Potential benefits of limiting global warming for the mitigation of temperature extremes in China

The study addresses how limiting global warming to 1.5 °C versus 2.0 °C above preindustrial levels affects temperature extremes across China. Motivated by the Paris Agreement targets and China’s “30–60 Dual-Carbon Target,” the authors note that China has already experienced substantial warming and frequent extreme events with major socioeconomic impacts. While CMIP6 models offer improved projections under SSP-RCP scenarios, uncertainties and biases persist, particularly at regional scales, necessitating bias correction. The research aims to evaluate CMIP6 models against observations, project changes in extreme temperature indices at 1.5 °C and 2.0 °C global warming, and quantify the benefits for China of limiting warming to 1.5 °C.

Prior work shows China’s mean temperature has increased by ~1.3–1.7 °C per century since 1900, with rising extremes. CMIP6 provides enhanced physics and scenario design over CMIP5 and has shown improved skill in simulating climate extremes, though results are mixed and substantial model spread remains. Studies highlight the need for bias correction to reduce regional-scale errors due to coarse resolution, process representation, and forcing uncertainties. Yet few studies have focused on bias-corrected projections of temperature extremes in China using CMIP6 at specified global warming levels, leaving a gap that this study addresses.

- Indices: Twelve ETCCDI temperature extreme indices were analyzed, covering intensity (TXx, TXn, TNx, TNn), absolute frequency (SU, ID, TR, FD), and relative frequency (TX90p, TX10p, TN90p, TN10p). - Data: Observational dataset CN05.1 (0.25° grid, 1961–2014) was used for validation. Daily max/min temperatures from 12 CMIP6 GCMs were obtained for historical, SSP245 (SSP2-4.5), and SSP585 (SSP5-8.5) scenarios. - Preprocessing and bias correction: Models were bilinearly interpolated to match the observation grid for comparison. Bias correction used the Equidistant Cumulative Distribution Function (EDCDF) quantile mapping method, assuming percentile-wise differences between simulations and observations in the historical period persist into the future. The correction adjusts the model CDF to the observed CDF in the reference period and applies the adjustment to future projections. - Warming level definition: Global annual mean temperature time series for each model were smoothed with a 20-year triangular moving average. The arrival years for 1.5 °C and 2.0 °C above 1850–1900 were identified per model as the first years crossing those thresholds. For each threshold, a 20-year window (−9 to +10 years around the arrival year) defined the future period. The historical baseline period was 1995–2014 (same length). Incremental impacts were quantified as changes from 1.5 °C to 2.0 °C periods. - Evaluation metrics: Model performance was assessed using inter-annual variability skill (IVS = (σm/σo)^2), root-mean-square error (RMSE), and standardized RMSE (RMSEstd) for spatial bias. Ensemble mean performance was compared with individual models before and after bias correction. - Analysis: Temporal evolution, spatial bias maps, regional averages, spatial patterns of changes at 1.5 °C and 2.0 °C under SSP245 and SSP585, and spatial maps of incremental changes (2.0–1.5 °C) were produced. Statistical significance was evaluated with Student’s t-tests at 5% level where shown.

Model evaluation after bias correction: - Temporal variability: Bias-corrected CMIP6 ensemble captured interannual variability of most indices; ensemble IVS generally < 1.0 and improved versus raw models. For the corrected ensemble mean, many indices had IVS < 0.4, notably TNx and FD. Some raw models (e.g., CNRM-ESM2-1, HadGEM3-GC31-LL, MIROC6) had larger IVS but improved after correction. - Spatial biases: Raw models showed warm biases for TXx/TNx and cold biases for TXn/TNn over most of China (west often >5 °C in magnitude). EDCDF correction reduced these to roughly −1 to 1 °C. For absolute indices, raw models overestimated ID, TR, FD and underestimated SU; post-correction biases were mostly within ~2 days, though FD remained overestimated in the west. Percentile indices were comparatively well simulated in both raw and corrected runs; TN90p tended to be underestimated in the west and slightly overestimated in the southeast; TN10p showed positive biases even after correction. - Overall, bias-corrected ensemble means outperformed individual models (lower RMSEstd; better IVS), with many models moving into the best (low-IVS, low-RMSEstd) quadrant after correction. Timing of warming levels (ensemble means): - SSP245: 1.5 °C arrives in 2030 (range 2021–2040); 2.0 °C in 2046 (2037–2056). - SSP585: 1.5 °C in 2026; 2.0 °C in 2039 (about 7 years earlier than SSP245). CanESM5 warms fastest (1.5 °C as early as 2012–2013; 2.0 °C by 2022–2024), and MIROC6 the slowest. Projected changes at 1.5 °C and 2.0 °C relative to 1995–2014 (area-averaged over China; multi-model means with model ranges): - Extremal indices (°C): - TXx: +1.35 (0.32–2.04) at 1.5 °C SSP245; +1.93 (1.18–2.53) at 2.0 °C SSP245. +1.50 (0.22–2.47) at 1.5 °C SSP585; +2.11 (0.96–3.05) at 2.0 °C SSP585. - TXn: +0.81 (−0.09–1.55) and +1.56 (0.76–2.38) under SSP245; +1.10 (0.31–1.79) and +1.52 (0.85–2.38) under SSP585. - TNx: +1.26 (0.25–2.56) and +1.90 (1.00–3.39) under SSP245; +1.07 (−0.98–2.43) and +1.71 (−0.35–2.98) under SSP585. - TNn: +1.02 (−1.40–2.02) and +1.80 (−0.45–2.52) under SSP245; +1.27 (−1.63–2.42) and +2.06 (−1.01–3.61) under SSP585. - Absolute frequency (days): - SU: +11.40 (−1.40–2.02) likely a typographic issue in range; multi-model means: +11.40 at 1.5 °C SSP245; +16.95 (11.23–22.92) at 2.0 °C SSP245; +11.95 (5.25–19.29) at 1.5 °C SSP585; +17.97 (11.03–25.52) at 2.0 °C SSP585. - ID: −6.19 (−10.57 to −3.02) at 1.5 °C SSP245; −9.56 (−14.10 to −5.06) at 2.0 °C SSP245; −6.46 (−10.89 to −3.59) at 1.5 °C SSP585; −9.85 (−14.71 to −6.39) at 2.0 °C SSP585. - TR: +8.40 (3.39–13.64) and +13.31 (9.05–20.95) under SSP245; +8.28 (−28.01–13.43) and +13.42 (−25.10–18.08) under SSP585 (large negative outlier from IPSL-CM6A-LR drives the wide range). - FD: −11.98 (−27.12 to −4.30) and −17.18 (−29.79 to −10.76) under SSP245; −12.35 (−30.57 to −0.64) and −17.70 (−34.59 to −6.95) under SSP585. - Relative frequency (% of days): - TX90p: +5.13 (1.21–8.50) at 1.5 °C and +7.92 (4.65–10.76) at 2.0 °C SSP245; +5.38 (1.04–9.41) and +8.19 (4.29–12.26) SSP585. - TX10p: −2.13 (−3.48 to −1.20) and −3.29 (−4.76 to −2.08) SSP245; −2.41 (−3.62 to −1.55) and −3.17 (−4.46–0.90) SSP585. - TN90p: +5.70 (1.29–10.93) and +8.63 (5.05–14.64) SSP245; +5.12 (−2.99–10.35) and +8.07 (−0.59–12.50) SSP585. - TN10p: −2.51 (−4.50–2.13) and −3.84 (−5.56 to −0.22) SSP245; −2.87 (−4.84–2.85) and −4.00 (−5.61–1.17) SSP585. The decrease in TN10p is most notable (~−4%) at 2.0 °C under SSP585. Spatial patterns: - Warming increases for TXx/TNx/TXn/TNn across China, with larger changes under higher warming and emission scenarios. SU and TR increase broadly (smaller changes on the Tibetan Plateau), while ID and FD decrease markedly, especially in western China (>20 days). TX90p and TN90p increase nationwide, while TX10p and TN10p decrease, most strongly in central-western regions. Incremental impact from 1.5 °C to 2.0 °C: - For most indices, incremental changes from the extra 0.5 °C exceed 25% across large parts of China, implying disproportionate intensification during the final 0.5 °C. Largest increments for TXx, SU, TX90p in the southwest; for TXn and TNx in the northwest. Smaller increments for TXx, SU, TX90p in the northeast and for TNn, TR, TN10p over the Tibetan Plateau. Region-averaged incremental changes exceed 25% for all indices; e.g., TXn increases by ~49.46% (SSP245) vs ~30.11% (SSP585). Under SSP585, increments are similar in pattern but smaller in magnitude and with larger inter-model spread (outliers for TXx, TNx, TR, TX90p).

The research question—how much China benefits in terms of reduced temperature extremes by limiting global warming to 1.5 °C rather than 2.0 °C—is addressed through bias-corrected CMIP6 projections at specified global warming thresholds. Bias correction substantially improves model skill in simulating historical extremes, lending greater confidence to projections. Projections show consistent intensification of warm extremes and reductions in cold extremes nationwide, with strong spatial heterogeneity (e.g., larger SU/TR increases outside the Tibetan Plateau; pronounced decreases in ID/FD in western China). Crucially, the additional 0.5 °C of warming from 1.5 °C to 2.0 °C disproportionately amplifies extreme indices across most of China, indicating significant benefits of stringent mitigation. The similarity of spatial patterns between SSP245 and SSP585 at the same warming levels suggests that the pathway to warming is less critical than the warming level itself for near-term extremes, though higher-emission scenarios show larger inter-model spread and uncertainty. These findings underscore the policy relevance of pursuing the 1.5 °C target to reduce heat risk, with particular attention to vulnerable regions in the southwest and northwest where incremental impacts are largest.

This study contributes a bias-corrected, multi-model CMIP6 assessment of temperature extremes in China at 1.5 °C and 2.0 °C global warming levels. Key conclusions are: (1) Bias correction markedly improves historical simulation of extreme indices; (2) Warm extremes (TXx, TNx, TX90p, TN90p, SU, TR) intensify and cold extremes (TXn, TNn, TX10p, TN10p, ID, FD) decline under both warming levels, with larger changes at 2.0 °C; (3) The additional 0.5 °C of warming leads to disproportionate (>25%) increases in extreme indices across most regions, particularly in the southwest (TXx, SU, TX90p) and northwest (TXn, TNx), highlighting substantial benefits to China from limiting warming to 1.5 °C. Future work should expand model ensembles and scenarios (including stabilized-warming experiments), improve representation and correction over complex terrains (e.g., Tibetan Plateau), and further investigate sources of uncertainty to enhance robustness of regional projections for adaptation planning.

Despite bias correction, notable uncertainties remain: residual biases persist for some indices (e.g., FD and TN10p in western China/Tibetan Plateau) likely due to complex topography and coarse GCM resolution; inter-model differences in forcings, internal variability magnitude, climate sensitivity, and warming timing definitions; observational uncertainties in complex terrain (station sparsity); scenario-related uncertainties more relevant for farther-future projections; inheritance of baseline-period biases; and the use of transient warming periods rather than near-equilibrium 1.5 °C/2.0 °C worlds. Under SSP585, greater spread and outliers (e.g., TR from IPSL-CM6A-LR) further increase uncertainty. Results should thus be applied with caution, especially at local scales.