OVERALL PERFORMANCE MEASURES
4th QUARTER METRIC COMPLETED:
The first three 2023 quarterly reports focused on the U.S. Department of Energy (USDOE) capability to simulate and understand atmospheric rivers and their impacts through the development of the Energy Exascale Earth System Model (E3SM), in conjunction with novel model evaluation and analysis methods. In this fourth quarter report, the focus is on USDOE advances in using a hierarchy of models to simulate and understand the impacts of heat waves on the U.S. West Coast.
It is well documented that heat waves are becoming more severe and more frequent (Masson-Delmotte et al. 2021) and pose risks to human and natural systems, including severe health impacts such as mortality and morbidity due to heat stress in urban heat islands (Vahmani et al. 2019); electric grid outages due to the compounding impacts of increased demand and/or negative impacts of heat on infrastructure (Stone et al. 2021); earlier snowmelt reducing reservoir storage (McEvoy et al. 2021); excess water temperatures affecting power plant cooling and ecosystems (Stillman et al. 2019); reduced agricultural yields due to lower soil moisture and plant mortality (Breshears et al. 2021); and wildfire risks due to reduced fuel moisture content (Pörtner et al. 2022). Increasingly severe heat waves affecting the western U.S. have been of particular concern (Duine et al. 2022; Osman et al. 2022; Chen et al. 2023).
USDOE investments in modeling human-Earth system interactions are increasing our capability to model and better understand the cascading multisectoral and multiscale impacts of heat waves. This fourth 2023 quarterly report documents modeling advances to understand how heat waves affect the electricity grid in the western U.S. In addition to increasing electricity demands through increased use of air conditioning, heat waves can negatively affect generation from renewable resources such as wind, solar, and hydropower (Field et al.; Pryor et al. 2013; Solaun et al. 2019), reduce the efficiency of thermoelectric generation, and cause failure of transmission lines (Dumas et al. 2019). These compounding impacts negatively affect the ability of grid operators to balance supply and demand. As demands increase and/or the most-cost-effective generation resources are compromised, the costs of operating the electricity grid increase. In the worst case, power outages (also called “unserved energy” or “loss of load”) occur (Burillo et al. 2016; Panteli et al. 2015).Traditionally, electricity grid planning has relied on historical, stationary conditions to guide investment and long-range planning decisions to forecast demand and study grid resiliency under extreme events (Jalaei et al. 2020; Yang et al. 2023; De Vilmarest et al. 2022; Craig et al. 2022). With climate change impacts becoming increasingly severe, USDOE foundational science research is informed by the societal need for open‑source, integrated human-Earth system modeling capabilities that address climate non-stationarity and can simulate the full range of demand and supply impacts that can occur during heat waves (Craig et al. 2022).
This report provides clear evidence of USDOE advancements in simulating the impacts of heat waves on the western U.S. electricity grid. This research has developed the first completely open-source, validated, and integrated hierarchy of models for this purpose. Using these tools, we can link future climate projections to models of national-to-global energy economy interactions, electricity demands, power plant siting, renewable resource generation, and grid operations.
As shown in Figure 1, we use a hierarchy of open-source, integrated human-Earth systems models to simulate heat wave impacts on the electricity grid of the western U.S. This modeling framework is unique in its ability to maintain internal consistency in climate and socioeconomic assumptions, while also increasing spatiotemporal and process resolution, from global to local scales. All the models and data used in this framework are open source, and all the models were developed for USDOE. We use this framework to simulate the impacts on electricity demands, solar and wind resources, and electricity system grid stress during a significant July 2018 heat wave that affected most of the western U.S. (NOAA July 2018 Global Climate Report). This heat wave produced record temperatures, requests from the major electric grid operator in California for power conservation, and electricity outages in Arizona due to temporary power shortages (Seattle Times July 25, 2018). We also simulate the impacts of two future heat waves that represent how the July 2018 heat wave would replay after 40 and 80 years of additional warming (in 2058 and 2098, respectively) to understand future grid stress. To isolate and understand the effect of the future, more extreme heat waves, we assume that 2018 electricity infrastructure and population remain fixed.
We provide summaries of each component of the framework below along with references providing more detailed information on each model and data set.
The modeling hierarchy uses global climate and socioeconomic scenarios based on the most recent data sets and assumptions used in the climate science community: the Coupled Model Intercomparison Project, Phase 6 (CMIP6) and the Shared Socioeconomic Pathway (SSP)–Representative Concentration Pathway (RCP) SSP-RCP framework (O’Neill et al. 2020).For the results presented in this fourth 2023 quarterly report, we focus on an RCP8.5 climate scenario (a very warm future climate driven by high greenhouse gas emissions consistent with the SSP5 socioeconomic scenario) so that we can investigate how future heat waves in such a climate could affect electricity demands, renewable resources, and electricity grid operations.
The national-scale modeling approach has two primary components. First, we use a thermodynamic global warming (TGW) approach to simulate future climate on an hourly basis for the contiguous U.S. over the 21st century at 1/8th degree spatial resolution (Jones et al. 2022).The simulations replay the weather events of a 40‑year historical period (1980-2019) under different levels of warming determined from an analysis of the CMIP6 archive (“hotter” versus “cooler” models) to produce future climate over the period 2020‑2099. This storyline, or analog, approach allows us to explore how past extreme events, such as heat waves, could play out under a warmer future. Each event re-occurs twice in the future simulations, first with 40 years of future warming and then with 80 years of future warming. These simulations provide high spatiotemporal and process resolution and physical consistency across climate variables over the U.S. for the entire 21st century.
The national-scale modeling also includes a simulation of 21st-century climate-energy-water-land-economy interactions using version 5.3 of the Global Change Analysis Model, with additional detail in the U.S. (GCAM-USA) (Binsted et al. 2022). GCAM-USA provides internally consistent future scenarios of energy prices and quantities, water consumption, and land use and land cover change for the U.S. and the rest of the world on a five-year time step. The model uses the SSP assumptions regarding population, including state-level projections for the U.S. (Zoraghein et al. 2020), and GDP change as key inputs to its simulation of future demands. In addition, we use the TGW simulations to model the impact of future climate on annual building electricity demands, water availability, and agricultural yields and apply those as inputs to GCAM-USA.
The modeling at the regional-to-local scale includes climate and population impacts on hourly electricity loads (“loads” is the term typically used instead of “demands” in grid operations modeling, so we will adopt it for the rest of this report), power plant siting, renewable resource generation, and electricity system grid operations. The electricity grid operations model takes the hourly loads and renewable resource information as inputs and then determines how the grid would operate and if grid stress occurs (i.e., high operating costs or outages). The local-scale modeling maintains internal consistency with the global and national-scale modeling because it uses the same climate and socioeconomic data and assumptions and the state-scale economic and energy system outcomes from the GCAM-USA simulation are used as boundary conditions.
The Total ELectricity Loads (TELL) model is a machine learning (ML)-based approach based on historical hourly weather and electricity loads. TELL uses population weighting of weather data because weather-driven electricity loads will only occur where people are located. TELL aggregates gridded 1-km population data to the county scale and weights county-scale hourly weather variables that have been derived from the 12-km TGW resolution (Burleyson et al. 2023). TELL combines its ML models with GCAM-USA projections of state-scale annual electricity demand and future hourly climate to produce projections of future hourly loads at the Balancing Authority (BA) scale (McGrath et al. 2022).BAs are the organizations within the U.S. electricity grid that balance supply and loads. The U.S. has three major interconnections (Western, Eastern, and the Electricity Reliability Council of Texas [ERCOT]) that operate independently, each with its own BAs. Figure 2 shows a map of the 28 BAs in the Western Interconnection that are the focus of this quarterly report.
3rd QUARTER METRIC COMPLETED:
Atmospheric rivers (ARs) account for the majority of global water vapor transport from the equator to the poles (Zhu and Newell 1998). These vast plumes, often thousands of kilometers long, carry more water than the world’s terrestrial rivers. Consequently, ARs are important sources of water to communities throughout the western United States. AR-driven precipitation is estimated to contribute to as much as 50% of western United States total annual water resources (Dettinger et al. 2011) and is largely responsible for the interannual variability of mountain snowpack, one of the West’s largest natural reservoirs of water (Siirila-Woodburn et al. 2021). However, ARs can also be hazardous to life and infrastructure, contributing to more than 80% of flood-related damage in the same region (Corringham et al. 2019).
U.S. Department of Energy (USDOE) investments in modeling of the Earth system have greatly improved our capability to simulate and understand ARs and their impacts. Development of the Energy Exascale Earth System Model (E3SM), in conjunction with novel model evaluation and analysis methods, has led to improvements in simulated AR climatology and a capability to simulate and experiment with individual decision-relevant AR events in a global Earth-system model. Investments in land-surface and hydrologic modeling systems, including modern process-based models and models built upon cutting‑edge machine learning (ML) approaches, have also allowed us to better capture total available water resources and estimate flood risk. Furthermore, support for the development of water management models has driven insights into the interplay between water availability and demand in the coupled human-Earth system. Using these tools, novel scientific research has led to a deeper understanding of AR impacts, the sensitivity of those impacts to changes in the large-scale environment, and potential vulnerability to AR hazards. These insights are important to inform practitioners and stakeholders about potential future socioeconomic and infrastructural risks from ARs.
Prior work has shown that in order to resolve landfalling AR-induced impacts on coastal communities, fine horizontal resolution (≤28 km) is needed in Earth system models (Rhoades et al., 2020a, 2020b, 2021, 2023). This is because finer horizontal resolutions allow Earth system model simulations to better resolve features and processes, such as coastal topography, air-sea contrast, rain-snow partitioning, and hydrological responses, which are responsible for shaping AR impacts (Demory et al. 2014). Finer-horizontal-resolution simulations are computationally affordable using regionally refined mesh capabilities of the Energy Exascale Earth System Model (RRM-E3SM). The RRM-E3SM has enabled simulations with local grid spacing as fine as 3km, ensuring land-atmosphere feedbacks related to rapidly varying topography are captured with high fidelity.
In the first 2023 quarterly report, the performance of E3SM in its low-resolution (LR) and high‑resolution (HR) configurations was examined for ARs, with results demonstrating substantially improved performance for the newer HR configuration. In the second 2023 quarterly report, RRM-E3SM was shown to exhibit similarly improved performance to the HR configuration at a fraction of the computational cost. In contrast, this third 2023 quarterly report focuses on AR impacts, which are examined in two ways: first, a suite of impacts-relevant climatological metrics and diagnostics for ARs in the western United States are used to assess E3SM’s ability to simulate AR climatology; and second, this report evaluates RRM-E3SM performance for modeling an AR event that produced widespread impacts throughout California – namely, the historic 1997 New Year’s Flood. For reference, this report uses the European Centre for Medium-Range Weather Forecasts (ECMWF) version 5 land analysis (ERA5-Land) at 9km horizontal grid spacing, U.S. Snow Telemetry (SNOTEL) data, and U.S. Geological Survey (USGS) stream gage data.
This report provides clear evidence of USDOE advancements in simulating AR impacts. At the climatological scale, E3SM performs well at capturing the fraction of precipitation, snowpack, and runoff from ARs, although at higher resolutions E3SM tends to produce more frequent and more intense storms than ERA5. These long-running simulations further include ARs that are similarly extreme to some of the most extreme ARs in the historical record. At the weather time scale, E3SM is demonstrably effective at simulating a high-impact historic AR event and, in conjunction with a suite of DOE-supported water management and hydrologic models, closely matched streamflow and reservoir inflow, and flagged areas of flood risk associated with the event. Meteorological inputs from RRM-E3SM even yield streamflow forecasts that better matched observations than a popular gridded meteorological product. All DOE‑supported water models show clear improvement over presently employed process-based and machine learning-based alternatives.
This report examines simulated AR impacts on both climatological and weather (i.e., event-focused) timescales. Notably, information derived from both analyses have utility to stakeholders for evaluating infrastructure vulnerabilities and for adaptation planning.
At climatological timescales, we analyze results from E3SMv2 in its low-resolution configuration (E3SM-LR, 110km grid spacing) and a regionally refined configuration (RRM-E3SM, 14km grid spacing over the contiguous U.S.) against observations from the western U.S. SNOTEL station network (Hufkens 2022) and reanalysis data from ERA5-Land (Muñoz 2019). Simulations cover the near-term historical period of 1986-2014. The E3SM simulations are run under Atmospheric Model Intercomparison Project (AMIP) protocols with prescribed sea surface temperatures and sea-ice extents, and fully coupled atmosphere (EAM) and land surface (ELM) models.
For the climatological study, we investigate to what degree our models capture AR impacts on water resource availability in the western U.S., including statistics of total precipitation and snow water equivalent. This study also explores ARs under the Ralph et al. 2019 AR category scale, which categorizes ARs into beneficial or hazardous to water resource management using integrated vapor transport and a USDOE-supported AR tracking algorithm, TempestExtremes (Ullrich et al. 2021). Finally, we examine the effect of ARs on runoff, snow cover fraction, and snow water equivalent: metrics relevant to both water resource management, specifically water supply reliability, and flood risk exposure.
At the weather timescale, we analyze results from E3SMv2 hindcast simulations of the 1997 New Year’s Flood, which was driven by a strong, persistent AR. Confirming the ability of a model to simulate specific historical weather events is important to build confidence in the model’s ability to represent more general events in a climatological context. Meteorological data from (1) RRM-E3SM, (2) E3SM at uniform coarse resolution and (3) a gridded meteorological product are subsequently fed into three impacts models: the Water Evaluation and Planning (WEAP) model, which simulates hydraulic connectivity between elements of the water management system, the Pennsylvania State University long short-term memory (LSTM) model, which is a pure data-driven ML model for streamflow, and a cutting‑edge, ML-based differentiable parameter learning (dPL) model. The LSTM and dPL models are subsequently used for simulating reservoir inflows during the flood event. These models are then evaluated against USGS streamflow observations during the flood event using conventional metrics such as Nash-Sutcliffe Efficiency and differences between peak flow.
2nd QUARTER METRIC COMPLETED:
Improving Simulations of Atmospheric Rivers and Heat Waves Using a Hierarchy of Models for High-Resolution Modeling
High-resolution climate data products are needed to develop a comprehensive understanding of weather and climate hazards in the Western U.S., both as they occurred historically and as they may occur in the future. Investments by the U.S. Department of Energy in global and regional climate modeling capabilities have provided new means to simulate these hazards and quantify their characteristics. In particular, the regionally refined modeling (RRM) capability in the Energy Exascale Earth System Model (E3SM) enables better resolution of the rough topography of the U.S. West, which in turn allows E3SM to better capture the impact of topography on local meteorology and associated extremes. The dynamical downscaling method (DDM), a technique that uses a secondary regional climate model forced at its boundaries, is another means to provide locally relevant climate data and better simulate extreme weather. As demonstrated in this report, these two simulation strategies can also be used together for cross‑validation or to assess regional structural uncertainties via a hierarchy of models.
Atmospheric rivers (ARs) and heat waves (HWs) are two prominent meteorological features that are sensitive to changes in the broader Earth system. ARs are long, narrow bands of atmospheric moisture transport that are responsible for the majority of extreme precipitation events in the western U.S. Consequently, an understanding of ARs is important for assessing flood risk and estimating water availability. HWs are periods of sustained high near-surface temperatures that pose significant risk for human populations, ecosystems, and agriculture. Future projections for both extremes, drawn from climate models, are necessary for quantifying risk and informing relevant practitioners.
In the previous quarterly report, the performance of E3SMv1 in its low-resolution (LR) and high‑resolution (HR) configurations was examined for ARs and HWs in the western U.S. In this report, we examine model performance for ARs and HWs in the western U.S. when simulations are conducted using E3SM version 2 in its regionally refined mode, and when simulations are performed using the dynamical downscaling method with the Weather Research and Forecasting (WRF) model. Results are compared with the LR configuration of E3SMv2 (LRv2) and against the European Center for Medium‑Range Weather Forecasts (ECMWF) version 5 reanalysis (ERA5). Both RRM and DDM enable higher resolution over regions of interest, and hence are suitable for studies on regional weather and climate.
The evaluation of ARs and HWs in the RRM, DDM, and LRv2 simulations reveals that several key metrics show improvement when the model resolution is increased from 1 degree to 0.25 degrees, whether through regional refinement or dynamical downscaling. AR frequency and landfalling AR precipitation are both substantially improved in RRM and DDM simulations. For HWs the RRM configuration exhibits modest improvements over the LRv2 configuration, in large part due to better resolution of topography in the Western U.S. HW improvements in the DDM configuration are largely limited to HW intensity and duration, with most improvement concentrated along the coast. This difference is likely due to a cold bias in the DDM simulation, which could be ameliorated through the use of a different land model. Despite the improvements mentioned above, other metrics show little variation across the three simulations. Some of these errors may be related to biases within ERA5, which could merit further study.
This report analyzes results from E3SMv2 in its RRM configuration and the WRF model, which is run using the dynamical downscaling method (DDM) and forced by E3SMv2-RRM data. E3SMv2-RRM is a global fully coupled modeling system that supports spatially variable grid spacing. The WRF model, in conjunction with NoahMP land surface model, is a regional atmosphere-land model that requires specification of meteorological fields along the domain boundary, and sea-surface temperatures at the atmosphere-ocean interface. For this study, we use the North America x4 grid, which has grid spacing of approximately 0.25 degree over and around the North American continent, and 1-degree grid spacing elsewhere. The chosen WRF grid has 0.25-degree grid spacing that extends from the western United States across much of the northeastern Pacific. The RRM and DDM grids used in this study are depicted in Figure 1. For comparison, an E3SMv2 simulation is also performed with 1-degree grid spacing (i.e., low resolution) over the entire globe, referred to as LRv2. This simulation is configured identically to the RRM simulation.
E3SMv2 simulations (both RRM and LRv2) are performed with active atmosphere, land, ocean, sea‑ice, and river components. The non-hydrostatic, compressible atmosphere model is described by Rasch et al. (2019) and makes use of a 72-level spectral element dynamical core to solve the primitive equations (Dennis et al. 2012, Taylor et al. 2020). Parameterized processes include deep convection (Neale et al. 2008; Richter and Rasch 2008, Zhang and McFarlane 1995); macrophysics, turbulence, and shallow convection (Golaz et al. 2002, Larson 2017, Larson and Golaz 2005); microphysics (Gettelman and Morrison 2015, Gettelman et al. 2015); aerosol treatment (Liu et al. 2016, Wang et al. 2020); and radiative transfer (Iacono et al. 2008, Mlawer et al. 1997). The ocean and sea-ice models use the Model for Prediction Across Scales (Petersen et al. 2019, Ringler et al. 2013), and a mesoscale eddy parameterization (Gent and Mcwilliams 1990) is used for E3SM-LR only. The land model is based on the Community Land Model version 4.5 (Oleson et al. 2013), run with satellite phenology. The river model is the Model for Scale Adaptive River Transport (Li et al. 2013, 2015), using runoff simulated by the land model to compute channel velocity, channel water depth, and water surface area.
The WRFv3.9.1 simulation, in the configuration used for this report, is performed with active atmosphere, land-surface, and lake components (Skamarock et al. 2008). The atmosphere is non‑hydrostatic, compressible, and was configured with 36 vertical levels, spectral nudging of wavelengths greater than ~700km, and a model top at 20hPa. Forcing data for the atmosphere (lateral boundary conditions and input for spectral nudging) comes from the E3SMv2-RRM simulation. The atmosphere grid is configured to have an approximate grid spacing of 28km. In this configuration, parameterized processes include deep convection (Kain 2004), microphysics (Hong et al. 2004), longwave (Mlawer et al. 1997) and shortwave (Chou and Suarez 1994) radiative transfer, and surface and planetary boundary-layer processes (Janjic 1994). The land-surface model is NoahMP, using TOPMODEL for runoff and groundwater, a single-layer, three-category urban canopy model, and the Community Land Model version 4.5 lake model (Niu et al. 2011, Subin et al. 2012). Ocean surface lower boundary conditions were taken from the E3SM-RRM run, and sea ice was set to form when sea surface temperatures reached a threshold of 272K or less.
For reference, the ERA5 data set (Hersbach et al. 2020) is used. ERA5 is the fifth-generation ECMWF reanalysis product, produced through assimilation of observations (e.g., aircraft, in situ, and
To ensure our analysis is sufficiently robust, the RRM and LRv2 simulations are conducted over the period 1985-2014 (30 years total). The DDM simulations use RRM data as forcing and also begin in 1985, but the first year is discarded, leaving 29 years for analysis.
TempestExtremes, a general-purpose and freely available feature detection and characterization software package, is used throughout this report to identify and characterize atmospheric rivers and heat waves (Ullrich et al. 2021).
Two metrics are used in this report for assessing agreement between model and reference. First, root mean square error (RMSE) is defined as
where oi refers to the observed field in grid cell i, pi refers to the model prediction, and Ai refers to the cell area. Second, index of agreement (IoA) complements RMSE, and is defined as
where the overline denotes the arithmetic mean. Values of IoA range from 0 to 1, with values closer to 1 indicating better agreement.
1st QUARTER METRIC COMPLETED:
The Western US is subject to many natural hazards, such as floods, droughts, heat waves, and fires. Continued investments in modeling efforts from the Department of Energy (DOE) have enabled examination of some of these hazards with the Energy Exascale Earth System Model (E3SM) at high spatial resolution – 25 km grid spacing compared to the standard, or low resolution, 100 km grid spacing.
Atmospheric Rivers (ARs) are filamentary bands of increased atmospheric moisture transport. Nearly all poleward moisture transport out of the subtropics occurs through ARs (Zhu & Newell, 1998). ARs have been shown to be a valuable source of freshwater supplies, while also having the capacity to bring about flooding, landslides, and wind damage, making them a key component of the hydrologic cycle to understand. Models like E3SM are a critical tool for predicting variabilities and changes in ARs, but uncertainties remain regarding which aspects of modeling, such as horizontal resolution, may improve the representation of ARs.
Heat waves (HWs) are periods of extreme near-surface temperatures that last on the order of days. As the global mean temperature has increased, so too have the frequency and intensity of HWs (Seneviratne, 2021). HWs are especially concerning, owing to their links to increased morbidity (Ye et al., 2012), only to be exacerbated by increasing average temperatures under a warming climate (Huang et al., 2011). Like ARs, models have biases in representing these events.
The DOE-sponsored TempestExtremes software product (Ullrich & Zarzycki, 2017; Ullrich et al., 2021) allows for tracking of ARs and HWs as objects. TempestExtremes has been used to track ARs repeatedly in the past (e.g., Liu et al., 2022; Rhoades et al., 2021). HWs, to date, have primarily been examined using timeseries analysis performed at individual locations. Here, we leverage TempestExtremes recently developed object-tracking capability within a new methodological framework to examine HWs as widespread events instead of using a traditional point-wise evaluation.
In this document, we evaluate the performance of E3SM in simulating ARs impacting the US West Coast and HWs in the Western US. As part of the Phase 1 water cycle simulation campaign, E3SMv1 was used to perform simulations at high resolution (~25 km grid spacing) and low resolution (~100 km grid spacing) to evaluate the impacts of model resolution on representing water cycle processes. Both ARs and HWs are thus examined at high and low resolutions and the differences are compared and summarized herein. Results show the impact of increased resolution on ARs is mixed, with some features showing improvements (capturing wind and precipitation extremes) while others showing degradation (frequency, position, and shape). Increased resolution improves HW intensity and duration, while HW frequency is rather insensitive to the model resolutions examined.
This report analyzes fully-coupled simulations from the E3SMv1 high-resolution (E3SM-HR) and low-resolution (E3SM-LR) experiments, as detailed in Caldwell et al. (2019) following a similar protocol of the Coupled Model Intercomparison Project Phase 6 (CMIP6) HighResMIP experiments (Haarsma et al. 2016). Both E3SM-HR and E3SM-LR have active atmosphere, land, ocean, sea-ice, and river components. The atmosphere model is described by Rasch et al. (2019) and makes use of a 72-level spectral element dynamical core to solve the primitive equations (Dennis et al., 2012). Parameterized processes include deep convection (Neale et al., 2008; Richter & Rasch, 2008; Zhang & McFarlane, 1995); macrophysics, turbulence, and shallow convection (Golaz et al., 2002; Larson, 2017; Larson & Golaz, 2005); microphysics (Gettelman & Morrison, 2015; Gettelman et al., 2015); aerosol treatment (Liu et al., 2016; Wang et al., 2020); and radiative transfer (Iacono et al., 2008; Mlawer et al., 1997). The ocean and sea-ice models use the Model for Prediction Across Scales (Petersen et al., 2019; Ringler et al., 2013), and a mesoscale eddy parameterization (Gent & Mcwilliams, 1990) is used for E3SM-LR only. The land model is based on the Community Land Model version 4.5 (Oleson et al., 2013), run with satellite phenology. The river model is the Model for Scale Adaptive River Transport (Li et al., 2013; Li et al., 2015), using runoff simulated by the land model to compute channel velocity, channel water depth, and water surface area and simulate streamflow.
E3SM-HR and E3SM-LR use an identical set of physical parameters but differ in the horizontal grid-spacing of each model component. Each component has approximately a factor of four grid refinement. The atmosphere and land components share a grid with spacing roughly equal to 25 km for E3SM-HR and 110 km for E3SM-LR. The ocean and sea-ice models also share a grid. The ocean/sea-ice grid varies in resolution globally, spanning 8-16 km for E3SM-HR and 30-60 km for E3SM-LR.
To track ARs and HWs, we use the TempestExtremes software (Ullrich & Zarzycki, 2017). ARs are tracked as features with large (area exceeding 850000 km2), localized vertically-integrated water vapor transport, that do not also fit the criteria for Tropical Cyclones (warm core low pressure systems). HWs are tracked as features whose daily maximum temperature exceeds the 95th percentile temperature for that location and during the May through September period for at least two consecutive days. In both cases, metrics like AR and HW frequency are sensitive to the specific feature definition chosen (e.g., Shields et al., 2018), which has implications for evaluating model performance in simulating these extreme events (Leung et al., 2022). Investigating the impacts of such sensitivity on our model evaluation results is beyond the scope of this study but should be explored in the future.
After tracking of ARs and HWs in the E3SM HR and LR simulations and in observations, different aspects of these extreme events are analyzed and compared between the HR and LR simulations and the observations.