components (e.g., with large temperature errors along the thermocline and large salinity errors in the upper ocean of the western warm pool region). The degradation of high-frequency variability relative to noDA is more evident when Argo measurements are assimilated. For example, while the RMSDs for high-frequency temperature are generally smaller than 0.6°C over most of the equatorial section in noDA (Fig. 10a), they are as large as 1.6°C over a sizable region/depth near 150°E in both Argo (Fig. 10c) and crtTAO+Argo (Fig. 10d); at the same place, large salinity RMSDs also emerge in the two experiments (Figs. 10g,h), which are absent in noDA (Fig. 10e).
As in Fig. 5, but for the high-frequency component.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
In summary, the above diagnostics suggest that it is challenging for both the current TAO/TRITON and Argo observing networks to be used for studies about high-frequency (<20 days) variability (except for mooring sites). For TAO/TRITON, even though it samples the ocean properties (and surface meteorological conditions) at frequencies as high as minutes, the coarse array of moorings may be unable to resolve high-frequency phenomena between sites by DA alone. In contrast, while the current Argo network has a better spatial sampling distribution (3° × 3°; Roemmich and Owens 2000), its temporal resolution of 10 days seems not good enough for resolving high-frequency variabilities (Gasparin et al. 2015); consequently, 10-day apart Argo measurements may lead to errors in the temporal representation of high-frequency variability.
4. Evaluation of ocean currents
This section presents a brief evaluation of ocean currents in four main experiments. Different from temperature and salinity, which are direct assimilation variables in the JEDI-based ODA system (see section 2c), ocean currents are not assimilation variables; instead, they are modulated purely through model adjustments. Thus, an evaluation of ocean currents provides a more stringent test on the quality of assimilations.
Figure 11 compares the subsurface structure of the mean zonal currents at the equator in noDA and three DA runs. Compared to the Nature Run (Fig. 11a), the core of the annual-mean Equatorial Undercurrent shifts slightly upward in noDA (Fig. 11b), which is consistent with a shallower thermocline depth in noDA than the Nature Run (Fig. 3a). Consequently, there is a zonal band of eastward current biases at the depth of 50–100 m (Fig. 11f). Meanwhile, the westward South Equatorial Current (SEC) at the surface is weaker in noDA than in the Nature Run (Fig. 11b vs Fig. 11a) which is more evident in the spatial distributions of surface zonal currents (figures not shown). When the TAO/TRITON synthetic temperature and salinity observations are assimilated (i.e., the crtTAO run), the eastward current biases at the 50–100-m depth are partially corrected (Fig. 11g vs Fig. 11f), but the surface SEC becomes even weaker (Fig. 11c vs Fig. 11b), and the Equatorial Undercurrent extends too deep in the eastern basin (Figs. 11c,g). Of particular concern is the Equatorial Undercurrent acceleration in between the mooring sites as this would induce subsurface divergence and reduced upwelling, perhaps resulting in the warm bias observed on the equator in the crtTAO run (Fig. 2b). When the Argo synthetic observations are assimilated (i.e., the Argo run), the simulated Equatorial Undercurrent and surface SEC are much closer to those in the Nature Run with biases smaller than 0.2 m s−1 over almost the entire equatorial Pacific (Fig. 11h). Significant current improvements relative to noDA is also seen in crtTAO+Argo (Fig. 11f vs Fig. 11h), but the improvement seems slightly smaller than that in Argo and their signs of biases are also generally opposite (Fig. 11i vs Fig. 11h).
Long-term mean zonal currents (m s−1) along the equator in (a) the Nature Run, (b) noDA, (c) crtTAO, (d) Argo, and (e) crtTAO+Argo. (f)–(i) The differences of (b)–(e) relative to (a), respectively. In (a), the 20°C isothermal is also included as a solid (dashed) curve for the Nature Run (noDA). The black squares in (c), (e), (g), and (i) and the small gray dots in (d), (e), (h), and (i) indicate where TAO/TRITON and Argo measurements are available, respectively.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
The low-frequency and intraseasonal components of zonal currents are evaluated in Fig. 12, which calculates the RMSDs relative to the Nature Run. In noDA, zonal currents feature large RMSDs in the upper ocean shallower than 150 m (largely within the mixed layer) at both time scales, with RMSDs greater than 0.21 m s−1 (Fig. 12a) and 0.12 m s−1 (Fig. 12e) over large regions, respectively. Compared to noDA, the crtTAO run did not reduce the large RMSDs; instead, greater RMSDs than in noDA are present in the eastern equatorial Pacific (Figs. 12b,f). When the Argo synthetic observations are assimilated (i.e., the Argo run), the zonal currents are clearly improved relative to noDA at both time scales. The RMSDs are smaller than 0.21 m s−1 over almost entire equatorial Pacific for the low-frequency zonal currents (Fig. 12c). For the intraseasonal zonal currents, RMSDs greater than 0.12 m s−1 are also only seen over a small region in the western Pacific (Fig. 12g). When the TAO/TRITON and Argo synthetic observations are jointly assimilated (i.e., the crtTAO+Argo), the RMSDs are slightly larger than in Argo, but smaller than in either crtTAO or noDA.
Root-mean-square differences (RMSD) of (left) low-frequency and (right) intraseasonal zonal currents (m s−1) along the equator with respect to the Nature Run in (a),(e) noDA, (b),(f) crtTAO, (c),(g) Argo, and (d),(h) crtTAO+Argo. The green squares in (b), (d), (f), and (h) indicate where temperature and salinity are measured by TAO/TRITION, respectively. The small gray dots in (c), (d), (g), and (h) indicate where Argo measurements are available.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
An evaluation of zonal currents at the surface (figures not shown) also suggests a similar result as for the subsurface zonal currents. That is, assimilating Argo measurements is superior to TAO/TRITON in constraining (indirectly through model adjustments) ocean currents in terms of their long-term states and low-frequency/intraseasonal variabilities. The performance of Argo, however, is slightly deteriorated by additionally assimilating TAO/TRITON observations.
There might be two factors contributing to the above difference among three DA experiments regarding ocean currents. First, the TAO/TRITON measurements might not be dense enough to effectively constrain the ocean currents. The issue not only comes from the smaller number of TAO/TRITON moorings than Argo floats, but also from a lack of salinity measurements at most moorings. The latter seems to be evidenced by the structures of RMSDs (Figs. 12b,d,f,h) that are not always smaller near the TAO/TRITON mooring sites than in between, a difference from the RMSD structures of temperature and salinity (see section 3). Second, the assimilation once a day (see section 2c) might introduce a nonnegligible shock in the experiments of crtTAO and crtTAO+Argo, which could contribute to errors in ocean currents. In contrast, data assimilation in Argo is conducted every 10 days, giving the ocean model a longer time to adjust. Consequently, the contribution of the above shock to ocean currents becomes much smaller in Argo. The effect of the shock might have contributed to the difference between Argo and crtTAO+Argo. The relative roles of the two factors, however, cannot be quantified without additional well-designed experiments.
At the high-frequency time scale, ocean currents were not improved in all three DA experiments (figures not shown), which could be expected from a similar lack of temperature and salinity improvements at the time scale (see section 3d).
5. Sensitivity results about DA parameters
In this section, two additional experiments are analyzed to evaluate the sensitivity of the above results to two DA parameters: correlation scale and observational errors. Regarding correlation scale, in all the above DA experiments we apply a default scale of 700 km. However, when evaluating the crtTAO performance in terms of both mean temperature biases (Figs. 2b and 4b) and RMSDs of low-frequency (Figs. 4b and 6b) or intraseasonal (Figs. 7b and 9b) temperature variabilities, errors are found to be significantly smaller at mooring sites than at gaps between them. Such error structures raise a possibility that the correlation scale of 700 km is too small for assimilating TAO/TRITON observations. To investigate the possibility, we conduct a one-month sensitivity experiment (i.e., crtTAO_RScl) by doubling the correlation scale (i.e., 1400 km) with other settings configured as the same as in crtTAO. It should be noted that even though a one-month integration may be inadequate to resolve uncertainties in fine scale error structures, the large-scale error structure is well captured by the one-month integration, as suggested by the comparison between the one-month (Fig. 13a) and 5-yr (Fig. 3a) diagnostics with crtTAO. When the correlation scale is doubled to 1400 km, the above large-scale error structure remains similar (Fig. 13b vs Fig. 13a). In particular, they both demonstrate warm temperature biases near the ocean surface and cold biases in the deep ocean sloping from the west to the east. Also, they both feature large temperature errors in mooring gaps. This comparison suggests that the choice of correlation length scale of 700 km is long enough to spread out the observational information gathered at TAO/TRITON sites, and the large errors between mooring sites are simply due to the sparse zonal deployments of the moorings.
Mean difference of temperature (°C) along the equator with respect to Nature Run in (a) crtTAO, (b) crtTAO_RScl, and (c) crtTAO_obserr. The black squares indicate where TAO/TRITON measurements are available, respectively.
Citation: Journal of Climate 34, 16; 10.1175/JCLI-D-20-0951.1
Another parameter tested is magnitude of observational errors. It is important but challenging for OSSEs to construct synthetic observations with realistic error properties (Fujii et al. 2019). In this study, relatively small observational errors (0.1°C for temperature and 0.03 psu for salinity) are applied for most experiments by considering several factors described in section 2b. To evaluate the sensitivity of above results to our choice of observational errors, we conduct another sensitivity experiment (i.e., crtTAO_obserr) by simply amplifying observational errors by a factor of 5 (i.e., 0.5°C for temperature and 0.15 psu for salinity). By comparing crtTAO_obserr with crtTAO (Fig. 13c vs Fig. 13a), it is evident that the above large-scale error structure remains similar, indicating the robustness of the above findings in section 3 with little sensitivity to the magnitude of observational errors.
6. Conclusions and discussion
This work was part of efforts in support of the TPOS 2020 project (Cravatte et al. 2016; Smith et al. 2019) by focusing on assessing the impact of current in situ ocean observing systems on the quality of ocean analysis in the tropical Pacific. For this purpose, a series of OSSEs was conducted to explore the relative roles of TAO/TRITON versus Argo in the TPOS, using the JEDI-based ODA system. Specifically, we performed four main experiments including a free run with no observations assimilated (noDA) and three assimilation runs in which the synthetic ocean temperature and salinity observations from TAO/TRITON moorings (crtTAO), Argo floats (Argo), and both TAO/TRITON and Argo (crtTAO+Argo) were assimilated. These experiments were validated against the Nature Run, which has high fidelity in reproducing important climate modes in the tropical Pacific (e.g., ENSO, MJO, barrier layers), and compared among each other in terms of their long-term mean states and variabilities. The evaluations of variabilities went beyond the traditional focus of OSE/OSSE studies about the tropical Pacific, namely the ENSO time scale variability) and extended to multiple time scales by additionally examining the variabilities at the intraseasonal (20–90 days) and high-frequency (<20 days) time scales.
For mean states and low-frequency ocean variations, both TAO/TRITON and Argo effectively correct the simulation biases in noDA, with Argo achieving more significant improvements. Specifically, both the subsurface cold bias and the large low-frequency temperature error along the thermocline in noDA were reduced by assimilating the TAO/TRITON or Argo synthetic observations. For salinity, the subsurface freshening bias was effectively removed by Argo, and was also partially corrected by TAO/TRITON in the upper ocean along the equator where five moorings observe the subsurface salinity evolutions. The findings suggest that the present TMA and Argo networks are generally suitable for ENSO applications, which has been the original motivation of the TPOS.
On the intraseasonal time scale, TAO/TRITON demonstrates a comparable skill to noDA in observing the subsurface thermal evolution. Considering that ocean models are generally good at capturing the response to wind-forced intraseasonal Kelvin wave activities in the equatorial subsurface variability, the comparable skill suggests that the TAO/TRITON array alone can be used to monitor the subsurface intraseasonal thermal evolutions in the tropical Pacific. For salinity, however, the current TAO/TRITON presents clear limitations because of its limited salinity observations. Compared to noDA and TAO/TRITON, Argo presents significant basinwide improvements in reproducing both temperature and salinity intraseasonal variability, suggesting that the current Argo observations are overall a good observational platform for intraseasonal studies (e.g., Gasparin et al. 2015). The improvements with Argo might be due to its homogenous distribution (Sivareddy et al. 2017).
For the high-frequency variability, however, both TAO/TRITON and Argo present evident shortcoming in observing them, even though improvements over noDA were present at the TAO/TRITON mooring sites. Our explanations for this feature are as follows. For the TAO/TRITON array, the moorings were deployed spatially (particularly zonally) so coarsely that the high-frequency observed information collected at mooring sites, which usually have small spatial scales, cannot be effectively mapped to the whole basin for high-frequency phenomena, which have usually small spatial scales. For the Argo network, on the other hand, its temporal resolution of 10 days is not good enough to capture the high-frequency phenomena. However, we note that such observations can be a valuable for the validation of models at point locations as discussed below. It should also be noted that our way of generating synthetic Argo measurements could underestimate (overestimate) the value of Argo in reproducing high-frequency variability over regions with higher (lower) density of Argo floats in reality.
We also evaluated ocean currents in above experiments. The analysis suggests that assimilating Argo measurements is superior to TAO/TRITON in constraining (indirectly through model adjustments) ocean currents in terms of their long-term states and low-frequency/intraseasonal variabilities. Equatorial deceleration at the mooring sites and acceleration in between the sites appears to lead to subsurface divergence and reduced upwelling and could account for the equatorial warm bias. The experiments conducted in this study, however, cannot address how much the superiority of the Argo DA is due to a higher observing coverage of Argo or due to a less initial shock in the Argo assimilation run.
Although our OSSE results suggest that Argo seems to play a more important role in the TPOS, many advantages of TAO/TRITON cannot be addressed in this study. For example, in addition to subsurface ocean states, TAO/TRITON also samples surface meteorological state at the same locations. The collocated ocean–atmosphere sampling is important in diagnosing and resolving the interaction of two boundary layers that are usually poorly modeled. The value of the meteorological data for DA was not addressed here. Second, the sampling frequency of TAO/TRITON is as high as 1 min for some flux fields (e.g., rainfall at the equatorial TAO flux moorings) and 10 min for oceanic fields. The high-frequency sampling aids interpretation of coarser measurements. Third, TAO/TRITON has collected longer climate records (at fixed spatial locations) than Argo and satellite. To maintain the continuity of the climate record, it will be important to maintain them as they can provide a long and consistent climate record to explore the lower-frequency climate variability (e.g., the decadal modulation of ENSO properties). In addition, the TAO/TRITON time series are also used to calibrate satellite winds, SSTs, etc., and their near-equatorial current measurements are vital for climate studies as well. All these advantages indicate the necessity of maintaining the TMA array in the tropical Pacific.
In addition, our DA experiments did not assimilate any other types of observations (e.g., altimetry, satellite SST, drifters). The choice has the advantage of highlighting the effects of the two important types of in situ observations on their own in reproducing the tropical Pacific variability at various time scales. The other types of measurements are also suggested to play important roles in oceanography for different aspects, such as altimetry in representing subsurface ocean (e.g., Verdy et al. 2014; Gasparin et al. 2015). If those observations are assimilated in our experiments together with TAO/TRITON and Argo, our capability to reproduce tropical Pacific variability is expected to be higher and the relative contributions from TAO/TRITON and Argo will be different, and perhaps less. The JEDI-based ODA system applied in this study is being actively developed with the capability of assimilating all the types of observations. We plan to conduct data denial experiments (i.e., observing system experiments) with the system to evaluate the respective roles of all the observational platforms in the TPOS in future.
Also, the vertical correlation scale in all DA experiments is updated every DA cycle based on the model layer thickness (i.e., adaptive to vertical stratifications). Since we did not see as much ground for concern from it as those from horizontal correlation scale when exploring results from our DA experiments, we did not conduct such sensitivity experiments in this study. But in the future such an experiment will be in our list.
As a further effort in support of the TPOS 2020 Project, we will repeat the above OSSEs to evaluate the proposed TPOS configuration recommended by the project, and toward that end, the current set of experiments describe development of a capability that will be essential. Additionally, the present set of experiments do provide as assessment of TAO and Argo measurements in resolving oceanic variability on different temporal scales.
Even though the proposed TPOS configuration has not yet been finalized, some changes in TMA can be expected based upon the existing TPOS 2020 report recommendations and discussions (e.g., Cravatte et al. 2016). For example, it is recommended that the meridional resolution of temperature and salinity in the equatorial zone be enhanced through a mix of additional moorings near the equator and targeted enhancement of Argo profiles in the equatorial zone. An enhancement in the vertical temperature and salinity resolution from the TMA is also recommended via additional sensors in the upper ocean on moorings from the top of the thermocline to the surface, and by having Argo measurements at 1-dbar resolution from 100 dbar to the surface. The density of Argo temperature and salinity profile observations is also recommended to be doubled throughout the tropical region (10°S–10°N), targeting a density of one profile per 3° × 3° square every 5 days or, equivalently, one profile per 2.1° × 2.1° square every 10 days. All the recommendations are expected to enhance our capability of observing the tropical Pacific variability on different time scales, particularly at the high-frequency time band (e.g., <20 days). Our proposed OSSEs will provide a more quantitative estimate of these influences and give the community an objective assessment about the strength and weakness of the proposed design for the TPOS.
Acknowledgments
The authors are grateful to Drs. William Kessler, Susan Wijffels, Sophie Cravatte, Shuyi S. Chen, Aneesh Subramanian, Ariane Verdy, Kelvin Richards, and Yan Xue for their comments. We also thank the anonymous reviewers and Drs. Zeng-Zhen Hu and Li Ren for reviewing this paper. Funding for this study is provided by NOAA’s Climate Program Office (CPO) through the Climate Variability and Predictability (CVP) program and Office of Oceanic and Atmospheric Research (OAR) through the Global Ocean Monitoring and Observation (GOMO) program. JZ is also partially supported by a NASA Ocean Salinity Science Team grant (NNX17AK09G). This is PMEL paper 5188.
REFERENCES
Adcroft, A., and R. Hallberg, 2006: On methods for solving the oceanic equations of motion in generalized vertical coordinates. Ocean Modell., 11, 224–233, https://doi.org/10.1016/j.ocemod.2004.12.007.
Ando, K., T. Matsumoto, T. Nagahama, I. Ueki, Y. Takatsuki, and Y. Kuroda, 2005: Drift characteristics of a moored conductivity–temperature–depth sensor and correction salinity data. J. Atmos. Oceanic Technol., 22, 282–291, https://doi.org/10.1175/JTECH1704.1.
Balmaseda, M. A., and D. Anderson, 2009: Impact of initialization strategies and observations on seasonal forecast skill. Geophys. Res. Lett., 36, L01701, https://doi.org/10.1029/2008GL035561.
Balmaseda, M. A., D. Anderson, and A. Vidard, 2007: Impact of Argo on analyses of the global ocean. Geophys. Res. Lett., 34, L16605, https://doi.org/10.1029/2007GL030452.
Beljaars, A. C. M., 1995: The parameterization of surface fluxes in large-scale models under free convection. Quart. J. Roy. Meteor. Soc., 121, 255–270, https://doi.org/10.1002/qj.49712152203.
Chelton, D. B., R. A. de Szoeke, M. G. Schlax, K. El Naggar, and N. Siwertz, 1998: Geographic variability of the first baroclinic Rossby radius of deformation. J. Phys. Oceanogr., 28, 433–460, https://doi.org/10.1175/1520-0485(1998)028<0433:GVOTFB>2.0.CO;2.
Chen, G., and G. Han, 2019: Contrasting short-lived with long-lived mesoscale eddies in the global ocean. J. Geophys. Res. Oceans, 124, 3149–3167, https://doi.org/10.1029/2019JC014983.
Cravatte, S., and Coauthors, 2016: First Report of TPOS 2020. GOOS-215, TPOS 2020, 200 pp, https://TPOS2020.org/first-report/.
Eisenman, I., L. S. Yu, and E. Tziperman, 2005: Westerly wind bursts: ENSO’s tail rather than the dog? J. Climate, 18, 5224–5238, https://doi.org/10.1175/JCLI3588.1.
Fujii, Y., K. Ogawa, G. B. Brassington, K. Ando, T. Yasuda, and T. Kuragano, 2015a: Evaluating the impacts of the Tropical Pacific Observing System on the ocean analysis fields in the Global Ocean Data Assimilation System for operational seasonal forecasts in JMA. J. Oper. Oceanogr., 8, 25–39, https://doi.org/10.1080/1755876X.2015.1014640.
Fujii, Y., and Coauthors, 2015b: Evaluation of the Tropical Pacific Observing System from the ocean data assimilation perspective. Quart. J. Roy. Meteor. Soc., 141, 2481–2496, https://doi.org/10.1002/qj.2579.
Fujii, Y., and Coauthors, 2019: Observing system evaluation based on ocean data assimilation and prediction systems: On-going challenges and a future vision for designing and supporting ocean observational networks. Front. Mar. Sci., 6, 417, https://doi.org/10.3389/fmars.2019.00417.
Gasparin, F., D. Roemmich, J. Gilson, and B. Cornuelle, 2015: Assessment of the upper-ocean observing system in the equatorial Pacific: The role of Argo in resolving intraseasonal to interannual variability. J. Atmos. Oceanic Technol., 32, 1668–1688, https://doi.org/10.1175/JTECH-D-14-00218.1.
Gasparin, F., M. Hamon, E. Rémy, and P.-Y. Le Traon, 2020: How deep Argo will improve the deep ocean in an ocean reanalysis. J. Climate, 33, 77–94, https://doi.org/10.1175/JCLI-D-19-0208.1.
Ge, X., W. Wang, A. Kumar, and Y. Zhang, 2017: Importance of the vertical resolution in simulating SST diurnal and intraseasonal variability in an oceanic general circulation model. J. Climate, 30, 3963–3978, https://doi.org/10.1175/JCLI-D-16-0689.1.
Hallberg, R. W., 2013: Using a resolution function to regulate parameterizations of oceanic mesoscale eddy effects. Ocean Modell., 72, 92–103, https://doi.org/10.1016/j.ocemod.2013.08.007.
Halliwell, G. R., Jr., A. Srinivasan, V. Kourafalou, H. Yang, D. Willey, M. Le Hénaff, and R. Atlas, 2014: Rigorous evaluation of a fraternal twin ocean OSSE system for the open Gulf of Mexico. J. Atmos. Oceanic Technol., 31, 105–130, https://doi.org/10.1175/JTECH-D-13-00011.1.
Holdaway, D., G. Vernieres, M. Wlasak, and S. King, 2020: Status of model interfacing in the Joint Effort for Data assimilation Integration (JEDI). JCSDA Quart., No. 66 (winter 2020), Joint Center for Satellite Data Assimilation Office, College Park, MD, 15–24, https://doi.org/10.25923/rb19-0q26.
Kawai, Y., and A. Wada, 2007: Diurnal sea surface temperature variation and its impact on the atmosphere and ocean: A review. J. Oceanogr., 63, 721–744, https://doi.org/10.1007/s10872-007-0063-0.
Kessler, W. S., 2012: The oceans. Intraseasonal Variability in the Atmosphere–Ocean Climate System, 2nd ed., W. K. M. Lau and D. E. Waliser, Eds., Springer, 199–246.
Lau, W. K.-M., and D. E. Waliser, 2012: Intraseasonal Variability in the Atmosphere–Ocean Climate System. Springer, 614 pp.
McPhaden, M. J., and Coauthors, 1998: The Tropical Ocean–Global Atmosphere (TOGA) observing system: A decade of progress. J. Geophys. Res. Oceans, 103, 14 169–14 240, https://doi.org/10.1029/97JC02906.
McPhaden, M. J., A. J. Busalacchi, and D. L. T. Anderson, 2010: A TOGA retrospective. Oceanography, 23, 86–103, https://doi.org/10.5670/oceanog.2010.26.
Moorthi, S., and M. J. Suarez, 1992: Relaxed Arakawa–Schubert: A parameterization of moist convection for general circulation models. Mon. Wea. Rev., 120, 978–1002, https://doi.org/10.1175/1520-0493(1992)120<0978:RASAPO>2.0.CO;2.
Philander, S. G. H., W. J. Hurlin, and R. C. Pacanowski, 1986: Properties of long equatorial waves in models of the seasonal cycle in the tropical Atlantic and Pacific Oceans. J. Geophys. Res., 91, 14 207–14211, https://doi.org/10.1029/JC091iC12p14207.
Riser, S., and Coauthors, 2016: Fifteen years of ocean observations with the global Argo array. Nat. Climate Change, 6, 145–153, https://doi.org/10.1038/nclimate2872.
Roemmich, D., and W. B. Owens, 2000: The Argo project: Global ocean observations for understanding and prediction of climate variability. Oceanography, 13, 45–50, https://doi.org/10.5670/oceanog.2000.33.
Roemmich, D., and Coauthors, 2019: On the future of Argo: A global, full-depth, multi-disciplinary array. Front. Mar. Sci., 6, 439, https://doi.org/10.3389/fmars.2019.00439.
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.
Saha, S., and Coauthors, 2014: The NCEP Climate Forecast System version 2. J. Climate, 27, 2185–2208, https://doi.org/10.1175/JCLI-D-12-00823.1.
Sivareddy, S., A. Paul, T. Sluka, M. Ravichandran, and E. Kalnay, 2017: The pre-Argo ocean reanalyses may be seriously affected by the spatial coverage of moored buoys. Sci. Rep., 7, 46 685, https://doi.org/10.1038/srep46685.
Smith, N., and Coauthors, 2019: Tropical Pacific Observing System. Front. Mar. Sci., 6, 31, https://doi.org/10.3389/fmars.2019.00031.
Tollefson, J., 2014: El Niño tests forecasters. Nature, 508, 20–21, https://doi.org/10.1038/508020a.
Trémolet, Y., and T. Auligné, 2020: The Joint Effort for Data Assimilation Integration (JEDI). JCSDA Quart., No. 66 (winter 2020), Joint Center for Satellite Data Assimilation Office, College Park, MD, 1–5, https://doi.org/10.25923/rb19-0q26.
Trenberth, K. E., G. W. Branstator, D. Karoly, A. Kumar, N.-C. Lau, and C. Ropelewski, 1998: Progress during TOGA in understanding and modeling global teleconnections associated with tropical sea surface temperatures. J. Geophys. Res., 103, 14 291–14 324, https://doi.org/10.1029/97JC01444.
Verdy, A., M. R. Mazloff, B. D. Cornuelle, and S. Y. Kim, 2014: Wind-driven sea level variability on the California coast: An adjoint sensitivity analysis. J. Phys. Oceanogr., 44, 297–318, https://doi.org/10.1175/JPO-D-13-018.1.
Vidard, A., D. L. T. Anderson, and M. Balmaseda, 2007: Impact of ocean observing systems on ocean analysis and seasonal forecasts. Mon. Wea. Rev., 135, 409–429, https://doi.org/10.1175/MWR3310.1.
Wang, C., C. Deser, J.-Y. Yu, P. DiNezio, and A. Clement, 2016: El Niño and Southern Oscillation (ENSO): A review. Coral reefs of the eastern tropical Pacific: Persistence and loss in a dynamic environment. P. W. Glynn, D. P. Mazello, and I. C. Enoch, Eds., Coral Reefs of the World, Vol. 8, Springer, 85–106.
Weaver, A. T., C. Deltel, E. Machu, S. Ricci, and N. Daget, 2005: A multivariate balance operator for variational ocean data assimilation. Quart. J. Roy. Meteor. Soc., 131, 3605–3625, https://doi.org/10.1256/qj.05.119.
Willett, C. S., R. R. Leben, and M. F. Lavín, 2006: Eddies and tropical instability waves in the eastern tropical Pacific: A review. Prog. Oceanogr., 69, 218–238, https://doi.org/10.1016/j.pocean.2006.03.010.
Xue, Y., C. Wen, X. Yang, D. Behringer, A. Kumar, G. Vecchi, A. Rosati, and R. Gudgel, 2017: Evaluation of tropical Pacific observing systems using NCEP and GFDL ocean data assimilation systems. Climate Dyn., 49, 843–868, https://doi.org/10.1007/s00382-015-2743-6.
Yan, C., J. Zhu, and G. Zhou, 2007: Impacts of XBT, TAO, altimetry and ARGO observations on the tropical Pacific Ocean data assimilation. Adv. Atmos. Sci., 24, 383–398, https://doi.org/10.1007/s00376-007-0383-4.
Zhang, Y., M.-P. Hung, W. Wang, and A. Kumar, 2019: Role of SST feedback in the prediction of the boreal summer monsoon intraseasonal oscillation. Climate Dyn., 53, 3861–3875, https://doi.org/10.1007/s00382-019-04753-w.
Zhu, J., and A. Kumar, 2019: Role of sea surface salinity feedback in MJO predictability: A study with CFSv2. J. Climate, 32, 5745–5759, https://doi.org/10.1175/JCLI-D-18-0755.1.
Zhu, J., W. Wang, and A. Kumar, 2017a: Simulations of MJO propagation across the Maritime Continent: Impacts of SST feedback. J. Climate, 30, 1689–1704, https://doi.org/10.1175/JCLI-D-16-0367.1.
Zhu, J., A. Kumar, W. Wang, Z.-Z. Hu, B. Huang, and M. A. Balmaseda, 2017b: Importance of convective parameterization in ENSO predictions. Geophys. Res. Lett., 44, 6334–6342, https://doi.org/10.1002/2017GL073669.
Zhu, J., A. Kumar and W. Wang, 2020: Intraseasonal surface salinity variability and the MJO in a climate model. Geophys. Res. Lett., 47, e2020GL088997, https://doi.org/10.1029/2020GL088997.
