Volume 10, Issue 12 e2022EF002967
Research Article
Open Access

Drivers and Mechanisms of the 2021 Pacific Northwest Heatwave

D. L. Schumacher

Corresponding Author

D. L. Schumacher

Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Correspondence to:

D. L. Schumacher,

[email protected]

Contribution: Conceptualization, Methodology, Software, Formal analysis, ​Investigation, Writing - original draft, Visualization

Search for more papers by this author
M. Hauser

M. Hauser

Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Contribution: Conceptualization, Software, ​Investigation, Writing - review & editing, Supervision

Search for more papers by this author
S. I. Seneviratne

S. I. Seneviratne

Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Contribution: Conceptualization, ​Investigation, Writing - review & editing, Supervision, Funding acquisition

Search for more papers by this author
First published: 25 November 2022
Citations: 13

Abstract

In late June 2021, western North America, and in particular the Pacific Northwest experienced a heatwave with temperatures usually only encountered in hot desert climates. Using a blend of reanalysis data and Earth System Model (ESM) simulations, we disentangle the physical drivers underlying this exceptional event. Our analysis highlights the role of the anticyclonic circulation aloft, which converted previously gained potential energy—some of which by intense latent heating thousands of kilometers upwind over the North Pacific—back into sensible heat through subsidence. We demonstrate that this upwind latent heat release did not only result in a hot troposphere above the heatwave region, but also contributed directly to escalating near-surface temperatures. Facilitated by the mountainous terrain and dry soils in the region, deep atmospheric boundary layers were established over the course of several days, connecting the air close to Earth's surface to a massive heat reservoir many kilometers above. Anomalous soil moisture acted to raise the heatwave temperatures by 3°C in a large region during the peak of the event, with local anomalies exceeding 5°C. Overall, we conclude that this heatwave was the outcome of an intricate interplay between dynamic and thermodynamic processes. ESM experiments suggest that the same large-scale atmospheric circulation fueled by enhanced thermodynamic drivers, such as more available moisture for condensation upwind, could enable even more extreme near-surface temperatures, in particular in a warmer climate.

Key Points

  • A strong “Omega Block” enabled the heatwave, yet near-surface air temperatures were more extreme than suggested by the large-scale flow

  • Sinking air aloft previously experienced strong latent heating over the North Pacific, contributing to the high near-surface temperatures

  • Deep atmospheric boundary layers brought the extreme heat down to the surface, with local soil moisture effects in excess of 5°C

Plain Language Summary

In late June 2021, western North America, and in particular the Pacific Northwest, experienced temperatures normally encountered in hot deserts. Our analysis highlights the role of the anticyclonic circulation aloft, whose downward spiraling air masses converted previously gained potential energy back into sensible heat. We show that in addition to heating through sinking, the air was previously heated by condensation in ascending air streams thousands of kilometers upwind, over the North Pacific. Together, these processes fostered a massive heat reservoir above the heatwave region, which contributed to escalating near-surface temperatures through strong vertical mixing. Dry soils in the region intensified surface heating, boosting maximum temperatures in excess of 5°C. Overall, we conclude that this heatwave was the outcome of an intricate interplay of the atmospheric flow and processes such as condensational and surface heating, further exacerbated by human-induced background warming. Our experiments suggest that if fueled by more available moisture for condensation upwind, the same large-scale atmospheric circulation could enable even more extreme near-surface temperatures.

1 Introduction

In the summer of 2021, a heatwave eclipsed existing temperature records in the Pacific Northwest (PNW). The extreme heat culminated in late June in the most densely populated coastal areas, where several hundred excess deaths were reported (BMJ, 2021). Epitomized by temperatures of nearly 50°C at about 50°N (e.g., Samenow & Livingston, 2021), and considered to be one of the most extreme weather events on record (Thompson et al., 2022), the sheer intensity of this heatwave mandates a detailed understanding of its causes. From an atmospheric perspective, the temperature escalations were fostered by a strong “Omega Block” in the region that slowly shifted eastwards. Atmospheric blocks are known to enable intense heatwaves in the midlatitudes, as their slowly subsiding—and hence warming—air masses shield the region underneath from storms and provide clear skies (Rex, 1950; Sousa et al., 2018; Trigo et al., 2005; Xoplaki et al., 2003). A rapid attribution study suggests that an event of this magntitude has become at least 150 times more likely due to anthropogenic climate change, and that an event with the same occurrence of probability in a preindustrial climate would have been 2 °C less hot (Philip et al., 2021). This analysis and other studies have provided evidence of human influence on both the occurrence and magnitude of extreme weather events (e.g., Hauser et al., 2016; Otto et al., 2012; Russo et al., 2015; Stott et al., 20042016; Trenberth et al., 2015; van Oldenborgh et al., 2021; Wehrli et al., 2019, 2022). Even in consideration of this human-induced aggravation, it remains unclear how exactly the 2021 PNW heatwave unfolded with such unprecedented magnitude, breaking temperature records by several degrees Celsius (Philip et al., 2021).

During the devastating 2003 and 2010 heatwaves in Europe (Barriopedro et al., 2011; Robine et al., 2008), a multiday heat accumulation took place; fueled by dry soils and hence strong surface heating, the atmospheric boundary layer (ABL) grew deeper and deeper and thereby entrained hot air from aloft (Miralles et al., 2014). For these and other compound hot and dry events, the relevance of both local and upwind drought conditions has already been highlighted (e.g., Fischer et al., 2007; Hauser et al., 2016; Schumacher et al., 2019; Seneviratne et al., 2013; Zampieri et al., 2009), showing that they enable stronger land-atmosphere feedbacks that can intensify hot conditions (e.g., Berg et al., 2014; Hirschi et al., 2011; Mueller & Seneviratne, 2012; Seneviratne et al., 200620102013; Stéfanon et al., 2014; Vogel et al., 2018; Zaitchik et al., 2006). A large fraction of the PNW had unusually dry soils already in late spring (Ansah & Walsh, 2021; Bumbaco et al., 2022), raising the question of what role soil moisture played during the heatwave. Further, given that the potent block setting the stage for extreme heat below remains a marked feature of a meandering circumglobal jet stream (e.g., Kornhuber et al., 2020; Rossby, 1939; Röthlisberger et al., 2016), an open question is whether remote processes and interactions, e.g., instigated by anomalous sea surface temperatures (Feudale & Shukla, 2011; Wang et al., 2014), may have contributed to the 2021 PNW event. In fact, a recent study points to a developing cyclone south of Alaska, whose heat released by condensation in ascending air masses strengthened the Omega Block in the PNW in late June 2021 (Neal et al., 2022). While the positive anomaly in geopotential height at 500 hPa (Z500) was unprecedented in that region, this distinctive anticyclonic circulation pattern alone does not seem to explain the temperature escalations (Philip et al., 2021). Moreover, existing assessments of the human-caused exacerbation of this heatwave largely rely on statistical relationships derived from observations or based on physical simulations of the Earth System of the same region. However, it is not a priori clear whether Earth System Models (ESM) simulate the same—potentially nonlinear (Philip et al., 2021)—processes and interactions as those responsible for the 2021 PNW heatwave.

To address these questions, factorial ESM experiments are used here, building upon the approach of Wehrli et al. (20182019). In this approach, different components of the Community Earth System Model (CESM; Hurrell et al., 2013) are either constrained or calculated interactively. Crucially, this framework includes simulations where the horizontal wind is nudged toward reanalysis data (Wehrli et al., 2018), so that the actual atmospheric circulation during the 2021 PNW heatwave is reproduced. This framework provides insights into the possibly unique set of processes and interactions underlying this extraordinarily hot event. We begin our investigation with an assessment of the suspected drivers of the event, which can be conceptually separated into “dynamic” and “thermodynamic” (see, e.g., Wehrli et al., 2018 for further background). Here, we refer to the atmospheric circulation as the “dynamic” component, whereas anomalous land and ocean surface states, as well as the background warming of the atmosphere and ocean since preindustrial times, are related to “thermodynamic” contributions. Building on this, we compare the dynamics of the 2021 PNW heatwave to observed and simulated events and highlight the occurrence of elevated temperatures throughout the entire troposphere. Next, using the atmospheric trajectory model TRACMASS (Aldama-Campino et al., 2020; Döös, 1995), driven with ERA5 data, we unravel the upwind heat budgets of anomalously hot air that was incorporated into the anticyclone, and isolate the effect of latent heating. It has already been shown that prior to the heatwave, large amounts of moisture were transported across the North Pacific, most of which precipitated in the process and thereby provided heat to the troposphere in the PNW through condensation (Mo et al., 2022). The implications of this trans-Pacific moisture transport for subseasonal predictability have also been investigated (Lin et al., 2022). Note that existing studies of upwind latent heat release in the warm season focus primarily on the formation and maintenance of blocks (Neal et al., 2022; Pfahl et al., 2015; Steinfeld & Pfahl, 2019; Steinfeld et al., 2020; Zschenderlein et al., 2020), and hence on dynamic aspects. Here, we follow a different approach and instead quantify the direct effect of upwind latent heating on downwind heatwave temperatures for the given large-scale circulation. While the causal link between upwind heat release and downwind state of the troposphere is established by the flow of air itself as well as processes occurring en route, which we approximate with backward trajectories, it is not a priori clear to what extent near-surface temperatures are affected.

2 Materials and Methods

2.1 Defining the Heatwave Region

To unravel the drivers of the 2021 PNW heatwave, we employ a heatwave definition designed to specifically target the occurrence of widespread extreme heat in summer (June–August). Daily maximum ERA5 (Hersbach et al., 2020) 2-m temperatures are evaluated on a 0.5° × 0.5° grid over northwestern North America for 40°–65°N and 100°–160°W. We define the heatwave region as all grid cells whose daily maximum temperatures (TX) correspond to a standardized anomaly of at least 3σ with respect to (w.r.t.) 1982–2008. A 15-day moving window is applied before calculating daily TX climatologies and standard deviations, using 1982–2008 as for other analyses here. We then determine the day with the largest heatwave area, 30 June 2021. Our analysis is based on UTC days, so 30 June corresponds to the local period of 29 June 5 p.m. to 30 June 5 p.m. Pacific time. In addition, the most extreme temperatures progressed eastward, with all-time temperature records at the comparatively densely populated coast with the metropolitan areas of Portland, Seattle, and Vancouver (CIESIN, 2018) already on 28 June according to Pacific Standard Time. We rely on two different TXʹ thresholds in this study to delineate a large 3σ region and a smaller area based on 5σ, encompassing even more extreme temperatures. Note that “heatwave region” always refers to the 3σ-threshold area in the following.

2.1.1 Defining Northwestern North American Hot Events

To enable a comparison of the 2021 PNW heatwave to other events in the Pacific Northwest and surroundings, we repeat the procedure described above for all summers from 1982 to 2021 and iteratively select the 150 events with the largest area of extreme heat (TXʹ ≥ 3σ). It is ensured that events are separated by at least 7 days, allowing us to capture the peaks of the 2021 PNW heatwave—ranked first—and 149 other heatwaves in northwestern North America. We then determine the respective peak locations by identifying the grid cell with the strongest standardized TX anomaly. This approach yields a majority of events within or close to inhabited areas situated either along the coast of the domain or east of the Rocky Mountains, and several events in the vicinity of the 2021 PNW heatwave peak location of 55°N, 118°W. All hot event peak locations are illustrated in Figure S1 in Supporting Information S1.

2.2 The CESM and Disentangling Framework

This study is based on global ESM simulations performed with CESM 1.2.2 spanning the period from 1979 to 2021 and employs the method introduced by Wehrli et al. (20182019). The disentangling framework rests on a set of simulations with interactively calculated or constrained components, namely, the atmosphere (Community Atmosphere Model version 5.3, CAM5), land (Community Land Model version 4, CLM4), and ocean surface (always constrained). With this, we estimate the influence of physical drivers such as atmospheric circulation, soil moisture, or SST anomalies, as well as the long-term human imprint on the Earth system. The framework, now driven with ERA5 instead of ERA-Interim data where applicable, is described in detail by Wehrli et al. (20182019), and an overview of how the different CESM components were forced here is provided in Table 1. In essence, the dynamic and thermodynamic contributions are determined by temperature anomaly differences in the heatwave region between simulations with differently forced components (e.g., actual versus climatological soil moisture). Note that the first three simulation years until 1981 are discarded to allow for model spinup. All anomalies presented in this study are calculated w.r.t. the climatological baseline period, 1982–2008, and Figure 1 depicts the disentangling framework used here with all employed CESM configurations.

Table 1. Overview of the Different Model Components Used in the Disentangling Framework
SST and sea ice Atmosphere Soil moisture
Reference (“actual”) oF aF sF
Climatological (1982–2008) oF aF | aI sC
Interactive oF aI sI
  • Note. The capitalized letter in each label (e.g., oF) indicates whether the respective component is constrained—or “forced”—toward actual reference data (F), the 1982–2008 climatology (C), or calculated interactively (I). Selected combinations of these components are used here to unravel the roles of dynamic and thermodynamic drivers (Figure 1). Italic font in gray indicates that a different component than suggested by the corresponding row is used; e.g., there is no simulations with a “climatological” atmosphere, and the atmosphere in simulations with climatological land and ocean surface states is thus either calculated interactively or nudged toward actual winds. The surface ocean is always constrained toward observations (oF), and is hence omitted in the overview of the disentangling framework in Figure 1.
Details are in the caption following the image

Disentangling framework to estimate physical drivers of the heat wave. The arrows denote contributions that are part of the additive disentangling framework presented by Wehrli et al. (2019), and indicate the pathway used to estimate the contributions of dynamic and thermodynamic drivers. For example, we start with the fully constrained simulation (aFsF) and compare it to a simulation with climatological soil moisture (aFsC); their temperature anomaly difference indicates the effect of anomalous soil moisture (brown box to the left). Next, by comparing the simulation with climatological soil moisture and constrained atmosphere to the ensemble mean of simulations where winds and soil moisture are calculated interactively, we obtain the circulation effect (purple box to the right), and so on. Note that contributions estimated based solely on simulations with interactive atmosphere are considered to be “generic” estimates (dashed border), contrary to “event-specific” effects estimated with a constrained atmosphere (toward actual winds according to ERA5). All these heatwave contributions are displayed for the 2021 Pacific Northwest event in Figure 2c.

Following Wehrli et al. (2018), SSTs and sea ice are constrained in all simulations using transient monthly observations, consisting of a merge of the Hadley Centre sea ice and SST data set version 1 and version 2 of the National Oceanic and Atmospheric Administration (NOAA) weekly optimum interpolation (OI) SST analysis (Hurrell et al., 2008). Note that since the ocean is always forced in this study, only (one-way) interactions rather than feedbacks between surface ocean state and large-scale atmospheric circulation are captured for interactively calculated atmospheres. Nevertheless, the winds in simulations with nudged circulation toward reanalysis data essentially represent the interplay of actual atmospheric internal variability and, e.g., ocean-atmosphere feedbacks. The employed solar and greenhouse gas (GHG) forcings, as well as atmospheric chemistry, aerosols, and land-use change, are largely based on historical data. We refer to Wehrli et al. (2018) for details, and note that global CO2, CH4, and N2O concentration observations were updated to cover the entire simulation period until the end of 2021.

2.2.1 Nudging of the Atmospheric Circulation

In this study, the atmospheric circulation is either calculated interactively, or “nudged” toward reference data for each model level. For the nudging, CAM5's prognostic horizontal wind equations are extended by a nudging term that relaxes the wind field toward the reference, i.e., the next 6-hourly reanalysis step for every model time step (30 min). The same vertical nudging profile as introduced by Wehrli et al. (2018) is used for the troposphere (Figure S2 in Supporting Information S1), in which only the zonal and meridional winds above roughly 700 hPa are forced toward the reference, here consisting of 6-hourly ERA5 data regridded to CESM's 0.9° × 1.25° horizontal resolution. Note that even though vertical winds are never nudged, mass conservation as expressed in the continuity equation ensures that the grid-scale vertical motion, especially above 700 hPa, also approaches the reference data set. Due to storage considerations, only the troposphere and lower stratosphere are nudged toward reanalysis data, while the four uppermost model levels—roughly corresponding to 24.6, 14.3, 7.6, and 3.6 hPa pressure levels—are calculated interactively. Note that the interactive calculation of upper model levels does not affect tropospheric winds.

We also perform two additional CESM simulations with slightly modified vertical nudging profiles. As visualized in Figure S2 in Supporting Information S1, the gradual increase from 0 to 1 of the vertical nudging coefficient, i.e., from no to full nudging, is centered on model level 21, which corresponds to 691 hPa for grid columns with a surface pressure of 1,000 hPa. We adjust the vertical nudging coefficient profile so that this transition occurs further away from the surface, at model level 19 and 17, corresponding to about 524 and 379 hPa, respectively. Due to computational constraints, we only simulate 2021 for these two experiments, and rely on our default simulation for the climatology.

2.2.2 Soil Moisture

To assess the impact of the land surface state on the 2021 PNW heatwave, soil moisture is either constrained or calculated interactively in our disentangling framework, following the approach of Wehrli et al. (2019). We constrain “actual” or “climatological” (1982–2008) soil moisture values, but never force CESM with ERA5 soil moisture: a direct transfer of such a highly model-dependent quantity between different models should be avoided (Koster et al., 2009). Instead, we first use meteorological input from ERA5 to drive the land surface model (CLM4) from CESM in offline mode. This produces an ERA5-driven soil moisture data set that is consistent with the fully coupled model (CESM) in terms of, e.g., soil levels and texture (Koster et al., 2009). The technical implementation of the soil moisture prescription in CLM4 was performed and assessed by Hauser et al. (2017). As in Wehrli et al. (2019), soil moisture is only constrained when the soil is not frozen. Otherwise, soil moisture is calculated interactively to avoid artificial ice generation and ensuing unrealistic ground heat fluxes (see Hauser et al. (2017) for details).

2.3 Composite Analysis of Anticyclones Associated With Hot Events

2.3.1 Constrained Circulation

To better understand the role of atmospheric dynamics during the 2021 PNW heatwave, we examine it together with 149 other northwestern North American hot events (Section 2.1.1). While anticyclones enabling extreme near-surface temperatures are often quasi-stationary (e.g., Kornhuber et al., 2020; Lau & Nath, 2014; Petoukhov et al., 2016; Yao et al., 2017), the block over the PNW slowly shifted eastward during the 2021 PNW heatwave. We thus employ an approach that is not bound to a static region, and instead track anticyclones in our fully constrained base CESM simulation (Section 2.2). Anticyclones are detected using Z500: first, a 15-day running mean of Z500 is calculated for every terrestrial grid cell in the selected latitudinal band including a ±10° buffer added to the meridional extent of our northwestern North American domain (30°–75°N), followed by the calculation of standardized anomalies w.r.t. 1982–2021 (Z500ʹ). Then, a rectangular domain with three sizes—4° × 4°, 10° × 10°, and 20° × 20°—is centered on the grid cell with the highest standardized Z500ʹ near the respective hot event peak location, i.e., within 5° in each horizontal direction. We only show results for 10° × 10° domains here, but this choice does not affect our conclusions. Iterative day-by-day tracking is initiated on the hot event's peak day and at the determined nearby Z500ʹ peak location, with anticyclone locations on the previous and subsequent day being assigned to Z500ʹ maxima within 5° horizontally. Going further back or forward in time, the respective, previously identified anticyclone locations are used, so that anticyclones are always tracked from day-to-day and with the same horizontal constraint. While some anticyclones may not be tracked properly due to this horizontal movement restriction, the focus here is on slowly moving anticyclones. The tracking is performed anywhere and without additional criteria for the 7 days before the peak and the 3 days after. The rationale is that for the 2021 PNW heatwave, temperatures plateaued only after the anticyclone had reached its maximum intensity (Philip et al., 2021). Finally, once all anticyclones have been identified and tracked backward and forward in time, the corresponding area averages of several variables—such as temperatures at different pressure levels—are logged to enable a composite analysis. Since Z500 increases for a warming atmosphere (e.g., Sánchez-Benítez et al., 2018), we have repeated the entire analysis for linearly detrended Z500, and found that the decision whether to detrend does not noticeably affect our results and conclusions. As detrending is not necessary for the simulations with interactive atmosphere (see details below), we only show results obtained without prior linear trend removal.

2.3.2 Interactive Circulation

After comparing the dynamics of the 2021 PNW heatwave to other historic hot events, we repeat the analysis for simulations with fully interactive atmospheres. Due to storage constraints, all the required data are only available from 2009 onward, and hence the analysis is restricted to 2009–2021, which also serves as the climatological base period. Instead of tracking anticyclones near hot events, we begin by calculating standardized Z500ʹ as described in Section 2.3.1, and then, within the same northwestern North American domain as employed for the definition of hot extremes, identify events for which Z500ʹ locally exceeds 2σ. The remainder of the tracking procedure is identical; the sole difference consists of the fact that tracking is initiated directly for strong anticyclones, as the historical hot events are not captured by these simulations with interactive atmosphere in the first place. A temporal overlap within the 11-day tracking periods is prevented by selecting the strongest anticyclone according to Z500ʹ, and the resulting events are then required to produce positive TX anomalies within the aggregation domain (10° × 10°) on the same day as the Z500 peak. The limited time period renders detrending obsolete and is more than compensated by many ensemble members (80 each for aIsI and aIsF; only results shown for aIsI).

2.4 Backward Trajectory Analysis of Hot Events

Motivated by the presence of anomalously high temperatures throughout the troposphere above the PNW during the 2021 event, we proceed to discover the origin of this heat anomaly and compare it to other North American heatwaves (Section 2.1.1). By doing so, we not only unravel where the air that enabled unprecedented temperature escalations in the PNW came from, but we can also infer if this air was primarily hot because of its origin, or perhaps due to heating processes occurring en route.

2.4.1 Tracking Air Back in Time With TRACMASS-ERA5

To follow the evolution of the air masses of the hot events backwards in time, we employ the Lagrangian trajectory model TRACMASS v7.0 (Aldama-Campino et al., 2020; Döös, 1995), which can be applied to study the atmospheric or oceanic circulation (e.g., Dey & Döös, 2020; Liang & Xue, 2020). TRACMASS represents reanalysis or General Circulation Model (GCM) winds as 3D mass transports between grid boxes by solving the continuity equation. Consequently, the resulting trajectories are mass conserving, and at the grid-scale or larger scales, consistent with the input circulation (Döös et al., 2017). Here, TRACMASS is employed to calculate ERA5-based 15-day backward trajectories, using 6-hourly horizontal winds at 0.5° × 0.5° on hybrid model levels and surface pressure. We largely follow the IFS setup (provided by the creators of TRACMASS, and accessed on 25 November 2021, through https://www.tracmass.org) for 6-hourly ERA-Interim input data, using the time step scheme with 120 iterations between model time steps. The release of about 11’000 air parcels—conceptualizations of coherent air masses—for each hot event is configured as follows: for a 4° × 4° area centered over each identified heatwave (see details above), air parcels are released 5 times (every 6 hr on the peak day of the hot event, i.e., 00, 06, 12, 18 UTC, and on 00 UTC of the following day). These parcels are seeded on a vertical wall (isec = 3), with each particle reflecting the air mass at the respective time (nqua = 3). About 3–4 parcels are seeded in each selected grid box (partquant = 50). This is performed such that air parcels residing over the center of each hot extreme, between 600 and 400 hPa, can be tracked backwards in time for at least 15 days. This pressure range is selected to represent the state of the middle troposphere, above any potentially deep ABL during hot events, as to avoid tracking substantial amounts of air within the boundary layer back in time. In addition to air parcel positions, we also trace the specific humidity and temperature to perform further analyses.

We note that, again motivated by storage considerations, we use temporally limited (6-hourly) and vertically incomplete ERA5 data on the lowermost 98 model levels (instead of the full 137, which cover the entire stratosphere and even most of the mesosphere). The 98 levels suffice, however, to track air parcels released in the middle and lower troposphere back in time for 15 days. Nevertheless, we compared trajectories calculated as described above, and repeated the calculation for the 2021 PNW heatwave using hourly ERA5 input at 0.25 × 0.25 horizontal resolution and at all 137 vertical levels and found only minor differences (not shown). This also holds for backward day –15 when trajectory errors are expected to be greatest, since they accumulate in the direction of integration (e.g., Stohl & Seibert, 1998).

2.4.2 Diagnosis of Latent Heating Along Trajectories

Even though hot events are exclusively diagnosed based on temperature here, many of them are associated with anticyclonic circulation aloft, which implies adiabatic warming due to subsidence particularly in the last 72 hr (Zschenderlein et al., 2020). While this adiabatic heating—converting potential into internal energy—is often considered to be a key driver of high near-surface temperatures in the midlatitudes (e.g., Bieli et al., 2015), it cannot explain the presence of high potential temperatures in the upper troposphere such as witnessed during the 2021 PNW heatwave (Neal et al., 2022). Therefore, to investigate the unprecedented free tropospheric heat during the 2021 PNW event, we consider the initial state of air 15 days prior to arriving, and processes occurring en route that shape the final state above the respective hot extreme. Specifically, we unravel the potential temperature budgets of all North American hot events including the recent PNW heatwave, and differentiate between diabatic heating and cooling. In addition, latent heat release as a particular kind of diabatic heating is also diagnosed.

In practice, we employ a process-based approach (Keune & Miralles, 2019; Sodemann et al., 2008) to estimate latent heating along air parcel trajectories. The underlying rationale is that if a specific process (e.g., radiative cooling) dominates within a 6-hourly trajectory step, this is reflected in a change of state (e.g., decreasing potential temperature, conserved specific humidity). From the perspective of an air parcel, latent heating should be accompanied by a rise in potential temperature while the equivalent potential temperature—i.e., conserved even for vertical motion including condensation—should either remain roughly constant or decrease due to dry air entrainment. We note that the trajectory calculations are imperfect, and associated inaccuracies, interpolation errors as well as numerical imprecision can cause noise (Sodemann et al., 2008). Therefore, we employ three sets of increasingly strict criteria to determine latent heating; loose, moderate, and strict. For all of these, we require the 6-hourly potential temperature change to be positive. We also enforce a relative humidity (RH) criterion; the RH must exceed either 70%, 80%, or 90% at the beginning or end of a 6-hourly time period. Furthermore, the equivalent potential temperature increase is limited to 2°C, 1°C, or 0°C. This results in three progressively conservative estimates of latent heating; unless specified otherwise, results are shown for the moderate criterion (80% RH required, and a maximum equivalent potential temperature increase of 1°C).

2.5 Exploring the Impact of Upwind Latent Heating on Downwind Temperatures With CESM

The analysis of TRACMASS-ERA5 trajectories enables an insight into the upwind diabatic heating budgets of hot events. However, it cannot provide an accurate estimate of the causal link between upwind latent heating and downwind heatwave temperatures—especially since the diagnosis of latent heating requires assumptions and associated parameter choices (see above). To this end, we first repeat the backward trajectory calculation with TRACMASS as described above but use fully constrained CESM output as a forcing instead. The only notable technical differences to TRACMASS-ERA5 are introduced by different horizontal and vertical input resolutions; hence, fewer air parcels are released for identical seeding settings. Next, having verified that the 2021 PNW heatwave backward trajectories are largely consistent between ERA5 and our base CESM run, we perform additional simulations to estimate the heatwave temperature impact of remote latent heating. Since trajectory analyses, whether forced directly by ERA5 or by ERA5-driven CESM data, reveal a predominantly oceanic origin of 2021 PNW heatwave air 15 days prior to arrival, latent heating is controlled in a large domain within the North Pacific Ocean (20°–60°N, 120°−240°E). In practice, we artificially increase or decrease specific humidity within the oceanic part of the domain, which results in less moisture being available for condensation and ultimately precipitation, our proxy for latent heating.

Considering that the lower troposphere contributes most of the moisture for precipitation (e.g., Sherwood et al., 2010), which is only partly affected by the default nudging procedure below ∼700 hPa (Section 2.2.1), we set up a new reference simulation, also fully forced (like oFaFsF), but with nudged specific humidity—in addition to horizontal winds—throughout the entire troposphere. This results in similar 2021 PNW heatwave temperatures as in our base simulation obtained for the vertical nudging profile (visualized in Figure S2 in Supporting Information S1) and without humidity-nudging (Figure S3 in Supporting Information S1), and serves as a reference for other simulations with a modified vertical specific humidity profile. The latter are obtained by increasing and lowering the values of 6-hourly specific humidity fields used for nudging, only within 500–800 hPa and the North Pacific domain from 19 to 26 June, using several scaling factors ranging from 0.5 to 1.25. This procedure acts to increase or decrease precipitation, but we refrain from attempting to fully suppress precipitation within our domain for multiple reasons. Enforcing dry air in proximity to the surface results in unrealistically deep marine ABLs and affects the surface energy partitioning; also, humidity is controlled only every 6 hr while the model has time steps of 30 min, so that moisture can still be gained by surface evaporation or enter the domain horizontally. Even more importantly, since all these experiments rely on the actual atmospheric circulation, enforcing a climatological—or any other reference—precipitation field would not yield a physically coherent simulation. Instead, we use a set of experiments with moderately altered latent heating to gauge the sensitivity on 2021 PNW heatwave temperatures. To the authors' knowledge, such an estimate has not yet been presented, even though the importance of upwind latent heating for downwind dynamics has long been noted (Davies et al., 1993; Hoskins et al., 1985; Pfahl et al., 2015; Steinfeld & Pfahl, 2019; Stoelinga, 1996; Zschenderlein et al., 2020).

3 Results

3.1 Record-Shattering Temperatures and Their Physical Drivers

In the last days of June 2021, much of the Pacific Northwest experienced near-surface temperature anomalies approaching—and even exceeding—20°C w.r.t. 1982–2008 (based on ERA5; Figure 2a). Averaged across the heatwave region (yellow contour in Figure 2a) and according to ERA5 data, the daily maximum temperature anomaly peaks at >15°C on 30 June (Figure 2b; black line) following a sharp increase after 25 June. Our base simulation from the temperature anomaly disentangling framework with constrained oceans, soil moisture as well as midlevel and upper-level tropospheric winds, named “(oF)aFsF,” slightly overestimates the event magnitude. Nevertheless, the simulated heatwave portrays a realistic temperature evolution. This is also the case for the simulation with climatological—rather than actual—prescribed soil moisture (aFsC). The remaining model simulations visualized here feature an interactive atmosphere and interactive or constrained soil moisture (aIsI, aIsF) and consist of 80 ensemble members. None of these simulations produced a heatwave that could rival the actual event; in this region and for late June 2021, even the ensemble maxima (colored shadings) are roughly 5°C less hot than the reference simulation (aFsF). This not only emphasizes the key role of atmospheric circulation, but also highlights the need to understand the full set of drivers involved in the occurrence of this remarkable hot extreme. Thus, using the presented set of simulations and the additive disentangling approach introduced by Wehrli et al. (2019; see also Section 2 and Figure 1), the effects of several physical drivers are estimated (Figure 2c).

Details are in the caption following the image

Unprecedented temperature anomalies in late June 2021 in the Pacific Northwest. (a) Daily maximum temperature anomaly in ERA5 on 30 June 2021. During the last days of June, the event reached its peak intensity in the heatwave region used throughout this study (yellow contour). Anomalies are calculated w.r.t. 1982–2008. (b) Anomaly time series for the heatwave region, depicting area-weighted mean daily maximum temperatures from ERA5 (black line). In addition, multiple simulations from CESM are shown: the “reference”—fully constrained atmosphere and soil moisture (aFsF), same but with constrained climatological soil moisture (aFsC), and interactive atmosphere with constrained or interactive soil moisture (aIsF, aIsI). The latter consist of 80 simulations each, whose ensemble mean is indicated by a solid line, while the shading visualizes minima and maxima. (c) Physical drivers in the additive disentangling framework (Figure 1), obtained by TXʹ differences visualized in (b). For soil moisture, an event-specific (SM) as well as a “generic” estimate (SM*) is provided, with corresponding circulation and circulation and imprinted SM effects. The sum of either set of soil moisture and circulation contributions together with the generic effects of recent warming and anomalous surface ocean is equal to TXʹ indicated by the fully constrained simulation.

Based on our model experiments with a nudged atmosphere, the effect of soil moisture amounts to 2.8°C during the peak of the event, while the circulation effect reaches >13.4°C. This soil moisture contribution is solely based on simulated (one-way) land-atmosphere interactions—but not (two-way) feedbacks—for actual (aFsF) and climatological (aFsC) soil moisture in CESM. In addition, the surface only affects winds in the ABL and lower troposphere due to enforcing the ERA5 large-scale circulation further aloft. Despite the inability of the atmosphere to influence soil moisture in these experiments, we emphasize that the prescribed soil moisture has been calculated in offline CLM simulations using ERA5 meteorological forcing and is hence consistent with the near-surface conditions arising from actual land-atmosphere feedbacks and other interactions. We also provide an alternative estimate of the circulation effect by first comparing the base simulation (aFsF) to the ensemble mean of simulations with interactive atmosphere and constrained soil moisture (aIsF); this yields an even stronger contribution of up to 14.7°C. By design, however, this estimate includes effects from both circulation as well as soil moisture, since the ERA5-constrained middle and upper tropospheric winds in aFsF can contain imprints of actual land-atmosphere and other interactions or feedbacks that shaped the large-scale flow. Land-atmosphere interactions were found to be particularly strong during other record-shattering midlatitudinal heatwaves (e.g., Fischer et al., 2007; Miralles et al., 2014), and strongly depend on the prevailing large-scale circulation. As such, temperature impacts of land-atmosphere interactions during the 2021 PNW heatwave tend to be underestimated in simulations with fully interactive atmospheres. Therefore, we consider the corresponding soil moisture effect (aIsF-aIsI) of at most 1.5°C to represent a “generic” estimate that likely fails to capture the interactions between anomalous soil moisture and event-specific circulation. This approach hence differs notably from the soil moisture contribution based on aFsF and aFsC, for which the same actual large-scale circulation is prescribed. The same limitation applies to the two remaining drivers in our additive framework; the effects of anomalous SSTs (w.r.t. the mean state of 2015–2020), and of our changing climate (compared to 1982–2008). Averaged over 26–30 June, the background warming acted to amplify the event's magnitude by 0.9°C, whereas the surface ocean only contributed 0.1°C. The circulation explains 81.0% of the event anomaly in the same 5-day period, while the event-specific soil moisture effect amounts to 11.7%. Even though the generic soil moisture effect is considerably weaker at the peak of the event, it explains nearly the same fraction of the 5-day mean temperature anomaly with 10.2%, whereas the corresponding circulation contribution accounts for 82.4%. The remaining event magnitude is attributable to anomalous ocean surface state (0.9%) and recent warming (6.4%).

According to our analysis, the 2021 PNW heatwave should thus be seen as largely enabled by dynamics (i.e., the atmospheric circulation), yet significantly exacerbated by thermodynamic drivers—particularly dry soils but also the warming climate, while surface ocean anomaly effects as identified here are negligible. We point out that the same limitations as already mentioned for soil moisture, i.e., ocean-circulation feedbacks cannot be targeted within our framework, also apply here and hence the true roles of land and ocean surfaces may be underestimated. Nonetheless, the results of our disentangling framework suggest that soil moisture acted to amplify heatwave temperatures, particularly toward the peak of the event at the end of June. This ramping up is only captured by the approach comparing climatological to actual soil moisture with constrained winds, as the generic soil moisture effect remains relatively constant at 1.4°C. While ongoing soil desiccation during a heatwave tends to strengthen land-atmosphere interactions (e.g., Miralles et al., 2014), large parts of the PNW already experienced drought conditions well before the event (Ansah & Walsh, 2021; Bumbaco et al., 2022). As such, it remains unclear how the event-specific soil moisture effect increased from about 0°C to nearly 3°C in a matter of days. We thus expand our investigation of the drivers behind extreme near-surface heat, focusing on the interplay of soil moisture and exceptional atmospheric circulation. The latter is well represented at large scales by our fully constrained CESM simulation, and since the winds below ∼700 hPa are exempted from the nudging procedure, the ABL can—in theory—freely interact with both the troposphere above and land or ocean surface below. In practice, this is complicated by the principle of continuity, which implies that even winds calculated on model levels not directly affected by the nudging can still be effectively governed by the flow aloft. Moreover, both the complex terrain in parts of the PNW and surroundings as well as the emergence of deep ABLs, which has been noted for other record-shattering heatwaves in the midlatitudes (e.g., Miralles et al., 2014), suggest that the assumption of an unconstrained ABL may not always hold.

In light of these considerations, there is ample reason to examine the impact of nudging by performing a sensitivity analysis. Essentially, the default nudging procedure employed here acts to constrain horizontal winds for most but not all of the troposphere, which is achieved by a gradual transition from no to full nudging centered at about 700 hPa. However, since the nudging is applied to hybrid sigma-pressure model levels that depend on surface pressure, this threshold is not constant and is higher over elevated surfaces. Therefore, we performed a total of four additional simulations with the same setup as aFsF as well as aFsC, yet nudging only winds further from the surface, above roughly 500 and 400 hPa, respectively. All these CESM experiments capture the spatial pattern of daily maximum temperatures on 30 June 2021 (Figure 3a), but the base simulation indicates more extreme heat particularly in the northeast of the heatwave region than the simulations with less constrained troposphere. This could be caused by an overestimation of heat entrainment from the free troposphere in this area, for which CESM, and in particular with the default nudging settings, simulates unusually deep ABLs (Figure 3b). According to ERA5, however, strong positive anomalies in ABL height during the peak of the event are largely restricted to the mountainous western part of the heatwave region (see also Figure S4 in Supporting Information S1). There, soil moisture effects (Figure 3c) are comparatively moderate, as CESM produces deep ABLs even in simulations with climatological soil moisture (aFsC, not shown). Further to the northeast, on the other hand, heatwave temperatures diverge clearly in simulations with prescribed actual and climatological soil moisture, locally exceeding 5°C.

Details are in the caption following the image

Unraveling the roles of soil moisture and atmospheric boundary layer (ABL) dynamics during the 2021 Pacific Northwest heatwave. (a) Near-surface maximum temperature anomalies on 30 June 2021 in ERA5, the base Community Earth System Model (CESM) simulation (aFsF), and two additional simulations with increased nudging threshold (from left to right). (b) As (a), but showing ABL height anomalies instead. (c) Soil moisture effects, always estimated by calculating the temperature anomaly differences between simulations with prescribed actual and climatological soil moisture. Since this is only possible with our CESM simulations, the leftmost subplot corresponding to ERA5 data is empty, and only visualizes the heatwave region employed throughout the paper (obtained with a 3σ threshold; yellow contour), as well as the additional smaller region delineated by even more anomalous heat (5σ; pink contour). (d) Regional averages of daily maximum near-surface temperatures for the 3σ and 5σ heatwave regions, for ERA5 and the same CESM simulations used in (a)–(c). (e) As (d), but for the ABL height instead. (f) As (e), but for simulations with prescribed climatological soil moisture. (g) Estimates of the soil moisture effect on TXʹ, given by the difference of simulations with actual (sF) and climatological (sC) soil moisture.

In this area, the choice of nudging threshold introduces considerably uncertainty to our disentangling approach at the grid-cell scale, whereas the estimated maximum temperatures and ABL heights, as well as soil moisture contributions, are similar for the entire heatwave region (left column in Figures 3d–3g). As expected, based on the overestimation of ABL depths in eastern parts of the heatwave region, all CESM experiments with prescribed actual soil moisture are associated with larger ABL height anomalies compared to ERA5 for the entire domain (Figure 3e). For a smaller region given by standardized ERA5 TX anomalies exceeding 5σ (rather than 3σ) on 30 June (pink contour in Figures 3a–3c), the simulated mean ABL height anomalies range from about 1 to 2 km. For the much larger 3σ region, even though the ABL height increase according to ERA5 is overestimated in simulations constrained toward actual soil moisture (Figure 3e), this increase ceases well before peak heat when enforcing climatological soil moisture (Figure 3f). This suggests that dry soils were crucial for the progressive ABL deepening prior to the peak of the heatwave, which also holds for the smaller heatwave region, despite a larger overall spread. As for the larger region, the experiment with nudged winds for pressures below ∼700 hPa produces the deepest ABLs. In fact, only this simulation still generates notable positive anomalies in ABL heights for climatological soil moisture in the 5σ heatwave region (Figure 3f), which is likely introduced by nudging within or in proximity to growing ABLs during the event. This effect is more pronounced in conjunction with prescribed actual soil moisture, since the dry soils in the region acted to maintain surface sensible heating despite escalating air temperatures, and hence fostered ABL growth.

Because of these considerations, the simulations with the most constrained ABL, i.e., nudging winds for pressures lower than 700 hPa, may provide a conservative estimate of the true soil moisture effect, particularly for the 5σ region, where the less atmospherically constrained simulations have notably higher contributions of up to 5°C. In the face of the discussed uncertainties, and the inability of CESM to accurately represent the complex orography in the region, we refrain from providing a “best estimate,” but rather point to the range provided by our nudging sensitivity analysis (Figure 3f). Overall, our analysis indicates that the soil moisture effects we target here—i.e., excluding interactions between the land surface and the large-scale circulation—are intimately linked to ABL dynamics, which explains why the strongest contributions are reserved for the peak of the heatwave, when ABL heights culminated according to both ERA5 and the CESM simulations. Moreover, as evident in the left column of Figure 3b, the most extreme heat (pink contour) occurred either where ABL height anomalies exceeded 2 km, or downwind, i.e., connected by horizontal advection, of such areas with deep ABLs (Figures S4a–S4c in Supporting Information S1). In fact, the 5σ heatwave region partly contains areas in which unprecedented ABL heights were achieved, exceeding climatological maxima from several hundred meters up to >1 km (not shown). This result provides nuance to the findings of Neal et al. (2022), who noted an initial decrease in ABL heights upon the arrival of the block on 26 June at 49°N, 119°W, due to the associated large-scale subsidence. While this is the case for certain grid cells, in general, there was a progressive deepening of the ABL in proximity to mountain ridges (Figures S4–S6 in Supporting Information S1). Interestingly, surface sensible heating was not anomalously high in many areas with extremely deep ABLs (Figure S4d in Supporting Information S1), yet it has been noted that ABLs over mountain ranges in dry climates often grow deeper than over flat terrain, with depths of 2 km or even 3 km (De Wekker & Kossmann, 2015). The deep ABLs were hence likely enabled by interactions between atmosphere and the complex terrain, and facilitated by widespread soil drought conditions, fostering heat entrainment from the free troposphere throughout the observed build-up of near-surface temperatures. We have hence established that the PNW experienced favorable conditions for entraining tropospheric air, yet it remains unclear to what degree the free troposphere itself was anomalously hot, implying an even stronger vertical advection of heat. To address this question, next, we examine the anticyclonic circulation behind the 2021 PNW heatwave.

3.2 Comparing the Dynamics to Other Hot Events

To better understand the large-scale dynamics of the 2021 PNW heatwave, we begin with an analysis of the 150 hottest historical events in northwestern North America since 1982 (between 40°–65°N and 100°–160°W, see Section 2.1), of which the 2021 PNW heatwave is ranked as number one. For all of these events and starting on the respective day of peak heat, we track anticyclones day-by-day and using a horizontal extent of 10° × 10° (see Section 2.3 for details), and average all variables over this domain. All events are tracked using Z500ʹ, and results are visualized w.r.t. the timing of peak heat. By doing so, we implicitly assume that every detected hot event is associated with anticyclonic conditions. The fact that during peak heat, Z500ʹ—our proxy of anticyclone intensity—ranges from 0.8σ to 3.3σ and is 2σ on average, suggests that overall, this assumption is justified. Furthermore, the timing of maximum anticyclone intensity does not necessarily align with peak heat as clearly demonstrated by the 2021 PNW heatwave, and in terms of maximum Z500ʹ, 120 and 147 out of all 150 hot events exceed 2.0σ and 1.5σ, respectively. For the 2021 PNW heatwave, the associated near-surface daily maximum temperature anomaly (TXʹ) exceeds every other event since 1982 (Figure 4a). Note that the discrepancy of peak heat timing, here 29 June rather than 30 June, can be explained by the fact that a moving 10° × 10° window centered on the respective anticyclone center is used rather than the static heatwave region introduced in Figure 1, and in addition, daily maximum temperatures in the PNW tend to occur close to 00 UTC. The peak in Z500ʹ during the 2021 PNW heatwave (pink lines in Figure 4a) occurs several days before the highest near-surface temperatures were reached, and is also unprecedented, albeit less clearly than for TXʹ. Temperatures in the upper (approximated by 250 hPa), middle (500 hPa), and lower troposphere (850 hPa) culminate about 5, 3, and 1–2 days prior to the TXʹ peak, respectively, consistent with the notion of sinking and adiabatically heated air. Nevertheless, the presence of warmer-than-usual air even far aloft—and occurring well before the anticyclone reaches its peak intensity—indicates that it not only matters how quickly the air is forced to sink, but also what initial state the air has even before it spirals downwards.

Details are in the caption following the image

Investigating the dynamics of the 150 hottest events in northwestern North America since 1982. (a) Results for nudged atmospheric circulation, with near-surface temperature anomalies, Z500 anomalies, and temperature anomalies in the upper, middle, and lower troposphere (from left to right). All data are based on a 10° × 10° domain centered on the respective anticyclone position and are plotted w.r.t. the day of peak TXʹ intensity. A range excluding the Pacific Northwest 2021 event (pink line) indicates previous minima and maxima. In addition, events are grouped based on their near-surface peak heat anomaly, using the 50th TXʹ percentile. (b) As (a), but based on simulations with interactive atmosphere and soil moisture. The 10th, 90th, and 99th TXʹ percentile groups are visualized along with the 50th percentile data group already employed for (a). (c) For each of the 150 hot events, the mean TXʹ in the 5 days leading up to and including peak heat and for static 4° × 4° domains is shown as a function of identically aggregated Z500ʹ. (d) As (c), but showing TXʹ as a function of the ABL height anomaly instead. (e) As (c), but using temperature anomalies at 500 hPa in place of Z500ʹ. (f) Expressing the midtroposphere temperature (T500ʹ) as a function of anticyclone intensity (Z500ʹ).

Next, we extend the analysis to simulations based on fully interactive atmospheres (Figure 4b). As the large-scale circulation evolves freely in these CESM experiments, instead of tracking anticyclones associated with historic heatwaves, we select anticyclonic events within northwestern North America, and ensure that these anticyclones are associated with positive temperature anomalies. As for the actual events simulated with nudged circulation, higher near-surface temperatures—as visualized by the 90th and 99th percentile data groups in Figure 4b—are associated with stronger anticyclones (Z500ʹ in Figure 4b) and pronounced temperature anomalies extending throughout the troposphere. For the hottest events represented by the 99th percentile group, a delay between Z500ʹ and TXʹ peak is evident, similarly to how the 2021 PNW heatwave achieved the highest temperatures only after the anticyclone intensity had reached a pinnacle (Figure 4a). Overall, our analyses of CESM simulations with interactive atmosphere suggest that the tropospheric temperature evolution during the actual 2021 PNW event was more than the mere manifestation of strong atmospheric subsidence, and instead directly contributed to the extreme near-surface conditions. In other words, air temperatures appear even more extreme than suggested solely by the exceptionally strong anticyclonic circulation behind the 2021 PNW heatwave. We investigate this further by comparing spatiotemporal averages of all 150 hot events, now calculated for a static 4° × 4° domain centered on the location of peak heat, and for the 5 days leading up to and including the day of most extreme temperatures. Figure 4c relates anticyclone intensity to near-surface daily maximum temperature, which shows that—based on the bulk of the remaining 149 events—the PNW 2021 event was hotter than implied by Z500ʹ, although only few events reach similarly extreme Z500ʹ in excess of 200 m. Similarly, the ABL height anomalies during late June 2021 also do not seem to account for the strong TXʹ, at least not single-handedly (Figure 4d). On the contrary, the aggregated anomalies in midtropospheric temperature (Figure 4e) are far higher than for any other event, which makes it difficult to extrapolate the relationship of T500ʹ and TXʹ beyond the observed range of the 149 hot events. This is why we do not perform a regression, and instead point out that a high T500ʹ is not necessarily associated with high TXʹ, yet convective instability would prevent extremely high near-surface temperatures in the absence of a sufficiently hot troposphere (e.g., Buzan & Huber, 2020). In other words, there is a physical constraint on near-surface temperatures imposed by free tropospheric conditions which also depends on the vertical humidity profile, yet this constraint likely only manifests for the most extreme heatwaves. This reasoning is supported by the fact that the T500ʹ-TXʹ relationship emerging for the 30 events (or less) with highest TXʹ is roughly in line with the unprecedented near-surface and midtropospheric temperate anomalies of the 2021 PNW heatwave. Moreover, as depicted in Figure 4f, the troposphere was hotter than implied by the local anticyclone intensity; hence, our next step is to determine the source of this tropospheric heat.

3.3 Unraveling the Role of Upwind Latent Heating

To shed light on the origin of tropospheric heat, air residing over the PNW during peak heat is tracked back in time for 15 days using ERA5 data (see Section 2.4 for details). This unravels the history of heatwave air, depicted as a function of time and height (Figure 5a) and of latitude and longitude (Figure 5b). In the last few days prior to peak heat, from 25 June, the air masses were already part of a strengthening anticyclone (red and orange colors in Figures 5a and 5b), and hence mostly subject to slowly descending vertical motion. Before becoming part of the anticyclone, the air crossed the entire Pacific Ocean as also reported by Mo et al. (2022), with a large amount of tracked air parcels originating in the lower troposphere in the tropical West Pacific. Therefore, the initial state of the air features a high mean temperature and pressure compared to the 149 other hot events in northwestern North America (pink lines in Figure 5c). Crucially, the high initial moisture content sustained intense latent heating until backward day −7 as the air masses ascended. The brunt of this upwind heat release occurred in two warm conveyor belts, i.e., slantwise ascending air streams in the warm sector of extratropical cyclones (e.g., Browning, 1986; Schäfler & Harnisch, 2015), to the southeast of Japan and south of Alaska. This finding is consistent with the notion of remote and nearby ascending heating branches (w.r.t. a block) by Zschenderlein et al. (2020). To facilitate a comparison, Figure 5c highlights event #13 with the strongest net diabatic heating—defined by the potential temperature difference of backward day 0 and –15—and the highest latent heating peak. The air tracked backwards for event #13 first gains moisture and reaches a specific humidity of almost 8 g/kg, comparable to the 2021 PNW heatwave at backward day −15, yet at a considerably lower potential temperature. This indicates that not only strong upwind latent heating contributed to the extreme final state of 2021 PNW heatwave air, but also its largely tropical origin, which sets it apart from other events such as #13. Next, we compare event #113 with the highest initial potential temperature, whose temperature evolution from backward day −7 to −4 is remarkably similar to the 2021 PNW heatwave. But whereas the overall downward motion and hence adiabatic heating is sustained in the case of air approaching the PNW in 2021, this is not the case for event #113. Moreover, even though event #13 had more intense adiabatic heating in the last few days prior to arrival (see the temperature progression in Figure 5c), the air en route to the PNW in 2021 descended longer and from a greater altitude.

Details are in the caption following the image

Backward trajectory analysis of the 2021 Pacific Northwest (PNW) heatwave and other northwestern North American hot events. (a) Vertical backward analysis of air residing over the heatwave region in late June 2021, with colors marking different time periods. Only 1,000 randomly selected trajectories are shown to facilitate their visualization. (b) Horizontal overview of backward trajectories expressed as gridded air mass and colored consistently with the time dimension in (a). (c) Evolution of temperature, pressure, potential temperature, and specific humidity for the air involved in 150 northwestern North American hot events (gray lines in the upper row), averaged over all respective backward trajectories. The 2021 PNW heatwave is emphasized (pink lines), as well as events #13 and #113 (elaborated in the main text). The lower row displays temporal changes of the same variables as displayed above. (d) Total latent heating diagnosed en route, from backward day −15 to 0, and the corresponding net diabatic heating rate for all 150 hot events. (e) Midtropospheric temperature (T500) from ERA5 in the last 5 days before and including peak heat for all hot events plotted against three predictions; once using only trajectory-based information on the potential temperature (red shading and dots), once using only local Z500 (blue shading and dots), as well as using both predictors (black markers).

This comparison indicates that the hot final state at backward day 0 was achieved by a well-timed interplay of large-scale circulation and tropical air: first, the near record-breaking initial specific humidity was consumed while the air masses ascended and traversed the North Pacific, boosting the potential temperature to its peak at backward day −6. Then, in a seamless transition as the air became engulfed in the anticyclone at a high altitude, subsidence and the associated adiabatic heating enabled a continuous conversion of potential into internal energy. Our findings confirm and expand on Neal et al. (2022) who focused on the dynamical implications of upwind latent heating but suggested that the unusual tropospheric warmth in late June 2021 in the PNW originated in lower latitudes. The authors also reasoned that heat released from condensation may have contributed to the elevated temperatures in the troposphere, but did not calculate backward trajectories or perform any other detailed analyses. We note, however, that our diagnosis of latent heating is subject to uncertainty and is thus performed thrice with increasingly strict criteria (see Section 2.4), with Figure 5 displaying results for the moderate criterion. This choice does not affect our conclusions. The net potential temperature change along the trajectories, visualized in Figure S7 in Supporting Information S1 for all events together with latent heating estimates, also depends on other sources of diabatic heating—most notably surface interactions—and cooling. Nevertheless, and as shown in Figure 5d, while the spread in net diabatic heating is large for events with low or moderate amounts of heat released from condensation, strong latent heating is always associated with increasing potential temperature. In a last step, we employ a linear regression to predict the T500 in the last 5 days leading up to and including peak heat for each hot event solely using trajectory-extracted information on the potential temperature. The latter informs on the final state of potential temperature at backward day 0, and allows us to reasonably predict midtropospheric temperatures (red shading in Figure 5e). We repeat this using only Z500 as a predictor averaged over the same time period as T500 for each event, resulting in an improved prediction (cf. blue and red shadings in Figure 5d). When both predictors are employed, which conceptualize the available potential energy and the inclination of the troposphere to heat adiabatically through subsidence, most T500 predictions are further improved. This is consistent with the expectation that tropospheric temperature is not only governed by processes occurring at regional scales, but also by the history of the air, and hence both its initial state as well as processes occurring en route such as latent heating.

To summarize, our trajectory analysis has identified exceptional upwind latent heat release fueled by air from the tropical West Pacific as a key driver behind the elevated midtropospheric temperatures of the 2021 PNW heatwave. Although this explains the previously noted apparent mismatch between local anticyclone intensity and tropospheric temperatures, it remains unclear to what extent this extra heat ultimately affected near-surface temperatures. We hence perform additional CESM experiments in which the specific humidity, and hence precipitation and latent heating, is artificially reduced or enhanced from 19 to 26 June over the Northern Pacific (domain visualized in Figure 6a). Within this time period, the air that would later be engulfed in the anticyclone over the PNW experienced the strongest latent heating (Figure 5). We point out that contrary to other contributions, it is not straightforward to, e.g., prescribe “climatological” specific humidity, since that would be inconsistent with the given large-scale circulation pattern. Instead, we gauge the heatwave temperature sensitivity to upwind latent heating by performing a sensitivity analysis. Figure 6a displays the outcome of the experiment where the artificial reduction of specific humidity is strongest; as one would expect, this causes precipitation reductions in the North Pacific compared to our base simulation. But it also results in lower temperatures in and around our heatwave region, with 5-day average differences (26–30 June) locally exceeding 2°C. Aggregated for the heatwave region, our experiments indicate that not only would the same event have been cooler with less condensational heat release over the North Pacific, but it could have been even hotter if there had been more available moisture. The strongest effects are found at the onset of the heatwave for our experimental setup, yet even several days after our specific humidity manipulations at the end of June, the lack (or surplus of) heat released far upwind still persists. Aggregated for both the heatwave region and through 26–30 June, and expressed as a function of precipitation over the North Pacific, our results indicate a downwind daily maximum temperature sensitivity to upwind precipitation on the order of several degrees Celsius (Figure 6c). As such, we have established a causal relationship between upwind latent heating and downwind heatwave conditions, and find that for the 5 days leading up to and including peak heat, the temperature impact is comparable to the previously discussed soil moisture anomaly effects.

Details are in the caption following the image

Downwind temperature impact of latent heating based on Community Earth System Model (CESM) simulations. (a) The domain in the North Pacific for which tropospheric specific humidity is modified to control precipitation is visualized by precipitation anomalies (120°−240°E, 20°–60°N; only over the ocean, hence all other ocean grid cells shown in gray). More specifically, the mean precipitation difference between the reference simulation (nudged toward ERA5 specific humidity) and the experiment where humidity is artificially decreased by a factor of 0.5 is shown (19–26 June). Similarly, land grid cells indicate the resulting mean differences in daily maximum temperatures around the peak of the 2021 Pacific Northwest heatwave (26–30 June). (b) Temporal evolution of heatwave region temperature differences for all employed specific humidity scaling factors. (c) Aggregated results, depicting heatwave region temperatures as a function of mean upwind precipitation rates in the North Pacific domain. Colors indicate the underlying specific humidity scaling factor, as shown in (b).

4 Conclusions

We investigated the physical drivers of the unprecedented heatwave in the Pacific Northwest in 2021 using sensitivity experiments with an ESM, reanalysis data, and backward trajectories. Our findings point to the crucial role of atmospheric circulation during the event, which is why only simulations with constrained winds in our framework can reproduce the extreme temperature anomalies. Indeed, in our framework, we cannot investigate the potential drivers and interactions underlying the large-scale circulation. Even though the role of upwind latent heating in shaping the dynamics of this event has already been recognized (Neal et al., 2022), the impact of, e.g., local and remote surface-atmosphere interactions remains unclear. This is one aspect of a bigger challenge; improving our understanding of atmospheric circulation responses to a changing climate, and the implications for extreme weather. Still, our disentangling framework indicates that anomalously low soil moisture was crucial for the establishment of deep ABLs by maintaining surface sensible heating, which was impeded by the escalating near-surface air temperatures throughout the heatwave. This implies enhanced vertical heat advection from the hot troposphere aloft, that, according to ERA5, was especially pronounced in the mountainous western parts of the heatwave region. Once mixed into the ABL, the hot air also enabled record-shattering temperatures elsewhere through horizontal advection. We performed additional CESM experiments with higher nudging thresholds compared to our base simulation to ensure that deep ABLs can evolve freely, and found that dry soils contributed >5°C during peak heat in some of the hottest areas. As underlined by a comparison of the dynamics behind the 150 hottest events in northwestern North America since 1982, the state of the troposphere in the PNW was characterized by excessive heat. Our trajectory analysis highlights exceptional latent heat release concentrated in two separate warm conveyor belts over the Pacific Ocean prior to the 2021 PNW heatwave, which explains why the tropospheric temperatures were so anomalous. Finally, using CESM simulations with modified upwind latent heating rates, we demonstrated that although this strong latent heating ceased about a week before the temperatures peaked in the PNW, it still affected heatwave temperatures, in some areas and averaged through 26–30 June by >2°C. Averaged over these 5 days leading up to and including peak heat the heatwave region used throughout the study, the mean temperature anomaly amounts to 13.7°C. Halving the upwind latent heat release results in a downwind heatwave 1.7°C less hot than that, whereas it would have been 1.6°C cooler for climatological soil moisture, and another 0.9°C cooler without recent background warming. Since the warming climate may have additionally contributed to both the strong latent heating and the soil drought, our estimate of the anthropogenic influence is likely conservative. We conclude that even though the dynamics remain a crucial feature and driver of this event, without additional processes and interactions, and in particular the contribution of drying soils to the temperatures, the observed record-breaking temperatures would not have occurred.

While existing research implies that (remote) thermodynamics in ascending humid airstreams helped shape the extreme dynamics over the PNW, this study shows that the same processes also directly exacerbated downwind heatwave temperatures. In other words, not only the local circulation pattern matters, but, owing to the origin of air and processes occurring en route to the heatwave region, also how it was established. We recognize that conceptualizing and compartmentalizing our Earth System and the extreme weather that it fosters has been essential to improve our understanding (e.g., Trenberth et al., 2015), yet events such as the 2021 PNW heatwave serve as a stark reminder of the underlying complexity. For reliable future projections, ESMs do not only need to adequately represent the circulation pattern, soil moisture, etc. where the extreme weather occurs, but should also accurately simulate spatiotemporal dependencies far beyond commonly considered scales, such as latent heating thousands of kilometers upwind. At last, we venture into the future and consider the implications of this event and the identified drivers. While the potential of even drier soils for the generation of more extreme heat has long been noted (e.g., Rasmijn et al., 2018; Seneviratne et al., 2006), we highlight the possible role of stronger upwind latent heating in future hot extremes. Our atmosphere is heating up, unmistakably manifesting the consequences of anthropogenic carbon release (e.g., Mann et al., 1999; Sippel et al., 2020; Zeebe et al., 2016). Rising air temperatures imply a nonlinear increase in saturation vapor pressure, as described by the Clausius-Clapeyron relationship (Hardwick Jones et al., 2010; Held & Soden, 2006). This suggests a higher potential for heavy precipitation in the future, since air ascending from the lower to the upper troposphere—as was the case prior to the 2021 PNW heatwave and for other events—tends to rain out effectively all its initial moisture. Therefore, in line with upward trends in extreme precipitation (e.g., Barbero et al., 2017; Westra et al., 2013), we expect that future hot events could be fueled by even more intense upwind latent heating. In fact, our research indicates that, despite eclipsing previous high temperature records in the PNW, the 2021 heatwave would have been even hotter with more moisture available over the Pacific Ocean.

Acknowledgments

The authors acknowledge funding from the European Union's Horizon 2020 research and innovation program for the XAIDA project under Grant Agreement 101003469. We also thank Michael Sprenger for providing most of the ERA5 data employed in the study.

    Data Availability Statement

    CESM version 1.2.2 can be downloaded from the University Corporation for Atmospheric Research (UCAR) website (https://www.cesm.ucar.edu/models/cesm1.2/). The full ERA5 data set is accessible through ECMWF's data catalogue (https://apps.ecmwf.int/data-catalogues/era5/?class=ea). While the full set of CESM simulations employed here is too large to be deposited on data repositories, all data underlying the main figures and the complete TRACMASS-ERA5 backward trajectories are publicly accessible through Zenodo at https://doi.org/10.5281/zenodo.7319828. Additional data are available upon request from the corresponding author.