ENSO Diversity and the Simulation of Its Teleconnections to Winter Precipitation Extremes Over the US in High Resolution Earth System Models

Abstract

Accounting for the diversity in El Niño Southern Oscillation (ENSO)'s spatial pattern, with the novel ENSO longitudinal index (ELI), we evaluate the simulation of its teleconnections to US winter precipitation extremes by seven global high-resolution (HR) Earth System Models (ESM). Six (four) HR ESMs simulate the observed increase in precipitation extremes over Southwest US (Southeast US) during ELI-defined El Niño events better than their low-resolution counterparts, which are low-biased. The stronger ENSO-dependence over the Southwest US and Southeast US in those models is associated with an improved simulation of moisture flux into the regions and/or storm track activity there. HR ESMs, however, generally overestimate the increase in precipitation extremes over the Pacific-Northwest during La Niña events. Model bias there is associated with bias in moisture transport into the region during La Niña events, which is amplified by the enhanced vertical mass fluxes in HR.

Key Points

• We evaluate seven high-resolution (HR) Earth System Models' simulation of diverse El Niño Southern Oscillation (ENSO) teleconnections to US winter precipitation extremes

• Simulation of ENSO-dependent Southwest-US (Southeast-US) precipitation extremes is improved in six (four) models with resolution increase

• This is due to improvements in ENSO dependent moisture transport, storm track activity, and vertical mass fluxes in HR models

Plain Language Summary

El Niño Southern Oscillation (ENSO) comes in many flavors with diverse spatial pattern of ocean warming and associated heavy tropical rainfall over the deep tropical Pacific. This ENSO diversity is one of the major reasons behind our limited understanding and prediction of its global impacts, particularly on climate extremes. A recent approach to characterize the full spectrum of ENSO using a simple index shows promise in improving our understanding. Here, we use this new index to evaluate if new state-of-the-art high resolution Earth System Models can capture the remote impacts of ENSO on US precipitation extremes in the winter. We find that some of these models can credibly simulate these ENSO teleconnections, across its many flavors, over the SE-US and SW-US, and generally improve upon their low resolution model counterparts.

1 Introduction

The diversity of the spatial pattern of El Niño Southern Oscillation (ENSO) events has limited our understanding of ENSO-associated predictability of regional water cycles. Various indices have been used to characterize and quantify the strength of unique facets of ENSO. A recent study (Williams & Patricola, 2018) proposed a unified approach to characterize ENSO's spatial diversity with a single non-linear index, namely the ENSO longitudinal index (ELI). It is calculated as the average longitude over the tropical Pacific where the sea surface temperature (SST) is above the threshold for deep convection (Williams & Patricola, 2018). Consequently, ELI represents the location of deep convection and the upwards branch of the Walker circulation over the tropical Pacific and tracks their zonal shifts associated with ENSO. These movements directly impact ENSO teleconnections to the mid-latitudes by modulating the extra-tropical wave-trains that impact moisture transport, storm track activity, etc. over remote regions (Patricola et al., 2020). ELI thus has been shown to be more effective at capturing teleconnections to seasonal mean and extreme precipitation regions like California and Southeastern US as compared to other conventional fixed domain indices like Niño 3.4 index (Patricola et al., 2020; Williams & Patricola, 2018).

The simulation of regional precipitation and its extremes remains a challenge for Earth System Models (ESMs). Higher resolution ESMs resolve more fine scale features than prevalent low resolution (100 km in the atmosphere) models representing more realistic orographic lifting, vertical mass fluxes, coastal processes, land use as well as mesoscale ocean eddies, although still relying on parameterization for sub-grid scale processes like convection. High-resolution (HR) global models generally appear to improve the simulation of mean and extreme precipitation as compared to their low resolution counterparts, producing more intense precipitation, which can also sometimes be unrealistic (e.g., Bador et al., 2020; Mahajan et al., 2015, 2018, 2022; Wehner et al., 2014, also see Figure S1 in Supporting Information S1).

However, only a few studies have evaluated remote teleconnections of low-frequency variability phenomena like North Atlantic Oscillation and ENSO to precipitation in HR ESMs (e.g., Delworth et al., 2012; Mahajan et al., 2018; Molteni et al., 2020; Roberts et al., 2018; Zadra et al., 2018). Fewer studies (e.g., Mahajan et al., 2018, 2022) evaluate the teleconnections to precipitation extremes. While these studies generally show improvements with resolution, this resolution sensitivity remains unclear (e.g., Molteni et al., 2020). Further, these studies and others evaluating conventional low-resolution ESMs (e.g., Beverley et al., 2021; Whan & Zwiers, 2017) have largely used the traditional static metrics of ENSO (like the Niño3.4 index) that either do not account for the spatial diversity of ENSO, or evaluate different ENSO flavors in isolation using multiple indices (e.g., Fredriksen et al., 2020). Moreover, as the tropical Pacific warms in future climate projections and idealized global warming experiments (for e.g., by 1–3 K from pre-industrial conditions in the transient 1pctCO₂ simulations, where carbon dioxide concentration increases at a rate of 1% per year until quadrupling in 140 years, J. R. Brown et al., 2020), there is a reduction of the climatological tropical Pacific zonal SST gradient (e.g., Beverley et al., 2021; J. R. Brown et al., 2020; Fredriksen et al., 2020; Williams & Patricola, 2018). These simulations exhibit little warming over the Niño3.4 region, but predict a likely increase in the frequency of extreme East Pacific ENSO events due to stronger surface warming over the cold tongue region (e.g., Beverley et al., 2021; Fredriksen et al., 2020; Williams & Patricola, 2018). It is important to evaluate if ESMs are able to capture the teleconnections to all the diverse forms of ENSO events to lay confidence in their projected future remote global impacts of a changing ENSO.

Here, we use the dynamic ELI index, which also captures ENSO diversity in future projections (Williams & Patricola, 2018), to evaluate seven HR ESMs' simulation of full-spectrum ENSO teleconnections to continental US winter season (November–February) precipitation extremes, when the ENSO anomalies and their impacts are strongest. We also analyze the associated large scale dynamics that may drive these remote impacts, namely moisture transport and storm track activity.

2 Simulations and Observational Data

We evaluate 100-year 1950-control simulations with fully coupled HR (25–50 km atmosphere model nominal resolutions) version of E3SM1 and five models (CNRM-CM6-1, EC-Earth-3P, ECMWF-IFS, HadGEM3-GC31, and MPI-ESM1-2) participating in the HighResMIP activity (Haarsma et al., 2016) with daily precipitation data available in the CMIP6 archive (Eyring et al., 2016). We also evaluate a segment (1950–2014) of a historical simulation conducted with the North American Regionally Refined Model version of E3SM2, where the atmosphere model nominal resolution over North America and the surrounding oceans is 25 km, and decreases to 100 km elsewhere (Tang et al., 2022). We also compare these HR simulations against their counterpart low resolution (50–250 km atmosphere model nominal resolution) production model simulations. We collectively refer to the high (low) resolution models as HR (LR) and also suffix references to the individual models with HR (LR), hereafter. More details about these models are included in Section S1 and Table S1 in Supporting Information S1.

We use NOAA Climate Prediction Center's (CPC) unified gauge-based analysis (Xie et al., 2007) at a resolution of 50 km for evaluating model simulated precipitation. We also use ERA5 reanalysis (Hersbach et al., 2020), which is computed at about 25 km horizontal resolution, to evaluate simulation of large scale dynamics by the models. We also evaluate the models against Global Precipitation Climatology Project's (GPCP) 1-degree daily (1DD) precipitation product (v1.3) (Huffman et al., 2001) and NASA's Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA2) reanalysis produced at a resolution of about 50 km (Gelaro et al., 2017), which yield qualitatively similar results. We, thus, generally limit model validation discussions to that against CPC data and ERA5 here. Model precipitation data was conservatively mapped to the CPC data grid for analysis. Using bilinear interpolation yielded similar results. Model variables validated against ERA5 were bilinearly interpolated to the ERA5 grid before analysis.

3 Methods

We model precipitation extremes using the Generalized Extremes Value (GEV) distribution, which is represented as follows:

$$$G(z)=\mathrm{exp}\left\{-{\left[1+\xi \left(\frac{z-\mu }{\sigma }\right)\right]}^{-1/\xi }\right\}$$$ (1)

Here, μ, σ and ξ are the location, scale and shape parameters of the distribution, respectively. The distribution is computed as its limit when ξ = 0 (Coles, 2001). A GEV is fit to the monthly maximum of daily precipitation for the months of November to February at each grid point following previous studies (Mahajan et al., 2018, 2022; Whan & Zwiers, 2017). The relationship of extremes to ENSO is captured by a linear dependence of the location parameter on the ELI, using the standardized ELI as a covariate:

$$$\mu ={\mu }_{0}+{\alpha }_{\mathit{ELI}}.EL{I}_{t}$$$ (2)

Here, ELI_t is the value of standardized ELI for the corresponding time, t, and α_ELI quantifies the linear dependence of μ on ELI—with larger magnitudes of α_ELI indicating a stronger dependence on ELI. All the GEV parameters, including α_ELI are estimated using the maximum likelihood method. This generalized linear model approach to capture trends and teleconnections of extremes to climate variability modes are now widely used (e.g., S. J. Brown et al., 2008; Evans et al., 2014; Mahajan et al., 2015; Whan & Zwiers, 2017). More details on GEV and associated null hypothesis testing are provided in Section S2 in Supporting Information S1.

ELI is computed following Williams and Patricola (2018). The deep convective SST threshold for each month in the time series is assumed to be the global average tropical SSTs for that month. This is based on the rapid homogenization of the heating over the deep convective regions in the tropics to the entire tropical troposphere by equatorial gravity waves, which allows the approximation of the entire tropics as a single moist adiabat. This makes the average global tropical SST to closely relate to the deep convective SST threshold. ELI is then computed as the average of the longitudes of each grid box with an SST greater than the computed deep convective threshold in the deep tropical Pacific (5°S–5°N, 120°–280°E) for that month. The computation of the deep convective threshold as the average global tropical SSTs allows it to account for differing background SSTs associated with the seasonal cycle or climate change (Williams & Patricola, 2018).

4 Results

4.1 ENSO Diversity in E3SM

Figure 1a shows the probability density function of the winter season (November–February) ELI for ERA5 data, LR and HR models. Mean and standard deviation of ELI for ERA5 and models are listed in Table S2 in Supporting Information S1. MERRA2 also show similar results (not shown). All LR and HR models demonstrate a eastwards skewed distribution of ELI, similar to ERA5. However, the peak of the density functions ranges from about 160° to 180°E in the models as compared to the about 170°E for ERA5. This implies that some models (e.g., CNRM-CM6-1, MPI-ESM1-2) have a higher than observed frequency of Eastern Pacific El Niño events (ELI ≥ 180°E), and some (e.g., E3SMv1, E3SMv2) simulate a higher than observed frequency of La Niña events (ELI ≤ 158°E), based on ENSO definitions using the ELI metric (Patricola et al., 2020). Also, most models exhibit reduced variability (Figure 1a, Table S2 in Supporting Information S1) and have a much stronger peak and narrower spread of ELI as compared to ERA5 implying more localized SST anomalies in simulations as compared to observations over the tropical Pacific. Some models also exhibit a strong resolution sensitivity with a shift in the density function with increase in resolution (e.g., MPI-ESM1-2).

Figure 1

Probability density function of winter season mean (a) El Niño Southern Oscillation Longitudinal Index and (b) Niño3.4 region averaged sea surface temperatures for ERA5 data (thick black solid lines) and low-resolution (dashed and colored lines) and high-resolution (solid and colored lines) counter-part 1950-control model simulations.

ELI only represents the average longitudinal location of strong SST anomalies, and does not directly account for the intensity of those anomalies. Figure 1b shows the density function of Niño3.4 region averaged winter season SST anomalies for ERA5 and model simulations and the standard deviations are listed in Table S3 in Supporting Information S1. We consider Niño3.4 region (5°S–5°N, 120°–170°W, Bamston et al., 1997) averaged anomalies to also serve as a proxy for the intensity of strongest whole tropical Pacific SST anomalies. Most models simulate stronger anomalies over the region with higher frequency than observations. Two models (CNRM-CM6-1 and MPI-ESM1-2) also exhibit strong resolution sensitivity, exhibiting weaker variability at HR with a much lower frequency of events stronger than ±1 K as compared to observations.

4.2 Extremes and ELI-Based ENSO Teleconnections

Consistent with previous studies (e.g., Mahajan et al., 2022; Wehner et al., 2014), HR models generally exhibit stronger extremes with finer spatial details compared to LR models, for example, over Southwest-US (SW-US), Pacific-Northwest (P-NW), and Southeast-US (SE-US, Figure S1 in Supporting Information S1, which shows the location parameter (μ) of winter precipitation extremes for CPC data, LR, and HR models). Figure 2 shows the dependence of the location parameter of winter precipitation extremes on ELI (α_ELI) for CPC data, LR, and HR models. Positive (negative) values indicate increase in the likelihood of stronger extremes with El-Niño (La-Niña) events. RMSE and pattern correlation of α_ELI of model data against that of CPC data over land in the plotted domain is also listed. CPC data shows strong ELI-dependence of precipitation extremes over SE-US and along the coastlines and over the Sierra Nevada mountains in California, with an increase in the likelihood of precipitation extremes during El-Niño events. This dependence is statistically significant at the 95% confidence level over several regions in the SE-US based on the likelihood ratio test and also accounting for multi-testing using the false discovery rate approach. This response pattern is generally consistent with GPCP data (not shown) and previous studies that use other observational data sets (Patricola et al., 2020; Williams & Patricola, 2018). ELI-dependence is found to be stronger than Niño3.4 dependence of extremes over the SW-US region (Figure S2 in Supporting Information S1), consistent with Patricola et al. (2020).

Figure 2

Precipitation extremes response to El Niño Southern Oscillation (ENSO). Geographic distribution of the ENSO longitudinal index (ELI)-dependent location parameter (α_ELI) of Generalized Extremes Value (GEV) fits to monthly block maxima of daily precipitation rates (units: mm day⁻¹) in the winter months for (a) Climate Prediction Center (CPC) data (1980–2018), and 1950 control simulation of low-resolution and high-resolution model configurations of (b) E3SM1, (c) E3SM2, (d) CNRM-CM6-1, (e) EC-Earth3P, (f) ECMWF-IFS, (g) HadGEM3-GC31, and (h) MPI-ESM1-2. Hatched areas denote grid points where the GEV model with ELI as a covariate is significantly different than a GEV model without it with false discovery rate corrections for multi-testing at the 0.05 global significance level. The root-mean squared errors and pattern correlations of simulations against the CPC data are also listed along with the nominal model resolution.

HR models generally (e.g., E3SM1, ECMWF-IFS, HadGEM3-GC31, MPI-ESM1-2), though not all, simulate a stronger response over parts of SE-US as well as Southwest-US (SW-US) as compared to the LR, which exhibit a low bias. But, overall LR and HR models exhibit similar domain RMSE and pattern correlation. Over the SE-US, some HR model responses are comparable in magnitude and spatial distribution to CPC data (e.g., E3SM1), but most models (e.g., ECMWF-IFS, HadGEM3-GC31, MPI-ESM1-2) exhibit ELI-dependence only toward the eastern part of SE-US and over Florida—where the dependence is stronger than observed in some models (e.g., HadGEM-3-GC31, MPI-ESM1-2). The RMSE in four HR models (E3SM1, CNRM-CM6-1, EC-Earth3P, and ECMWF-IFS) is reduced over their LR counterparts over the SE-US region (Box A in Figure 2, Table S4 in Supporting Information S1), between 3% and 41% with the largest reduction noted in E3SM1. HR response remains weaker than observed over SW-US for most models, but improves over their LR counterparts for all models (except MPI-ESM1-2) with the RMSE over the region (Box B in Figure 2, Table S5 in Supporting Information S1) reducing between 4.5% and 36%.

HR models also simulate a statistically significant increase in precipitation extremes during La Niña events over the P-NW, which is stronger than that seen in CPC data. This bias is also seen in some LR models. Using Niño3.4 index as a covariate instead of ELI, Mahajan et al. (2022) showed improvements in ENSO-dependent precipitation extremes in E3SM1 over SE-US and an amplification of P-NW bias. For other models, Niño3.4 index-dependent extremes qualitatively yield weak resolution sensitivity results to that of ELI over those regions (Figure S2 in Supporting Information S1). Also, the pattern of precipitation extremes response to ELI and Niño3.4 index over the US is generally similar to the mean precipitation response in the model simulations (Figures S3 and S4 in Supporting Information S1).

4.3 Moisture Transport

Figure 3 shows the Pacific North America (PNA)-like pattern teleconnection of the mean winter season geopotential height at 500 mb (Z500) when regressed against the standardized ELI for ERA5 data, LR, and HR models. The reduction in Z500 over the North Pacific and over Southern US with increase in ELI is weaker than ERA5 in all models but extends further eastwards into the North Atlantic in some, with implications on moisture influx over land. The increase in Z500 with ELI increase over North America exhibits different biases in different models. Overall, the RMSE (pattern correlation) is reduced (increased) in some HR models as compared to their LR counterparts.

Figure 3

Dependence of geopotential height at 500mb (Z500) and moisture transport response on El Niño Southern Oscillation (ENSO). Regression of Z500 against the seasonal ENSO longitudinal index (ELI) index for (a) ERA5 data (1979–2018) and (b–h) 1950 control model simulations. Hatched areas denote grid points where the regression coefficient is statistically different than zero at the 0.05 global significance level with false discovery rate corrections for multi-testing. Regression of vertically integrated moisture transport (IVT) against the ELI index is also overlayed for simulations for which IVT data was available on CMIP6 archive at the time of writing.

Figure 3 also shows the regression of zonal and meridional vertically integrated moisture transport against the ELI as vectors for ERA5, and for LR and HR models with available data. ERA5 exhibits an increase in the transport of moisture into the SE-US from the Gulf of Mexico and the North Atlantic Ocean with an increase in the ELI coherent with the increase in mean and extreme precipitation. ERA5 also exhibits an increase in poleward moisture transport along the P-NW coastline over the North Pacific with an increase in ELI, with some influx of moisture into California from the Pacific Ocean. This is consistent with the increase in the likelihood of precipitation extremes during El Niño events in the region (Figure 2). This pattern of changes in moisture transport associated with ELI is in accordance with the Z500 response and is also noted in MERRA2 (not shown) as well as ERA-20C (Patricola et al., 2020).

Corresponding to the ELI associated Z500 pattern, most LR and HR models simulate an increase in eastwards transport in the Gulf of Mexico and along the Southern Gulf Coast into Florida, but lack the observed moisture transport into other parts of SE-USA from the Gulf of Mexico and Atlantic ocean with an increase in ELI. Models that simulate ELI dependent moisture influx from the Gulf of Mexico into the mainland (e.g., E3SM1-HR) are associated with a stronger increase in mean and extreme precipitation over the Gulf Coast part of SE-US with increase in ELI, similar to observations. And, those with ELI dependent influx from the Atlantic Ocean (e.g., HadGEM3-GC31-HR) simulate ELI dependence of extremes in the eastern part of SE-US. Two (E3SM1-HR and HadGEM3-GC31-HR) of the four HR models that exhibit lower RMSE for ELI dependence on extreme precipitation (Table S2 in Supporting Information S1) are also indicative of some improvements in moisture transport into the region compared to the LR models. Models that produce strong extremes over the P-NW during La Niña events also simulate a stronger than observed increase (decrease) in moisture transport into that region during La Niña (El Niño) events.

Over California, some LR and HR models (e.g., E3SM1, HadGEM3-GC31) simulate an increase in moisture influx across the coastlines with an increase in ELI, similar to observations, and consistent with the simulated precipitation mean and extremes dependence. This is despite the weaker ELI associated Z500 response over the North Pacific in both LR and HR models compared to ERA5, but likely due to the eastward displaced center of the Z500 response in the models. Nonetheless, most HR models (except E3SM2-HR) are indicative of an improved moisture transport into the region as compared to their LR counterparts. However, the generally weaker than observed simulated precipitation extremes there suggests other processes at work in the region, like atmospheric rivers that operate on shorter time scales there (e.g., Patricola et al., 2020). We do not investigate those here, but plan to study the resolution sensitivity of the simulation of their ELI dependence in the near future. Overall, the above suggests that the improvements or discrepancy in the simulation of ELI dependence of precipitation extremes is related to ELI associated moisture transport simulation.

4.4 Storm Track Activity

We use the standard deviation of the 2–6 days band-pass filtered geopotential height as a metric for storm track activity. Figure 4 shows the regression of this metric against the standardized ELI for the winter season (November–February) for ERA5 data, and for five models where Z500 daily data was available. Positive (negative) values indicate an increase in storm track activity during El-Niño (La-Niña) events. ERA5 exhibits a southwards shift of the storm track, similar to MERRA2 (not shown). There is an increase in storm track activity with the ELI over SE-US, where precipitation extremes also exhibit a similar relationship with ELI. There is also a weak increase in storm track activity with ELI over Western California, although its not statistically significant. ERA5 data also exhibits a statistically significant increase in storm track activity during La Niña events over the P-NW.

Figure 4

Dependence of storm track activity on El Niño Southern Oscillation (ENSO). Regression of the standard deviation of 2–6 days band-pass filtered winter season geopotential height at 500mb against the ENSO longitudinal index index for (a) ERA5 data (1979–2018) and (b–h) 1950 control model simulations. Hatched areas denote grid points where the regression coefficient is statistically different than zero at the 0.05 global significance level with false discovery rate corrections for multi-testing.

Both LR and HR model simulate a southward shift of the storm track activity with increase in ELI, similar to ERA5. However, the increase in storm track activity with increase in ELI is generally weaker than observed in model simulations (except ECMWF-IFS-HR) over both SE-US and California which plausibly results in the weakened simulation of extremes over those regions in most models during El Niño events. Over the SE-US, three (E3SM1-HR, E3SM2-HR, and ECMWF-IFS-HR) of the five HR models show an increase in storm track activity as compared to their LR counterparts. The associated stronger ELI dependence of mean and extreme precipitation (Figure 2) implies a role for storm track activity in the resolution sensitivity of E3SM1 and ECMWF-IFS there. The bias in moisture transport in E3SM2-HR likely counteracts the noted improvement in ELI dependent storm track activity to produce a weaker precipitation response as compared to E3SM2-LR (Figure 2). The weaker storm track activity in EC-Earth3P-HR likely contributes to the weak precipitation response (Figure 2). ELI-dependent vertical updrafts are also stronger in most HR models as compared to their LR counterparts over the region (Figure S5 in Supporting Information S1). HR models simulate stronger vertical updrafts than LR models, owing largely due to fluid continuity (Mahajan et al., 2018, 2022; O’Brien et al., 2016; Rauscher et al., 2016; Yang et al., 2014). Combined with the stronger influx of moisture and storm track activity in the SE-US region with increase in ELI, stronger vertical mass fluxes—like those forced by extra-tropical cyclones—can generate stronger winter time large scale precipitation in HR (e.g., Mahajan et al., 2018).

Some HR models (e.g., E3SMv1, ECMWF-IFS) simulate stronger (but not statistically significant) ELI dependent storm track activity over California compared to their counterpart LR models, and are comparable to ERA5 (Figure 4). The associated stronger ELI-dependent precipitation extremes in those HR models as compared to LR over the region suggests that the stronger storm track activity is also contributing to the improvement, along with stronger moisture flux into the region in HR (e.g., E3SMv1, Figure 3). The ELI associated vertical velocities around the California coastlines are also stronger in HR versions than LR (Figure S5 in Supporting Information S1) in those models, resulting in more upwards mass fluxes causing more precipitation from larger moisture fluxes into the region in the models (Figure 3).

Over the P-NW, the increase in storm track activity during La-Niña events is generally weaker than observed in model simulations with no apparent resolution sensitivity (Figure 4). This implies that the amplified simulated ELI dependent precipitation mean and extremes in some LR and HR model simulations is largely due to the stronger than observed increase in the moisture transport into the region (Figure 3). The stronger response in HR also appears to have an added contribution from enhanced vertical mass fluxes over the complex terrain over the P-NW (Figure S5 in Supporting Information S1), that produce more large scale precipitation, despite similar moisture fluxes into the region.

5 Summary and Discussion

We find that HR models generally improve upon the LR models in the simulation of ENSO teleconnections to precipitation extremes over SE-US and SW-US, where observational data exhibit strong relationships to ENSO. In our evaluation, we also account for ENSO diversity by using the ELI to represent ENSO. We find that HR ESMs simulate a diversity of ENSO as represented by the ELI, but generally with a weaker than observed spatial variability, and are generally similar to their LR counterparts. HR simulated ENSO teleconnections to precipitation extremes over the SE-US is found to improve over their LR counterpart for four models (E3SM1, ECMWF-IFS, HadGEM3-GC31, MPI-ESM1-2), although marginally in some cases and with the RMSE over the region degrading in others. Over the SW-US, resolution sensitivity is clearer, with all models (except one) showing a reduction in RMSE with increase in resolution. The improvements noted in these HR models largely appear to be due the improved simulations of ELI associated mean moisture transport into the two regions and/or ELI dependent extra-tropical cyclone activity there. These results provide support for the development and use of HR models to study seasonal predictability; climate dynamics, variability and change and inform stake holders in climate change impacts.

The HighResMIP HR models were not specifically tuned to improve the quality of their simulations, with only minimal changes made mainly to maintain numerical stability. While E3SM1-HR was tuned, this was largely to improve the mean climatology (Caldwell et al., 2019). Despite this, we observe substantial improvements in the simulation of ENSO teleconnections to extremes in most models at higher resolution compared to their LR counterparts. Targeted tuning efforts focused on improving the simulation of tropical Pacific SSTs and other model features, such as diabatic heating rates, could further increase the fidelity of ENSO teleconnections in HR models and enhance the accuracy of seasonal forecasting of regional water cycles and predictions of ENSO-related changes under climate change.

Acknowledgments

This research was supported as part of the Energy Exascale Earth System Model (E3SM) project, funded by the U.S. Department of Energy (US-DOE) Office of Science's (SC) Office of Biological and Environmental Research. This manuscript has been authored by UT-Battelle, LLC which is supported by SC under Contract DE-AC05-00OR22725. Work at Lawrence Livermore National Laboratory was performed under the auspices of the US-DOE under Contract DE-AC52-07NA27344. This research used the resources of the Oak Ridge and Argonne Leadership Computing Facilities at the Oak Ridge and Argonne National Laboratories, respectively, and the National Energy Research Scientific Computing Center, which are supported by the SC under Contracts DE-AC05-00OR22725, DE-AC02-06CH11357, and DE-AC02-05CH11231, respectively.

Open Research

Data Availability Statement

All simulation output analyzed here is available through the Earth System Grid Federation (ESGF) nodes (https://esgf.llnl.gov/nodes.html) and can be found by searching for the model and field name. CPC Global Unified Precipitation data were provided by the NOAA/OAR/ESRL/PSL, Boulder, Colorado, from their website at https://psl.noaa.gov/data/gridded/data.cpc.globalprecip.html. ERA5 data were obtained from the Copernicus Climate Change Service Climate Data Store (CDS) at https://doi.org/10.24381/cds.f17050d7, https://doi.org/10.24381/cds.6860a573, and https://doi.org/10.24381/cds.bd0915c6.