Quantitative attribution of historical anthropogenic warming on the extreme rainfall event over Henan in July 2021

The ‘21·7’ Henan extreme rainfall event (HNER) caused severe damage and many fatalities. The daily precipitation during this event (from 1200 UTC on 19 July 2021–1200 UTC on 20 July 2021) was 552.5 mm and the maximum hourly precipitation was 201.9 mm (at 0900 UTC on 20 July 2021). Previous studies have suggested that an evaluation of the role of anthropogenic climate change in extreme rainfall events is crucial in disaster prevention and mitigation under the current global climate crisis. We examined the changes in the coverage and intensity of extreme rainfall during the ‘21·7’ HNER event under anthropogenic climate change using a set of convective permitting simulations. Our results showed that the regional-average magnitude of the 48 h accumulated rainfall during the ‘21·7’ HNER was increased by 5.7% (95% confidence interval: 4%–11%), which is in agreement with the Clausius–Clapeyron rate, while the area of extreme rainfall (⩾500 mm) increased by 29.9% (95% confidence interval: 21%–40%) as a result of anthropogenic climate change over the Henan region during the late 20th century. Anthropogenic climate change has led to a warm moist tongue over the target region, which has increased the column-integrated water vapor content and induced an anomalous cyclone–anticyclone pair. Anthropogenic warming has caused stronger southerly and southeasterly winds, leading to stronger convergence in the lower troposphere, stronger updrafts in the mid-troposphere and stronger divergent winds in the upper levels. These effects have all contributed to the increase in rainfall. These results enhance our understanding of the dynamic effects of anthropogenic warming on the ‘21·7’ HNER and provide additional evidence that anthropogenic warming increased the magnitude of the ‘21·7’ HNER in China.


Introduction
Heavy rainfall frequently occurs in China as a result of interactions among the East Asian summer monsoon, typhoon and the country's complex terrain (Meng et al 2019, Zhao et al 2021).The frequency and intensity of heavy precipitation events are likely to increase over most land regions under the effects of global climate change (IPCC 2021).Anthropogenic warming has increased the probability of daily-scale extreme precipitation events by increasing the water vapor content of the atmosphere (Zhou et al 2021) and will lead to a wetter and more variable hydro-climate on daily to interannual timescales (Zhang et al 2021(Zhang et al , 2022)).Henan Province was affected by a persistent heavy rainfall event from 17-23 July 2021.The capital city, Zhengzhou, recorded a maximum hourly rainfall of 201.9 mm (Zhao et al 2022), the highest rainfall ever recorded over mainland China (Xu et al 2022a(Xu et al , 2022b)).This event, referred to as the '21•7' Henan extreme rainfall event (HNER), triggered devastating floods.About 15 million people were affected, with 398 fatalities and direct economic losses of about 120 billion yuan over Henan Province (www.mem.gov.cn/xw/yjglbgzdt/202201/t20220123_407199.shtml).
Previous studies of the '21•7' HNER have mainly focused on synoptic weather systems, such as the upper level high-pressure ridge and mid-level trough (Sun et al 2022), the water vapor channel modulated by the subtropical high and the westerly belt (Gao et al 2022), the persistent double low-level jets (Luo and Du 2022), the impacts of the upper tropospheric cold low (Cai et al 2022), the interactions of binary typhoons Infa (2021) and Cempaka (2021) as well as their remote impact (Xu et al 2023).Besides, the mesoscale convective systems are also of interest, such as the embedded mesoscale vortex (Fu et al 2022), the well-organized meso-γ-scale convective system (Yin et al 2022).These comprehensive inquiries have shed significant light on this extreme weather event, providing a robust foundation for future research.In essence, it is believed that the multi-scale interactions facilitated by multiple simultaneous favorable factors have led to the '21•7' HNER.Here the links between anthropogenic climate change and this record-breaking rainfall event are investigated.
Increasing attention is being paid to the detection of extreme weather and climate events and their attribution to anthropogenic climate change.Stott et al (2004) analyzed the extreme heatwave in Europe in 2003 that killed thousands of people.The Bulletin of the American Meteorological Society has published a special issue on the theme of explaining extreme events from a climate perspective annually since 2012, which has greatly advanced climate change attribution research (Peterson et al 2013, Herring et al 2018).Attribution methods can be broadly split into two categories: (1) methods using long-term observational data to determine changes in the probability and magnitude of specific events; and (2) methods using numerical simulations to contrast how an event would manifest in a world with and without anthropogenic climate change (NASEM 2016).In the attribution analysis of specific extreme weather events, extreme heat and cold events have the highest level of confidence, whereas only medium levels of confidence can be achieved in the attribution of extreme precipitation events; strong convective storms have little or no confidence (Zhai et al 2018).This is mainly due to the smaller spatiotemporal scale of the precipitation field and the lower ability of models to simulate precipitation fields compared with temperature fields.
With respect to the '21•7' HNER, the concern of this study, Qin et al (2022) has investigated the attribution of climate change and revealed that convective organization resulted in much higher precipitation extremes than unorganized convection.Wang et al (2022) showed that anthropogenic warming and wetting increased the average rainfall by about 7.5% within the target area during this event.Furthermore, their results emphasized a positive thermodynamic feedback between the convection and the release of latent heat.Studies have also indicated that the system directly producing this extreme rainfall event was a mesoscale convective storm with strong vertical velocity (Chen et al 2022, Yin et al 2022), which motivates us to further explore the dynamic mechanisms.Here, we conduct an in-depth, comprehensive study to explore the effects of the dynamic and thermodynamic mechanisms of anthropogenic warming on its occurrence and development.These findings will help us to better understand extreme rainfall events under the current global climate crisis.

Data and methods
Hourly multisource integrated precipitation data (0.1 We used the Weather Research and Forecasting model with the Advanced Research core (WRF-ARW) version 4.4 (Skamarock et al 2019) to examine the impact of anthropogenic climate change on '21•7' HNER.Using the initial and lateral boundary conditions of the ERA-5 from the European Centre for Medium Range Weather Forecasts (Hersbach et al 2020), 72 h simulations with double two-way interactive nested domains (D01 and D02) were initialized at 0000 UTC on 18 July 2021 with the outermost domain (D01) centered at (30.0 • N, 120.0 • E).The targeted region is bounded by 31.5 • N-36.5 • N and 110.0 • E-116.0 • E, which corresponds to the administrative region of Henan Province, as denoted by the thick black curve in figure 1(a).This area is much smaller than the model domains D01 and D02, which ensures the results of the following analysis are minimally affected by the domain boundaries.The simulation covered the development, enhancement and subsequent weakening stages of this extreme rainfall event.The model top was set at 50 hPa with 50 sigma layers in the vertical direction.The model configuration, hereafter referred to as the control (CTL) experiment, is given in table 1.We focus on the 48 h period between 0000 UTC on 19 July 2021 and 0000 UTC on 21 July 2021, the one covering recordbreaking precipitation at both daily and hourly scales.The simulated radar reflectivity was calculated by NCAR Command Language (NCL) built-in function for WRF real-data which incorporates the intercept  The widely used pseudo-global warming (PGW) approach was used to quantitatively evaluate the warming and wetting aspects of anthropogenic climate change (Shepherd 2016, Patricola and Wehner 2018, Michaelis et al 2019).A companion sensitivity experiment (NAT experiment) was conducted with identical model configurations to the CTL experiment, apart from adding a 'delta' to the initial and boundary conditions of the CTL experiment.The value of 'delta' was derived from the ensemble mean differences of five global climate models (CESM2, GFDL-CM4, FGOALS-g3, ACCESS-CM2 and ACCESS-ESM1-5) in the Coupled Model Intercomparison Project Phase 6 (CMIP6) between the all-forcing historical run and its natural forcing only counterpart during the time period 1984-2014.The CTL and NAT experiments only differ in the thermodynamic states of the initial and boundary conditions while keeping the same large-scale circulation, thus the concerned differences could be attributed to anthropogenic climate change.Variations in the timescale up to the decadal range and the bias of individual climate models were reduced through the use of the long-term average and model ensemble mean.
It is important to highlight that there are two main sources of uncertainty in these types of attribution studies.First, the uncertainty caused by global climate models' potential inaccuracies in depicting atmospheric internal variability at a regional scale.This can be mitigated to some extent by employing an ensemble mean of diverse climate models, as has been done in this study.The second source of uncertainty arises from unresolved processes and their interactions at the sub-grid scale in convective permitting simulations.In response to this, a 30-member ensemble simulation using the stochastic kinetic energy backscatter scheme was performed for both the CTL and NAT experiments in our research.Any resultant uncertainties were subsequently quantified using the 30 members of each experiment in our later analysis.Here, we employed the Student's t-test to assess the statistical significance of the differences between the CTRL and NAT experiments.Differences with a significance level exceeding 95% were deemed statistically significant.

Control simulation and verification
Figures 1(a) and (b) show the spatial distribution of the 48 h accumulated rainfall from the observations and CTL simulation from D02 (3 km resolution), respectively, between 0000 UTC on 19 July 2021 and 0000 UTC on 21 July 2021.The model reproduced the general distribution pattern of the rain belt reasonably well, especially in the area of heavy rainfall (⩾250 mm) near the city of Zhengzhou.The model simulated a maximum 48 h accumulated precipitation of 832.5 mm, close to the maximum of 818.0 mm in the observations.The center of heavy rainfall located at (33.9 • N, 112.4 • E), southwest of Zhengzhou, was larger in the CTL simulation than in the observations.Given our study's primary focus is on the evolution of rain bands and systems inducing rainfall, rather than the intensity of precipitation at a single point under the anthropogenic warming, thus the simulation results are generally acceptable.
Figure 1(c) compares the time series of the regional-average accumulated and 3 h rainfall between the observations and simulation over the target area.Both the accumulated observed and simulated rainfall showed similar increasing trends, although the simulation showed slightly more rainfall than the observations.The simulated 3 h rainfall matched the observations well during the concentrated period of heavy rainfall (1800 UTC on 19 July 2021-0900 UTC on 20 July 2021).To evaluate the simulated pattern of the direct rainfall-producing system, namely the convective belts, we compared the real-time radar observations with the simulated composite radar reflectivity to evaluate the simulated pattern of the direct rainfall-producing system, the convective belts.The 48 h averaged composite radar reflectivity analysis reveals that the intensity and distribution pattern of the CTL simulation closely resemble the observed data (figures 1(d) and (e)).
In general, the CTL experiment simulated well the pattern of the rain belt and the intensity of the HNER and their evolution over time.The CTL experiment also simulated the distinct convective cell structures reasonably well, although these were stronger than in the observations.Given this consistency between the CTL simulation and the observations, we used the large-area output from D01 (12 km) and highresolution output from D02 (3 km) to assess the impacts of anthropogenic climate change on the HNER.

Anthropogenic climate change impact
Compared with the CTL simulation, the domainaveraged precipitation decreased slightly in the NAT experiment from 186.3 mm (figure 1(b)) to 176.5 mm (figure 2(a)).However, the extent and intensity of heavy precipitation changed significantly.For example, the maximum 48 h accumulated rainfall increased from 703.6-832.5 mm in the NAT and CTL experiments when anthropogenic climate change was considered.Figure 2(b) shows the difference between the CTL and NAT experiments.Precipitation increased across most of Henan Province, with the largest increase in the central region near the city of Zhengzhou, whereas precipitation decreased in some of the northern areas.This indicates that anthropogenic climate change increased the intensity of rainfall in the HNER and also shifted the location of rainfall southward.
Figure 2(c) shows the probability distribution functions of the regional-average 48 h accumulated rainfall over the target area for the CTL simulation and the NAT experiment.We used these two probability distribution functions to quantitatively assess the variability of extreme precipitation.The two probability distribution functions were roughly equivalent when the simulated precipitation was <350 mm, indicating no significant impact.We consider 350 mm as a potential threshold for damage.The probability distribution function of the CTL simulation was clearly shifted to the right and was greater than that of the NAT experiment over the high extreme end.This change was more significant in the area affected by heavy rainfall (figure 2(d)), with the areas of heavy rainfall of 400, 500 and 600 mm increased by 23.8% (13%-31%), 29.9% (21%-40%) and 33.8% (25%-45%), respectively.All quoted uncertainties fall within the 5%-95% range.All these signals indicate that the magnitude of the HNER was increased under forcing by anthropogenic climate change.

Dynamic and thermodynamic contribution
Figure 3(a) shows the 'delta' value, which reflects the anthropogenic signal.This will help us to understand the cause of the increased rainfall associated with anthropogenic climate change.A warming trend was clearly seen throughout the whole troposphere and a moistening trend in the lower troposphere.The spatial distribution averaged over 1000-700 hPa (figure 3(b)) showed a warm moist tongue over the target region.These changes were evident and consistent with previously estimated results using different model ensembles of the CMIP6 (Qin et al 2022, Wang et al 2022) and CMIP5 (Wang et al 2019), which demonstrates that the downscaling estimation of the climate model simulations allows a quantitative assessment of the impacts of long-term anthropogenic climate change on the HNER.
The Clausius-Clapeyron (CC) relationship shows that global warming increases the moisture-holding capacity of the atmosphere (Trenberth et al 2003).Figures 3(c) and (d) show the column-integrated moisture flux and its convergence in the CTL and NAT experiments, respectively.Both experiments have showed that abundant water vapor was continuously transported to the target area from the south and southeast, resulting in a convergence at the region.This was much stronger in the CTL simulation.For instance, the maximum columnintegrated moisture flux is 700-800 kg m −1 s −1 in the NAT simulation (figure 3(d)) while it reached 800-900 kg m −1 s −1 in the CTL experiment (figure 3(c)), with an enhancement of more than 10%.Given the approximate 0.6 K increase in temperature (figure 3(b)), the expected enhancement is only 4.2% based on the CC equation, which suggests that the dynamic impact is also crucial under the anthropogenic warming in addition to the thermal impact.Figures 3(e) and (f) quantitatively show the difference between these two simulations.We identified an interesting anomalous circulation feature: a cyclone-anticyclone pair in our target area Zhengzhou, the area with the strongest wind convergence and water vapor flux, is exactly in the middle of this pair of anomalous circulations.This corresponds well with the development of the HNER.
We compared the characteristics of the circulation at different heights in the troposphere for the CTL and NAT experiments and their difference (figures 4(a)-(f)).The southerly and southeasterly winds at 850 hPa in the lower troposphere over the Henan region were stronger in the CTL experiment (figure 4(a)) than in the NAT experiment (figure 4(b)).The convergence was stronger, which, in turn, created a stronger cyclonic circulation (figures 4(a) and (c)).The vertical upward motion at 500 hPa in the mid-troposphere was stronger in the CTL experiment and the centers with a large upward motion were more concentrated (figure 4(a)).The velocity potential and divergent winds at 200 hPa in the upper troposphere were are also significantly stronger in the CTL simulation (figure 4(d)) than in the NAT experiment (figure 4(e)).This showed that, under anthropogenic warming, there was a consistent positive dynamic feedback (figures 4(c) and (f))throughout the troposphere during the HNER, which resulted in a significant increase in precipitation.Furthermore, the CTL experiment displayed increased convective instability (as shown in figures 5(a) and (b)).

Concluding remarks
We analyzed the influence of anthropogenic climate change on the '21•7' extreme rainfall event in Henan using observations and convective permitting simulations under the conditional extreme event attribution framework.The results show that the regional-average magnitude of the two-day accumulated precipitation increased by 5.7% (4%-11%) and the area of heavy rainfall (⩾500 mm) increased by 29.9% (21%-40%).This indicates that anthropogenic climate change increased the magnitude of this event.This increase is consistent with the thermodynamic expectations from the CC relationship.Such thermodynamic processes play a vital role in amplifying the intensity of rainfall events in a warmer world.Beyond the thermodynamic considerations, our study uncovers the dynamical impacts of anthropogenic climate change on the HNER.
The estimated anthropogenic influence showed a warm moist tongue over the target region, which increased the column-integrated water vapor content and induced an anomalous cyclone-anticyclone circulation pair.The difference between the two experiments suggests that anthropogenic warming and wetting enhanced the southerly and southeasterly winds.The convergence of these winds in the lower troposphere accelerated the updraft in the midtroposphere and the divergent winds in the upper level.This provided consistent positive dynamic feedback and enhanced the HNER.
There are two caveats to this study.First, our results cannot explain how anthropogenic warming affected the frequency of occurrence of this extreme rainfall event as the simulation is only for one specific event.Second, because the PGW conditional attribution framework is only a partial attribution (Shepherd 2016), the potential dynamic variation in the initial and lateral boundary fields was not explicitly addressed because the large-scale dynamics can interact with the thermodynamics (Qian et al 2022).Given these uncertainties, most previous studies on the detection and attribution of extreme weather events have focused on the thermodynamic aspects (NASEM 2016, Zhai et al 2018), whereas our findings emphasize the dynamic effects of anthropogenic warming on the HNER.These results are an extension of the work of Wang et al (2022) and Qin et al (2022) and provide additional evidence that anthropogenic warming increased the magnitude of the HNER.

Figure 1 .
Figure 1.The 48 h accumulated rainfall (units: mm) between 0000 UTC on 19 July 2021 and 0000 UTC on 21 July 2021 for the observational precipitation product (a), and the thick black curve denotes our target region of Henan province, the white star represents the city of Zhengzhou and the red plus signs indicate the location of radar stations and the control experiment (b).(c) Time series of the regional-average accumulated (solid lines) and 3 h rainfall (histograms, units: mm) in the observations (black) and control simulation (red) over the area (31.5-36.5 • N, 110.0-116.0• E) from 0000 UTC on 18 July 2021-0000 UTC on 21 July 2021.The corresponding ensemble member ranges are shaded (light pink) for the control simulation.48 h averaged composite radar reflectivity (units: dBZ; shading) between 0000 UTC on 19 July 2021 and 0000 UTC on 21 July 2021 for the (d) radar observations and (e) the control experiment.

Figure 2 .
Figure 2. (a) The 48 h accumulated rainfall (units: mm) of the natural run sensitivity experiment between 0000 UTC on 19 July 2021 and 0000 UTC on 21 July 2021 and (b) the difference between these results and the control simulation.The black dots represent areas where the difference is significant at a 95% confidence level.(c) Probability distribution functions of the regional-average 48 h accumulated rainfall over the area (31.5-36.5 • N, 110.0-116.0• E) for the control simulation (red solid line) and the natural run sensitivity experiment (blue solid line).The corresponding ensemble member ranges are shaded (light pink, control simulation; light blue, natural run sensitivity experiment).(d) Probability distribution functions of the area affected by heavy rainfall over the same area as in part (c) for the control (red solid line) and natural run sensitivity (blue solid line) experiments and the relative percentage increase calculated by (CTL minus NAT)/CTRL, note log scaling of the left y-axis.And the corresponding ensemble member ranges are shaded as (c).

Figure 3 .
Figure 3. (a) Effects of anthropogenic climate change on the average air temperature (units: K) and specific humidity (units: g kg −1 ) profiles over the WRF D01 region (5.0-40.0• N, 90.0-150.0• E) estimated by the CMIP6 selected model ensemble mean (historical run minus natural run) from 1984 to 2014.(b) And its horizontal distribution averaged over 1000 hPa-700 hPa for temperature (contours, units: K) and specific humidity (shading, units: g kg −1 ).The horizontal distribution of the column-integrated moisture flux (vectors), its magnitude (shaded, unit: kg m −1 s −1 ) and its convergence (contour, unit: 10 −5 g kg −1 s −1 ) produced by the (c) control and (d) sensitivity experiments at 0000 UTC on 20 July 2021 and (e), (f) their differences.The red rectangles in parts (b)-(e) are the target domain of the Henan region (latter means the same).(f) Enlarged display of the Henan region shown in part (e); the brown dots in parts (e) and (f) show the location of Zhengzhou while the green dots represent areas where the difference for column-integrated moisture flux is significant at a 95% confidence level.

Figure 4 .
Figure 4. Horizontal wind fields (vectors; units: m s −1 ) at 850 hPa and the vertical velocity (shading; units: m s −1 ) at 500 hPa produced by the (a) control and (b) sensitivity experiments and (c) their difference at 0000 UTC on 20 July 2021.Velocity potential (shading; units: 10 6 m s −1 ) and divergent wind fields (vectors; units: m s −1 ) at 200 hPa produced by the (d) control and (e) sensitivity experiments and (f) their difference.The black dots in (e) and (f) represent areas where the difference for velocity potential is significant at a 95% confidence level.

Table 1 .
Description of the model details and experimental setup.