Unprecedented 21st century heat across the Pacific Northwest of North America

The study addresses the increasing frequency and intensity of extreme heat events driven by anthropogenic climate change. Global mean temperatures in the early 21st century are ~0.99 °C above pre-Industrial levels and projected to exceed 1.5 °C by 2040. The Pacific Northwest (PNW; 42–53°N, 115–124°W) experienced an exceptional heatwave in late June–early July 2021 associated with an omega block pattern, producing temperatures up to 49.6 °C, elevated wildfire activity, and increased heat-related mortality. 2021 also registered the highest June–August (JJA) near-surface air temperatures on record (1950–2021), about +3.6 °C above the 1951–1980 mean (ERA5). The PNW’s historically temperate climate, limited air conditioning prevalence, and infrastructure not adapted to extreme heat heighten vulnerability. Prior rapid-attribution studies suggest such an event would have been virtually impossible without anthropogenic forcing, but the rarity is difficult to quantify given the short instrumental record. The authors aim to augment observations with tree-ring proxies to reconstruct millennial-scale PNW summer temperatures, place 2021 and recent warming rates in a long-term context including the Medieval Climate Anomaly (MCA; 950–1250 CE), and evaluate future risks using CMIP6 projections.

The paper situates its work within literature documenting global increases in extreme heat and warming trends (IPCC AR6; studies on heatwave frequency, intensity, and duration). It references analyses of the 2021 PNW heatwave indicating its exceptional nature and strong anthropogenic contribution, while noting limitations due to short observational records. It highlights prior dendroclimatic research showing tree-ring wood density and related metrics as robust summer temperature proxies across the Northern Hemisphere, and regional reconstructions (e.g., Alberta Icefields) characterizing MCA warmth. It also references CMIP6 scenario modeling, concerns about some models’ warm biases, and underestimation of internal variability in projections of extratropical extremes.

Tree-ring data collection and preparation: The team sampled living conifers from temperature-limited environments at 17 PNW sites (42.0–53.0°N, 124.0–115.0°W), extracting two increment cores per tree. They targeted total-ring width and latewood blue intensity (LWB). Samples were dried, mounted, microtome-shaved to a planar surface, and polished. To avoid BI contamination by mobile compounds, they performed resin extraction (passive soak in 99.5% ethanol at 24 °C for 96 h), then re-prepared surfaces. Total-ring width and LWB were measured using an Epson Expression XL 12000 scanner and CooRecorder. Cross-dating accuracy was assessed with COFECHA. Series were detrended using the SignalFree framework with age-dependent splines (end behavior constrained to non-increasing) and variance stabilization on power-transformed residuals to preserve climate signals in BI. Climate data: The predictand is ERA5 JJA mean 2 m air temperature (t2m) averaged over the PNW for 1950–2021. For future comparison, CMIP6 JJA projections under SSP2-4.5 and SSP3-7.0 were compiled over the PNW from available ensemble members; multi-model medians and 5th/95th percentiles were calculated (40 and 36 members, respectively). Reconstruction model: A nested principal components regression (PCR) framework was used to account for changing predictor availability through time. Site-level detrended chronologies were screened against local observed climate variables; only chronologies significantly correlated (p < 0.05) with current-year summer temperature and the regional ERA5 target (1950–2017) under Pearson, robust Pearson, and Spearman were retained, yielding 29 chronologies. Predictors and ERA5 were initially pre-whitened to mitigate autocorrelation inflation; non-prewhitened significant predictors were then used in PCR. PCs with eigenvalues >1 (Kaiser–Guttman) were candidates; the final subset minimized AIC. Calibration/validation used early (1950–1985) and late (1986–2018) splits. Skill metrics included CRSQ/VRSQ and validation RE/CE; uncertainty was quantified by maximum entropy bootstrap with 300 replicates. As nests dropped back through time, variance stabilization re-scaled each nest to the preceding nest over their overlap. Final reconstruction values were bias-corrected to ERA5. Because predictors end in 2018, ERA5 values were appended for 2019–2021. Statistical analysis: All reconstructed values were converted to anomalies relative to the 1951–1980 mean. Gaussian PDF-based probability densities were estimated for reconstructed (950–2021) and ERA5 (1950–2021) values. Multi-year warm/cool periods were identified via a 30-year LOESS smoother, with bivariate characterization of duration and mean anomaly. Extreme annual departures were defined using the 50th, 90th, and 99th percentiles of warm/cool anomalies. Warming rates were assessed using backward 20- and 50-year moving windows. Low-frequency comparisons were made to independent Northern Hemisphere and Alberta Icefields reconstructions via z-scores and rolling 51-year correlations. Kernel density estimates compared distributions across pre-Industrial reconstruction, post-Industrial pre-observational, ERA5, and CMIP6 SSP2-4.5. Future risk of 2021-like conditions (+3.6 °C anomaly) was computed annually from CMIP6 ensemble distributions assuming Gaussianity, yielding probabilities and 5th/95th percentile bounds; risks were smoothed with 20-year LOESS and 95% CIs.

- The leading principal component of tree-ring predictors correlates strongly with ERA5 PNW JJA t2m (r = 0.87, p < 0.01); the calibration model explains 78% of variance over 1950–2017. Despite fewer predictors in earlier centuries, the model retains skill back to 950 CE (e.g., ~45% variance explained prior to 1400 CE; positive RE and CE). - 2021 exhibits the highest reconstructed PNW JJA temperatures since 950 CE. ERA5-based probability for 2021 within 1950–2021 is ~0.004%; reconstruction-based probabilities place 2021 even further outside historical variability. - The modern warm period (1979–2021) is among the longest continuous warm intervals, with the highest mean anomaly: +1.27 °C relative to 1951–1980; even excluding 2021, 1979–2020 remains the warmest. The second-warmest 43-year period occurs during the MCA (1045–1087 CE) with +0.68 °C; MCA warmest decades remain ~0.59 °C cooler than post-1979. - 2021 is at least +1.5 °C warmer than any identified MCA positive departure; within the instrumental era, 2021 exceeds the previous record (2015) by +0.40 °C. - Distributional shift: comparing 950–1900 to 950–2021 shows a positive skew in summer temperatures, consistent with anthropogenic CO₂ forcing impacts on North American summer temperature distributions. - Warming rates are unprecedented: the fastest 20-year increase occurs 1993–2012 with +3.3 °C; the next fastest is 1107–1126 with +2.9 °C. Over 50-year windows, warming since the 1970s is the highest in the record. - Agreement exists among reconstruction, ERA5, and CMIP6 historical simulations over shared periods. Observed 2021 (+3.6 °C) exceeds the 95th percentile of CMIP6 SSP2-4.5 2005–2035; 2015 (+3.2 °C) exceeds the 95th percentile for 1999–2029. - CMIP6 SSP2-4.5 projections (2022–2100) show median JJA warming > +3.0 °C relative to the long-term pre-Industrial (range +1.5 to +6.0 °C), with increased mean and variance. - Future risk: Under SSP2-4.5 and SSP3-7.0, annual probability of 2021-like summers (+3.6 °C) reaches ~50–60% by 2050. Even at ensemble 5th percentiles, 50% risk is reached by 2070 (SSP2-4.5) and 2057 (SSP3-7.0). Under SSP3-7.0, risk rises from ~5% contemporarily to ~94% by 2100 (90–97%).

The millennial-scale reconstruction demonstrates that the 2021 PNW summer heat is unprecedented in both magnitude and rate of change, reinforcing attribution studies that identify a dominant anthropogenic influence. The reconstruction captures regional internal variability and multi-decadal warm episodes (notably during the MCA) while showing modern extremes exceed past natural variability. Agreement among reconstruction, observations, and CMIP6 over shared periods supports the robustness of projections for the PNW, despite concerns about some CMIP6 models’ warm biases. The results imply that without rapid mitigation, 2021-like summers will become common by mid-century, compounding societal and ecological risks in regions historically unaccustomed to extreme heat. The study underscores the value of empirical tree-ring data to better constrain internal variability, thereby improving projections of future extreme heat risk and informing adaptation planning.

This study provides a high-skill, millennial-length reconstruction of PNW summer temperatures, establishing that the 2021 heat and recent warming rates are without precedent over the last ~1070 years. It contextualizes modern extremes relative to MCA variability and quantifies accelerating warming since the late 20th century. Coupled with CMIP6 scenarios, the analysis projects a rapid increase in the likelihood of 2021-like summers, reaching about a coin-flip annually by 2050 under intermediate-to-high forcing. These findings highlight urgent needs: emissions mitigation to limit warming (ideally SSP1-2.6 trajectories) and targeted adaptation strategies in temperate regions not historically prepared for extreme heat. Future work should further integrate proxy-based constraints on internal variability into climate model projections and refine regional risk estimates for extreme seasonal heat.

- Proxy network sparsity increases back in time (e.g., only two chronologies before 1400 CE), elevating uncertainty in early reconstruction segments despite positive validation metrics. - The reconstruction assumes Gaussian distributions for probability and risk estimates; real-world temperature distributions, especially under climate change, may deviate. - Tree-ring predictors end in 2018; ERA5 data were appended for 2019–2021, which may introduce a methodological discontinuity despite bias correction. - Some CMIP6 models exhibit warm biases (“hot model” problem) and current models may underestimate internal variability of extratropical extremes, affecting future risk quantification. - Regional coverage is limited to 17 sites and 29 chronologies; while highly correlated with summer temperature, tree growth may also reflect non-temperature influences and site-specific factors.