ENSO-Induced Teleconnection: Process-Oriented Diagnostics to Assess Rossby Wave Sources and Ambient Flow Properties in Climate Models

a. Background

During El Niño–Southern Oscillation (ENSO) winters an equivalent barotropic stationary wave train pattern dominates over Pacific and North American region (Horel and Wallace 1981) and this pattern can be interpreted as an external Rossby wave forced by equatorial Pacific heating (Hoskins and Karoly 1981; Held and Kang 1987; Trenberth et al. 1998). Rossby wave generation and propagation provides the basis for many theories of how the tropics influence midlatitudes (Hoskins et al. 1977; Held and Kang 1987; Branstator 1983). A process-oriented diagnostic (POD; Maloney et al. 2019; Annamalai 2020) package that addresses the chain of processes, that is, intermediate between equatorial Pacific heating and the circulation pattern over the Pacific and North American regions that are usually not addressed in model evaluations (Deser et al. 2016), is developed. The POD will help to address this critical model evaluation gap, by quantifying the roles of anomalous upper-tropospheric divergent wind patterns in the generation of stationary Rossby waves, and the zonally and meridionally varying ambient flow properties that determine the horizontal propagation of these waves (e.g., Branstator 1985; Ting and Sardeshmukh 1993; Di Carlo et al. 2022). As part of model evaluation, we apply the POD to a subset of climate models that participated in phases 5 and 6 of the Atmospheric General Circulation Model Intercomparison Project (AMIP). Besides assessing the representation of relevant processes in individual models, improvements or degradations in models that participated in both AMIP5 to AMIP6 versions are identified, and their implications to ENSO-induced teleconnections are discussed.

During El Niño winters, positive sea surface temperature (SST) anomalies along the central-eastern equatorial Pacific (e.g., Rasmusson and Carpenter 1982) favor local enhancement of moist static energy, promoting deep convection and increasing the release of latent heating throughout the troposphere (see Wallace et al. 1998 for a review). The resultant anomalous vertical velocity and upper-level divergence alters the generation of the horizontal component of atmospheric vorticity (Hoskins et al. 1977; Hoskins and Karoly 1981), giving rise to anomalous Rossby wave sources (RWS′; Sardeshmukh and Hoskins 1988). These RWS′ excite normal modes of the zonally varying ambient flow (Simmons et al. 1983). Kang and Held (1986) and Sardeshmukh and Hoskins (1988) emphasized that at upper-tropospheric levels, the contribution by advection of the climatological meridional gradient in absolute vorticity by the anomalous divergent winds can also lead to a significant RWS′ in westerly wind regions, in which Rossby wave motion is possible. Trenberth et al. (1998) provide a comprehensive review on all aspects of tropical–extratropical interactions. Atmospheric general circulation models (AGCMs) forced with ENSO-related SST anomalies exhibit various levels of success in representing this circulation pattern (Shukla and Wallace 1983; Lau and Nath 1994; Hoerling et al. 1997; Kumar and Hoerling 1997; Hoerling and Kumar 2002; Shukla et al. 2000; Barsugli and Sardeshmukh 2002; Peng and Kumar 2005; Annamalai et al. 2007; Peng et al. 2014; Deser et al. 2016; Li et al. 2020, among others).

During boreal winters, a prominent midlatitude circulation over the Pacific–North American region is the so-called Pacific–North American (PNA) pattern defined by Wallace and Gutzler (1981). One established interpretation is that PNA is an internal mode of the midlatitude circulation (Lau 1981) that is amplified by ENSO (Horel and Wallace 1981; Kumar and Hoerling 1995; Lau and Nath 1994; Hoerling et al. 1997; Hoerling and Kumar 2002; Peng and Kumar 2005; Li et al. 2020), that is, ENSO can create preferred teleconnection response patterns, such as the PNA (Trenberth et al. 1998). An alternative interpretation is based on differences in the position and structures of the upper-level height anomalies, particularly during El Niño (e.g., Lopez and Kirtman 2019). Based on this paradigm, Straus and Shukla (2002) suggest that ENSO does not force PNA. Intriguingly, the height anomalies bear close similarity to traditional PNA during La Niña winters (Fig. 1 in Lopez and Kirtman 2019). In the following subsection, we clarify our approach with reference to the above interpretations.

b. Present study

Guided by linear and nonlinear model results available in the literature, we hypothesize that the efficacy of climate models’ ENSO-induced teleconnection can be assessed by their fidelity in representing anomalous tropical precipitation and associated upper-level divergence, RWS′, and ambient flow characteristics. To test this hypothesis, we have developed a diagnostics package (section 2). Besides assessing climatological basic flow properties and seasonal anomalous conditions during ENSO winters, the barotropic vorticity equation is solved at mid-to-upper-troposphere levels, and terms contributing to the total RWS′ are quantified. Here, to circumvent coupled models’ simulated SST biases cascading into anomalous precipitation and RWS′ characteristics, we focus only on the AMIP-type experiments in which all the participating models are forced by observed SSTs (Gates et al. 1992). This limits the attribution of errors solely to limitations in the representation of atmospheric model processes and their interdependencies. However, the presence of error compensation can make such direct attribution challenging.

The POD is applied to five reanalysis products to assess inherent uncertainties among them, and for the purposes of model assessment, robustness in their results is sought. The reason being that in data sparse regions such as the deep tropics, estimates of diabatic processes are not well constrained by in situ measurements. Then, they depend on the first guess provided by the forecast models that is sensitive to the physical parameterizations employed (Annamalai et al. 1999; Zhang et al. 2008).

Figure 1a shows El Niño winter (DJF) composites of geopotential height anomalies at 200 hPa (HGT200) constructed from ERA5 reanalysis (details are given in section 2). One notes alternating signs in HGT200 that arch from the tropical Pacific poleward to the North Pacific, and then eastward and southward across North America exhibiting a wave train pattern. One clear difference from the traditional PNA pattern of Wallace and Gutzler (1981) is the longitudinal shift of centers (∼10°) over North Pacific and North America (boxes 2–4 in Fig. 1a), in agreement with Lopez and Kirtman (2019). Guided by Fig. 1a and following the Wallace and Gutzler (1981) approach, a circulation index termed El Niño–induced circulation index (ENCI) over the Pacific–North America is calculated using standardized HGT200 as 0.25 × (HGT200_REG1 − HGT200_REG2 + HGT200_REG3 − HGT200_REG4), and the results are shown for the period 1979–2019 (Fig. 1b). For boreal winters, the traditional PNA index obtained from the Climate Prediction Center (CPC) is also shown (Fig. 1c). Both indices, barring differences in magnitude, show positive (negative) values during warm (cold) phases of ENSO, and the cross-correlation between them is 0.8. Our goal here is not to examine how accurately El Niño forces PNA. Our intention is to assess the model processes that are deemed necessary for the forced midlatitude circulation over the Pacific–North American regions during El Niño winters.

View Full Size

(a) El Niño winter (DJF) composites of geopotential height (m) anomalies at 200 hPa; (b) boreal winter El Niño–induced circulation index based on four regions [boxed outlines in (a)] for the period 1979–2019 and is calculated as 0.25 × (HGT200_REG1 − HGT200_REG2 + HGT200_REG3 − HGT200_REG4) and (c) boreal winter traditional PNA index provided by CPC/NOAA. Results in (a) and (b) are derived from ERA5. Here, REG1 is 10°–20°N, 160°–130°W; REG2 is 40°–50°N, 170°–140°W; REG3 is 50°–70°N, 120°–80°W; and REG4 is 30°–40°N, 110°–90°W.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

In this model evaluation study, we assess the models’ fidelity in accurately representing (i) strength and location of RWS′ that depend on anomalous tropical precipitation and associated divergent flow properties, and climatological vorticity gradients in the subtropical jets, and (ii) local vorticity gradient of the ambient zonal flow ( ${\partial }^2\overline{U}/\partial y^2$) in guiding the propagation of Rossby waves. Of relevance, model biases in precipitation (e.g., Kumar et al. 2005; Annamalai et al. 2007; Deser et al. 2016), upper-level divergence and RWS′ (e.g., Jin and Hoskins 1995), and basic-state flow properties (e.g., Branstator 1985; Palmer and Mansfield 1986; Ting and Sardeshmukh 1993) are all expected to influence ENSO-induced teleconnection (Trenberth et al. 1998; Barsugli and Sardeshmukh 2002).

The remainder of the paper is organized as follows: Section 2 deals with the POD and some details of the AMIP simulations and reanalysis products. In section 3, models’ ability in representing equatorial Pacific precipitation (EPP) anomalies to generation of RWS′ are discussed. Section 4 deals with the models’ fidelity in simulating basic-state properties. El Niño–induced teleconnections are interpreted in section 5. In section 6, model biases are quantified, along with improvements or degradations in the transitions from AMIP5 to AMIP6 are shown, and POD’s efficiency to model developers along with key implications are presented. Conclusions and future directions are highlighted in section 7. Further discussions on model interpretations on midlatitude circulation are summarized in appendix.

a. ENSO_RWS

The diagnostics package termed ENSO_RWS consists of four levels or steps that are sequentially performed with monthly data either from reanalysis or model integrations. To attain robust composite results a reasonable sample of ENSO winters is needed. However, the POD can be applied even for a single El Niño winter (e.g., seasonal prediction solutions). Similarly, the POD is applicable to any number of pressure levels (e.g., to identify the level at which maximum upper-level divergence and associated RWS′ are located).

In level 4, results from levels 1–3 are condensed into scatterplots. Specifically, the sequential plots illustrate the model’s ability in representing the chain of processes. The POD is expected to inform the sensitivity of RWS′ to EPP anomalies and ambient flow characteristics and the importance of ${\partial }^2\overline{U}/\partial y^2$ and ${\beta }_{*}$ in the refraction and reflection properties of Rossby waves.

b. Model solutions and reanalysis

The AMIP5/6 suite consists of solutions from a total of 55 models, and the simulation period is 1979–2005 for AMIP5, and 1979–2014 for AMIP6, respectively. Compared to AMIP5, there are additional models in AMIP6 pool (Eyring et al. 2016). The POD is applied to solutions available from all AMIP5/6 models, and results are summarized for AMIP5 and AMIP6 separately as well as for the models that participated in both phases.

To validate model results, similar diagnostics are applied to multiple reanalysis products for the period 1979–2014. We emphasize that reanalysis uncertainties need to be taken into account while validating models. The products we diagnose include the European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERAi; Dee et al. 2011); the fifth major global reanalysis produced by ECMWF (ERA5; Herbasch et al. 2020); the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2; Gelaro et al. 2017); the Japanese 55-year Reanalysis (JRA-55; Kobayashi et al. 2015); and the Climate Forecast System reanalysis (CFS-R; Saha et al. 2010).

For models and reanalysis, variables diagnosed include precipitation, three-dimensional winds and geopotential height. All the diagnostics are performed for composite El Niño winters [years when Niño-3.4 (5°S–5°N, 120°–170°W) SST anomalies exceed 1.0 standard deviation]. Since observed SST is prescribed in AMIP simulations, El Niño winters are identical in both reanalysis and models, and thus offer direct comparisons. Due to space constraints, however, spatial plots from only two reanalysis, CFS-R and ERAi, are shown, since in most of the diagnostics presented here, results from these two reanalyses show lower/upper bounds on the five reanalyses considered.

a. El Niño–related precipitation and circulation anomalies at 200 hPa

Figures 2a–e show anomalous precipitation (shaded), 200-hPa divergence/convergence (contours/hatching) and divergent winds (vector) from reanalyses. Observations (Fig. 2f) show that in response to warm SST anomalies (contours) along the central and eastern equatorial Pacific, precipitation (shaded) is increased with a local maximum around the date line (10°S–0°, 160°E–140°W), termed the EPP region. Besides differences in intensity, all the reanalyses represent the EPP pattern. In response to increased convection, locally concentrated anomalous upper-level divergence and divergent winds are seen. Poleward, the meridional component of the divergent wind anomalies converges over the subtropical North Pacific (STNP; 25°–40°N, 150°E–160°W) and along the South Pacific convergence zone, implying a strengthened local overturning Hadley circulation. In the equatorial plane, zonal weakening of the Walker circulation with enhanced precipitation over the EPP and suppressed precipitation over the Maritime Continent–tropical western Pacific (MC-TWP), resulting in convergence anomalies over MC-TWP are also apparent. Thus, two well-separated upper-level convergence anomalies, one over the STNP and another over the MC-TWP are evident. In the reanalyses, while the structures of divergent wind anomalies are in good agreement, their intensities differ, which is consistent with the varying precipitation anomalies. Note that a direct comparison of divergent winds in reanalyses will depend on the first guess supplied by the forecast models (Dee et al. 2011). This is sensitive to the physical parameterizations used in each reanalysis model (Annamalai et al. 1999).

View Full Size

(a)–(e) El Niño winter (DJF) composites of anomalous precipitation (shaded; mm day⁻¹), 200-hPa convergence/divergence (hatching/contours; 10⁻⁶ s⁻¹) and 200-hPa divergent winds (m s⁻¹; reference vector is shown) constructed from reanalysis products: (a) ERAi, (b) CFSR, (c) MERRA2, (d) ERA5, and (e) JRA55. (f) Composites of observed precipitation from GPCP and SST anomalies (contours with an interval of 0.5°C for positive and 0.2°C for negative values) are shown. Observed SST is taken from Optimum Interpolated SST products. Green vertical lines correspond to westward extension of positive precipitation anomalies (longitudinal position of 1.0 mm day⁻¹ contour) along the equatorial Pacific.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

Figure 3 shows results from select AMIP5 (left) and their corresponding AMIP6 (right) models while Fig. 4 summarizes all AMIP models’ fidelity in capturing anomalous precipitation and STNP convergence. Compared to observations (Fig. 2f), both in GFDL-AM3 (Fig. 3a) and CAM5 (Fig. 3c), EPP anomalies are meridionally confined but zonally elongated, extending well into the western equatorial Pacific (green vertical lines) with implications for westward shifts in anomalous convergence in the STNP. Also, suppressed precipitation and convergence anomalies protrude into the eastern equatorial Indian Ocean. In contrast, EPP anomalies in MIROC5 and MPI-ESM-LR are stronger than observations (Fig. 4c), but STNP convergence anomalies are stronger in MPI-ESM-LR (Fig. 4a) while they are unorganized with multiple local maxima in MIROC5 (Fig. 3e). In GFDL-AM4 (Fig. 3b) and CAM6 (Fig. 3d), notable improvements include a reduced westward extension of EPP anomalies and spatially organized convergence anomalies in the STNP. Based on EPP and MC-TWP precipitation anomalies (Figs. 3b,d and 4d), the anomalous Walker circulation is realistic in GFDL-AM4 while both precipitation anomalies are weaker than observations in CAM6. The reduced precipitation anomalies protruding into the eastern equatorial Indian Ocean still persist from CAM5, and a narrow tongue of unrealistic convergence anomalies extends along the subtropical Asian jet (around 30°N). This is a feature present in all CAM6 model versions including Finite Volume 2 degree resolution (FV2), WACCM, and WACCM-FV2 (not shown). Compared to MIROC5 (Fig. 3e), considerable improvements in MIROC6 (Fig. 3f) include STNP convergence and positive precipitation anomalies along the East Asian monsoon (EAM; 20°–30°N, 100°–140°E). Of all the AMIP5 models, anomalous EPP and STNP convergence are by far the strongest in MPI-ESM-LR (Fig. 4a), while they are weaker in its AMIP6 version (Fig. 4b).

View Full Size

El Niño winter (DJF) composites of anomalous precipitation (shaded; mm day⁻¹), 200-hPa convergence/divergence (hatching/contours in units of 10⁻⁶ s⁻¹), and 200-hPa divergent winds (m s⁻¹) constructed from a few models: (left) AMIP5 and (right)AMIP6. Green vertical lines correspond to westward extension of positive precipitation anomalies (longitudinal position of + 1.0 mm day⁻¹ contour) along the equatorial Pacific.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Scatterplots between simulated anomalous precipitation (mm day⁻¹) averaged over the EPP (10°S–0°, 160°E–140°W) and 200-hPa divergence (10⁻⁶ s⁻¹) averaged over the STNP (25°–40°N, 150°E–160°W): (a) AMIP5 models; (b) AMIP6 models. (c),(d) As in (a) and (b), but for EPP and precipitation anomalies averaged over MC-TWP (5°S–18°N, 110°–150°E). (e),(f) As in (a) and (b), but for MC-TWP precipitation and 200-hPa divergence (10⁻⁶ s⁻¹) averaged over the EAM region (20°–30°N, 100°–140°E). The vertical and horizontal dashed lines correspond to reanalysis uncertainties measured as lower and upper bound values among them, and a model simulation is interpreted to be realistic if its values fall within these uncertainties.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

Scatterplots (Fig. 4) summarize models’ abilities in representing anomalous local Hadley (Figs. 4a,b) and planetary Walker (Figs. 4c,d) circulations. Compared to reanalysis, simulated anomalous tropical precipitation and associated upper-level circulation in most models are weaker. The intermodel spreads between EPP (∼2.5 to 8.0 mm day⁻¹) and STNP convergence (0 to −2.2 × 10⁻⁶ s⁻¹), as well as a varied convergence response for similar EPP forcing and vice versa are readily apparent (Figs. 4a,b). The strong association (both in sign and amplitude) between the EPP wetness and MC-TWP dryness (Figs. 4c,d) can be interpreted as follows: The anomalous Walker circulation with ascent over the equatorial central-eastern Pacific leads to descent and low-level divergence over the MC-TWP. Subsequently, the low-level anomalous westerlies advect climatological moisture from the equatorial west to central Pacific further feeding the convection and enhancing ascent (Annamalai 2020). Figures 4e and 4f summarizes the linkage between MC-TWP dryness and regional divergence anomalies along the EAM front. The low-level anomalous anticyclonic circulation that develops in response to reduced MC-TWP precipitation, through horizontal moist advection, leads to a strengthened EAM and, hence, upper-level divergence (e.g., Annamalai et al. 2005). However, large scatter between MC-TWP dryness and EAM-divergence may be attributed to models’ inability in representing latitudinally confined EAM anomalous precipitation (Figs. 2 and 3). Within reanalysis uncertainties, it must be emphasized that GFDL-AM4 captures realistic simulations of both anomalous EPP and STNP convergence (Fig. 4b), and MC-TWP dryness (Fig. 4d). As will be shown next, both convergence over the STNP and divergent wind anomalies along the EAM contribute to $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ and $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ terms of RWS′, respectively.

b. Anomalous Rossby wave sources at 200 hPa

Figures 5 and 6 show the geographical distributions of RWS′ (shading) from two reanalysis (Figs. 5a and 6a) and a selection of AMIP5 (Figs. 5b–e) and AMIP6 (Figs. 6b–e) models. In the right panels, 200-hPa climatological $\overline{\zeta }\hspace{0.17em}$ (contours), ${\upsilon }_{\chi }^{\prime }$ and total RWS′ (shaded) are shown. Note that the left and center panels have different shading scale. In both reanalyses, compared to subtropics, values of RWS′ are smaller in the deep tropics owing to smallness in $\overline{\zeta }$. Elsewhere, the contribution due to the vortex stretching term ( $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$) dominates the total RWS′ (right panels). Combining earlier results (Figs. 2–4), anomalous convergence over the STNP determines the stretching dominance, resulting in a cyclonic vorticity source. This will be termed the primary RWS′. Note that the primary source is situated in the climatological large-scale trough region (right panels), and this favors Rossby wave rays to be directed more toward the pole (Branstator 1983). ${\upsilon }_{\chi }^{\prime }$ is perpendicularly oriented to the $\overline{\zeta }$ contours and determines the zonal structure and sign of $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ (middle panels) and its contribution with a local maximum along the Asian jet is important for accurate representation of RWS′ over 25°–40°N, 80°E–170°W. Contributions from transient eddy convergence of vorticity [last two terms in Eq. (3)] to RWS′ are small and hence not shown.

View Full Size

200-hPa anomalous Rossby wave source (10⁻¹¹ s⁻²; shading) due to (left) stretching term, (center) advection term, and (right) total from (a) CFSR, (b) AM3, (c) CAM5, (d) MIROC5, and (e) MPI-ESM-LR. Results from a few select AMIP5 models are shown. In the right panels, 200-hPa climatological absolute vorticity (10⁻⁵ s⁻¹; contours with an interval of 2 × 10⁻⁵ s⁻¹ and zero contour as thick) and anomalous divergent winds (m s⁻¹) are also shown. Reference vector is shown. Note that the center column has a different shading scale.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

200-hPa anomalous Rossby wave source (10⁻¹¹ s⁻²; shading) due to (left) stretching term, (center) advection term, and (right) total from (a) ERAi, (b) GFDL-AM4, (c) CAM6, (d) MIROC6, and (e) MPI-ESM2-LR. Results from a few select AMIP6 models are shown. In the right panels, 200-hPa climatological absolute vorticity (10⁻⁵ s⁻¹; contours with an interval of 2 × 10⁻⁵ s⁻¹ and zero contour as thick) and anomalous divergent winds (m s⁻¹) are also shown. Reference vector is shown. Note that the center column has a different shading scale. The boxed region in the top-left (top-center) panel shows the averaging area used to estimate $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ ( $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$) for results shown in later figures.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

In GFDL-AM3, the STNP maxima in $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ is weak and located west of the date line. Along the Asian jet the tongue of positive vorticity due to $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ is also shifted west. Both are factors attributed to errors in representing perturbations to the local Hadley and planetary Walker circulations as implied in the divergent winds (right panels in Figs. 5 and 6). These limitations are seemingly overcome in GFDL-AM4 (Fig. 6b). Similarly, compared to CAM5, there is a clear improvement in the representation of the primary RWS′ in CAM6; however, the westward extension of zonally elongated signatures in both $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ and $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ along the Asian jet (Fig. 6c) reflect the need for further model improvements. Compared to its AMIP5 counterpart (Fig. 5d), components of RWS′ are similarly better represented in MIROC6 (Fig. 6d). Of all the AMIP6 models, amplitudes of both $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ and $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ are the strongest in MIROC6, resulting in a too-strong RWS′ over the STNP (Figs. 7c,d). In contrast, realistic representations of RWS′ in MPI-ESM-LR (Fig. 5e) are degraded in its AMIP6 version (Fig. 6e).

View Full Size

(left) Scatterplots between simulated anomalous 200-hPa divergence (10⁻⁶ s⁻¹) averaged over the subtropical North Pacific (25°–40°N, 150°E–160°W) and RWS′ due to stretching term, $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ (10⁻¹¹ s⁻²); (right) scatterplots between simulated anomalous precipitation (mm day⁻¹) averaged over the East Asian monsoon front (22°–32°N, 110°–140°E) and RWS′ due to advection term, $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ (10⁻¹¹ s⁻²). (a),(b) All AMIP5 models and (c),(d) all AMIP6 models. The vertical and horizontal dashed lines correspond to reanalysis uncertainties measured as lower and upper bound values among them, and a model simulation is interpreted to be realistic if its values fall within these uncertainties. The averaging areas for $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ and $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ are shown in Fig. 6a.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

To assess the role of simulated biases in anomalous precipitation, divergence, and climatological $\overline{\zeta }$ and its gradient in representing leading RWS′ in models, we show scatterplots among these variables (Fig. 7). While there is considerable model scatter between simulated anomalous EPP and STNP convergence (Fig. 4), the subsequent relationship of STNP divergence on the primary RWS′ is largely linear for most models. The diverse response in $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ in AMIP6 for divergence anomalies around ∼−1.25 (Fig. 7c), and −0.7 in AMIP5 (Fig. 7a) is perhaps due to diversity in models’ climatological relative vorticity averaged over the STNP (not shown). Representing anomalous positive precipitation along the EAM front and associated divergent winds that determine $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ are indeed difficult. This is in large part due to existing precipitation biases in models, but also due to how models relate precipitation anomalies to circulation anomalies given the vertical profile of heating is key to this connection. Even in reanalyses (Fig. 7d; note the spread in precipitation anomalies from 0.7 to 1.2 mm day⁻¹ for $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ values ∼5.0 × 10⁻¹¹ s⁻²) the relationship is poorly constrained. Reanalyses results suggest that contributions from $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ lie in the range of 4.5–5.5 × 10⁻¹¹ s⁻² while that of $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ is 11–13 × 10⁻¹¹ s⁻². In both AMIP versions, large scatters and ranges in both $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ and $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$ imply errors in divergent flow properties that depend on models’ accurately representing tropospheric diabatic processes. Compared to their AMIP5 counterparts, many models (GFDL-AM4, CAM6, bcc-csm2-mr, MIROC6, NorESM2-LR, CMCC-CM2-SR5) show improvements in $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$ in their AMIP6 versions (Fig. 7c). More discussions are provided in section 6 (Figs. 14–16).

In summary, the POD identifies model STNP RWS′ deficiencies as being due to the following:

• Limitations in ${\upsilon }_{\chi }^{\prime }$ over the STNP highlighting limitations in representing perturbations to the local Hadley circulation, and hence on $-\overline{\zeta }\nabla \cdot {\upsilon }_{\chi }^{\prime }$.

• Westward extension of EPP anomalies and associated anomalous convergence, accounting for spatially incoherent and zonally elongated RWS′ along the jet.

• Limitations in representing the structure of $\nabla \overline{\zeta }$ along the Asian jet, and enhanced precipitation along the EAM front, which have implications for RWS′ through contributions from $-{\upsilon }_{\chi }^{\prime }\cdot \nabla \overline{\zeta }$.

a. Circulation response over North Pacific and North America

Figures 10 and 11 show 200-hPa boreal winter climatological distribution of ${\beta }_{*}$ superimposed on total RWS′ (left panels), along with HGT200 and V200 (right panels). Since the deep tropical circulation response is robust across the models, we show HGT200 poleward of 20°N only. Spatial variations in ${\partial }^2\overline{U}/\partial y^2$ are shown in Fig. 12. The same was previously shown for GFDL AM3/AM4 in Fig. 9. The observed circulation response comprises centers of alternating signs in HGT200 (e.g., Figs. 10f and 11f) and arches from the tropical Pacific poleward to North Pacific, and then eastward and southward across North America exhibiting a wave train pattern. V200 shows patterns that are in quadrature with HGT200. Based on the above results, we will assess the following:

View Full Size

200-hPa boreal winter (DJF): (left) climatological ${\beta }_{\ast }$ (shading; 10⁻¹¹ m⁻¹ s⁻¹) and RWS′ (dark hatching for positive values and light hatching for negative values; 10⁻¹¹s⁻²); (right) 200-hPa geopotential height (shaded; m) and meridional wind (contours; m s⁻¹) anomalies during El Niño. Vertical lines in the left column correspond to longitude of zero values of ${\beta }_{\ast }$ in North Pacific. Results from ERAi and a few AMIP6 models are shown.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Spatial maps of ${\partial }^2\overline{U}/\partial y^2$ (or local vorticity gradient) of the ambient zonal flow ( $\overline{U}$) at 200 hPa. Results are from (a),(b) two reanalysis and (c)–(f) select AMIP5 (AMIP6) models are presented in the left (right) panels. Units are 10⁻¹¹ m⁻¹ s⁻¹.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

• The coherent versus diffuse nature of the primary RWS′ region and its location in the STNP (west or east of the date line or zonally elongated along the jet).

• Spatial distributions of ${\partial }^2\overline{U}/\partial y^2$ and ${\beta }_{*}$ on their impact on reflection/refraction properties of the waves.

Of the AMIP5/6 models, GFDL-AM4 and CAM6 capture the most realistic El Niño–induced circulation anomalies (Figs. 11g,h) with a characteristic meridional propagation from the tropics to North Pacific and then arching northeastwards covering North America, and finally southern parts of the United States of America. Over the north-central Pacific, compared to AMIP5 versions (Figs. 10b,c), there are improvements in representing ${\partial }^2\overline{U}/\partial y^2$ (Figs. 9, 12) and ${\beta }_{*}$, and a realistic RWS′ over STNP (Figs. 11b,c, 13c,d), all factors favoring improved midlatitude circulation response over the Pacific–North American region

In many models, a very common erroneous feature is zonally elongated (northwest–southeast oriented) HGT200 over the Aleutian low region. If the primary RWS′ is positioned to the west of the date line (e.g., MIROC5; Fig. 10d) then owing to the presence of positive values in ${\partial }^2\overline{U}/\partial y^2$ (Fig. 12) or negative values of ${\beta }_{*}$ to the west and poleward of this RWS′, meridional propagation of Rossby waves are not permitted. Therefore, waves generated there will get reflected and become zonally elongated in the east–west direction, as ray tracing theory would suggest (Hoskins and Karoly 1981; Hoskins and Ambrizzi 1993). If the primary RWS′ is diffused (MPI-ESM2-LR; Fig. 6e), and/or if negative values of ${\beta }_{*}$ over central North Pacific extend too eastward (MIROC6; Fig. 11d) HGT200 over the Aleutian low region are weak and zonally elongated as well (Figs. 11i,j). Specifically, instead of pure meridional propagation from the forcing region, both HGT200 and V200 patterns in Figs. 11i and 11j suggests northeast arching wave train indicating reflected waves due to inaccurate ambient flow conditions.

In GFDL-AM3 (Fig. 10b), negative values in ${\beta }_{*}$ are not well defined over the central-North Pacific and thus meridional propagation of Rossby waves forced by RWS′ positioned to the west of the date line is possible. Therefore, we infer waves emanating from west of Japan arch northeastward depict a different propagation path (Fig. 10g). Furthermore, in both GFDL-AM3 and MIROC5 the primary RWS′ reside along the Asian jet (Figs. 10b,d) in which K_s is higher (Fig. 9g), and ${\beta }_{*}$ is strongest (Fig. 9e). Then the radiated waves get trapped along the waveguide and tend to move poleward and equatorward as they exit the jet region. Thus, interference from multiple waves with different wavelengths may contribute to unrealistic circulation anomalies, as implied in barotropic model solutions (Held and Kang 1987). Therefore, in assessing climate models’ fidelity to represent PNA or ENCI it is crucial to assess models’ ability in accurately representing ${\beta }_{*}\text{LON}$. Ultimately, this metric depends on a model’s eastward extension of the Asian jet, and more precisely on ${\partial }^2\overline{U}/\partial y^2$.

b. Metrics to assess models’ circulation performance

We identify two metrics, namely, RWS′ and ${\beta }_{*}\text{LON}$, and to assess models’ PNA/ENCI dependence on either of them, scatterplots are shown in Fig. 13. Results from AMIP5 (AMIP6) are shown in the top (middle) panels for PNA, and only AMIP6 for ENCI are shown in the bottom panels. In the left panels, 0 in ${\beta }_{*}\text{LON}$ refers to the date line. In both metrics, there are clear uncertainties among reanalyses, and assessing the models within these uncertainties suggests that there is a clear improvement in AMIP6, that is, the number of models with realistic PNA/ENCI or RWS′ is higher compared to AMIP5. For both circulation indices, similar response for diverse RWS′ and varied response for similar RWS′, and similar diversity in ${\beta }_{*}\text{LON}$ and PNA/ENCI are common. Even if the two metrics are realistic, simulated PNA/ENCI can be either stronger (e.g., GFDL-AM4 and CAM6) or weaker [e.g., Taiwan Earth System Model, version 1 (TaiESM1) and Meteorological Sciences Community System Model 1-0 (MS-CSM-1-0)]. Additionally, even for weaker RWS′ a realistic PNA/ENCI is simulated (e.g., ACCESS-ESM1-5). Generally, sensitivity of the two metrics is similar to both ENCI and PNA with one clear exception. Of all the AMIP6 models, MIROC6 represents the strongest RWS′ while the simulated PNA is the weakest, but ENCI appears realistic (Fig. 13f). A careful examination of contribution from all the four regions (Fig. 1a) indicate that due to unrealistic eastward extension of negative HGT200 over North Pacific and continental United States (Fig. 11i), contribution from the fourth region overwhelms, resulting in a realistic ENCI. Results presented in section 5 suggest that the two metrics are necessary but not sufficient conditions to represent a realistic midlatitude circulation response during El Niño winters. More discussions on this aspect are provided below.

View Full Size

(left) Scatterplots between RWS′ (10⁻¹¹ s⁻²) averaged over the STNP (25°–40°N, 150°E–160°W) and PNA; (right) scatterplots between ${\beta }_{\ast }\text{LON}$ in the North Pacific and PNA. (a),(b) All AMIP5 models and (c),(d) all AMIP6 models. Scatterplot (e) between RWS′ and ENCI and (f) between ${\beta }_{\ast }\text{LON}$ and ENCI in AMIP6 models. The vertical and horizontal dashed lines correspond to reanalysis uncertainties measured as lower and upper bound values among them, and a model simulation is interpreted to be realistic if its values fall within these uncertainties.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

a. Quantification of model biases

Figure 14 shows model biases (expressed in %) with respect to multireanalysis mean anomalies. AMIP5 (AMIP6) model biases are shown in the left (right) panels, and the key variables discussed in previous sections are shown. EPP wetness and MC-TWP dryness, in about two-thirds of all AMIP models, are weaker than observations. Irrespective of a model’s ability to represent equatorial precipitation anomalies, biases in STNP convergence (and associated RWS′) are systematically weaker than in reanalysis. This implies a role for other processes in determining actual STNP convergence. In some models, the maximum level of anomalous divergence/convergence may not be at 200 hPa, an issue currently being investigated. As regards to ${\beta }_{*}\text{LON}$, biases in most models are inclined toward being east of the date line in AMIP6 compared to west of the date line in AMIP5. In both AMIP versions, biases in both circulation indices are similar in any given model, suggesting model consistency in representing either of the two indices. In summary, model biases quantitatively range from (i) −55% to 40% for tropical precipitation, (ii) −80% to 20% for STNP convergence that directly impact primary RWS′ (−75% to 30%), and (iii) −60% to 40% for circulation response over the PNA region.

View Full Size

(left) AMIP5 and (right) AMIP6 model biases with respect to multireanalysis (expressed in %). The variables are (first row) EPP, (second row) MC-TWP precipitation, (third row) STNP divergence at 200 hPa, (fourth row) RWS′ averaged over STNP, (fifth row) ${\beta }_{\ast }\text{LON}$ in the central-North Pacific (0 values in ${\beta }_{\ast }\text{LON}$ refers to date line), (sixth row) PNA, and (seventh row) ENCI indices.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

b. Improvements or degradations from AMIP5 to AMIP6

Model results of the chain of processes presented above suggest that some of them are improved in AMIP6 and some are degraded. To highlight improvements for each modeling center, Fig. 15 shows results from models that participated in both AMIP versions. Here, biases are estimated with respect to their corresponding multireanalysis means. If the AMIP6 bias falls closer to reanalyses (boxed area in Fig. 15) compared to AMIP5 biases, it is interpreted as improved (left panels). Conversely, if the bias falls more distant, it is interpreted as degraded (right panels). In all panels, number 5 stands for AMIP5 and 6 refers to AMIP6, and colors correspond to the model’s name. If only one of the two variables in Fig. 15 is either improved or degraded, that model’s results are not shown.

View Full Size

Scatterplots to illustrate (a)–(d) improvements or (e)–(h) degradations in both variables in each panel from AMIP5 to AMIP6. Variables considered are RWS′ (10⁻¹¹ s⁻²) averaged over the STNP, ${\beta }_{\ast }\text{LON}$ in the North Pacific, PNA and ENCI circulation indices. Model biases are estimated with respect to their corresponding multireanalysis means. (left) If the AMIP6 bias falls closer to reanalyses (boxed area) compared to AMIP5 biases, it is interpreted as improved. (right) Conversely, if the bias falls more distant, it is interpreted as degraded. In all panels, number 5 stands for AMIP5 and 6 refers to AMIP6, and colors correspond to the model’s name. The 0 values in ${\beta }_{\ast }\text{LON}$ refer to the date line.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

AMIP5 model values that lie in the bottom left quadrants (weak in both variables; Figs. 15a,c), and more eastward values of ${\beta }_{*}\text{LON}$ (Figs. 15b,d) converge toward reanalysis values (boxed area) in AMIP6. These results convince us that there is an overall improvement in AMIP6 models’ representations of RWS′, ${\beta }_{*}\text{LON}$, PNA, and ENCI. In each of the panels, while the number of models showing improvements in both variables is different, four models show consistent improvements and they are BCC-CSM2-MR, NCAR-CAM6, GFDL-AM4, and CMCC-CM2-SR5. In these models, improved basic-state as evidenced by ${\beta }_{*}$ (Figs. 15b,d) and representation of RWS′ (Figs. 5 and 6) in regions of lower K_s (Figs. 8 and 9), favor meridional propagation. While six models show improvements in RWS′ and PNA (Fig. 15a), ${\beta }_{*}\text{LON}$ in CanESM5, for example, is positioned too westward of the date line (Fig. 13d), and an unrealistically strong PNA/ENCI is simulated. We offer possible reasons for these extremes in the appendix. It should be mentioned here that models that show improvements or degradations in RWS′ and PNA pair, or ${\beta }_{*}\text{LON}$ and PNA pair, show similar results in RWS′ and ENCI pair, or ${\beta }_{*}\text{LON}$ and ENCI pair.

In considering ${\beta }_{*}\text{LON}$ and RWS′ as two independent metrics that collectively determine PNA/ENCI, we summarize their joint relationships in Fig. 16 with conventions used as in Fig. 15. In this cubic view, if both independent metrics are clustered around the reanalysis values (gray cubes), then height anomalies averaged over the PNA and ENCI are realistic, and improvements in ${\beta }_{*}\text{LON}$ and RWS′ and ENCI or PNA are readily apparent in their AMIP6 versions for the four models earlier identified in Fig. 15. Similarly, degradations in two models (IPSL-CM6A-LR and NorESM2-LM) are outliers. Furthermore, this plot suggests that if one of the two metrics is unrealistic so is the PNA/ENCI response. Figures 13 and 14 suggest that other AMIP6 models show realistic values either in RWS′ [e.g., Seoul National University Atmosphere Model, version 0, with a Unified Convection Scheme (SAM0-UNICON)] or ${\beta }_{*}\text{LON}$ [e.g., Energy Exascale Earth System Model, version 1.0 (E3SM-1-0)] or both (e.g., TaiESM1). We discuss some aspects of these models’ fidelity in the appendix.

View Full Size

Three-dimensional (3D) plots to highlight AMIP5/6 models representation of two metrics and their impact on (a) PNA and (b) ENCI indices. Independent variables are ${\beta }_{\ast }\text{LON}$ in the North Pacific and RWS′ (10⁻¹¹ s⁻²) averaged over the STNP east of the date line (25°–40°N, 180°–160°W). Model biases are estimated with respect to their corresponding multireanalysis means. If the AMIP6 bias falls closer to reanalyses (boxed area) compared to AMIP5 biases, it is interpreted as improved. Conversely, if the bias falls more distant, it is interpreted as degraded. In all panels, number 5 stands for AMIP5 and 6 refers to AMIP6, and colors correspond to the model’s name. Reference unit “line” represents 0.25 standard deviation. Models in which both metrics and PNA/ENCI indices improved are highlighted at the top.

Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-22-0346.1

• Download Figure
• Download figure as PowerPoint slide

c. Implications for model developers

In models, both in basic-state and anomalous conditions, changes in the response of moist processes either to parameterization modifications or tuning and calibration, can often change the nature of the seasonal distributions of tropical precipitation and heating, and by association moistening and divergence profiles. While validation of an integrated scalar quantity such as precipitation is mostly straightforward, an understanding of the circulation consequences in the tropics, and particularly the nonlinear wave–mean flow interactions of the extratropics is not. The intention would be to monitor simulation improvements over time. Equally, however, when inadvertent or unintended degradations are discovered, decisions need to be made on how to move forward. This could involve either a concerted effort to understand the processes leading to the degradation or accepting the response and moving forward regardless. By having a suite of PODs as an automated tool, a broad range of simulation features could be invoked that span the range of performance features for which a model is being designed. In the case of climate models this would include large-scale circulation features (the focus here), modes of variability and climate feedback processes (e.g., cloud feedbacks, precipitation, and temperatures). Thus, the assessment presented here is intended to provide useful feedback to model developers where POD performance changes can be linked to parameterization improvements, even though these linkages can often be circuitous and difficult to disentangle.

Model developers primarily focus on a very limited set of requirements or metrics in order to produce a useful model, often for CMIP (and also AMIP) purposes (Schmidt et al. 2017; Hourdin et al. 2017). The focus is often biased toward the larger climate change requirements, in order to capture historical, and therefore reliable, future evolution of a warming world, even though there is disagreement in priorities across the modeling centers. The analysis presented here is different, in that it captures a succession of processes critical to both climate change and Earth system prediction in the Northern Hemisphere, but it does so with a process-oriented focus that follows this succession from the tropical Pacific to the West Coast of the United States. Applying this particular POD continually through the development process will keep a regular check on the fidelity of the modeled response to anomalous El Niño convection (such a progression in coupled models is obviously a much greater challenge given the variation in Pacific SSTs and associated model biases). Applying a set of PODs is expected to reduce the frequent occurrence in model development of a particular process or mode of variability degrading over time. This degradation can be unknown to the developers earlier due to lack of availability of appropriate PODs. The hope is that it will become a standard method to assess models, in the same way that ENSO, the North Atlantic Oscillation (Simpson et al. 2020), and Madden–Julian oscillation (Chen et al. 2022) have become to date.