Attribution of drought trends on the Mongolian Plateau over the past decades

The Mongolian Plateau (MP) has undergone a significant drought trend in recent decades, presenting a substantial threat to local ecosystems and environments. The debate persists on whether this observed drought trend stems from external forcings or is a result of internal variability. Utilizing the large-ensemble simulations of the climate model and dynamical adjustment method, we have identified that the atmospheric circulation anomalies are the main drivers of drought trends in MP. A zonal atmospheric wave train, triggered by internally-generated warming of the North Atlantic sea surface temperature (NAS), is responsible for nearly 57% of the drought trend observed in MP. While external forcings could potentially induce a moistening trend in MP via direct thermodynamic processes, the atmospheric circulation anomalies linked to the forced NAS warming can not only offset its direct effect but also further amplify the drought trend in MP, accounting for 43% of the drought trend observed in MP.


Introduction
The recent drought in arid and semi-arid ecosystems has been aggravated based on the elevated temperature and variable precipitation, leaving farreaching imprints on the regional environment, water resources, and agriculture, including migration of several hundred thousand herders to the capital city, loss of lakes, and declines in grassland productivity (e.g.Sternberg 2011, Peng et al 2013, Tao et al 2015, Huang et al 2016b, Liu et al 2019).The Mongolian Plateau (MP), located in arid and semi-arid areas, is considered fragile and susceptible to desertification and degradation.In the early 21st century, the MP has been experiencing the hottest drought in past decades (e.g.Jin et al 2019, Liu et al 2019).Over the past 42 years, it encountered rapid warming about by 0.4 • C per decade, while the precipitation had a pronounced interannual and interdecadal fluctuation during 1979-2020 (Hessl et al 2018).Because climate trends exhibit apparent spatio-temporal inconsistency responding to its internal variability as well as to external forcing (e.g.Wallace et al 2012, Meehl et al 2013, Deser et al 2014, 2016, Gong et al 2019, 2021, Siler et al 2019, Wu et al 2021), on regional scale, understanding the aridity in MP caused by internal climate variability or external forcing has been a significant challenge (e.g.Piao et al 2017, Jin et al 2019, Liu et al 2019).
Many studies have emerged to investigate dryland changes in the past and predict their future trajectories, the results of which are highly inconsistent (e.g.Dai 2013).For instance, legions of observations and model simulations assess that an expansion of the drylands under continuous global warming (e.g.Dai 2011, 2013, Feng and Fu 2013, Sherwood and Fu 2014, Scheff and Frierson 2015, Huang et al 2016a, Park et al 2018, Zhou et al 2019, Hari et al 2020, Ukkola et al 2020, Li et al 2021), but almost the same time, other scholars believe that drought has been relieved over the past years and even growing green with enhanced precipitation in the coming decades (e.g.Fensholt et al 2012, Sheffield et al 2012, Donohue et al 2013, Zhu et al 2016, He et al 2019, Piao et al 2020, Berg and McColl 2021, Lian et al 2021).Knowledge of how climate change affected the drylands on regional scales in the past is essential for judging whether the drylands will expand in the future on a global scale.The MP is a typical dryland with the aridity index (AI) below 0.65.AI is a multivariate index to depict the comprehensive metrics of aridity (e.g.Feng and Fu 2013, Scheff and Frierson 2015, Huang et al 2016a, Park et al 2018, Berg and McColl 2021, Lian et al 2021).It combined the effects of precipitation deficits and high potential evapotranspiration (PET) in drought according to AI definition (e.g.Dai 2011, 2013, Scheff and Frierson 2015, Huang et al 2016a, Park et al 2018).Regional precipitation forecasts remain less confident, but prolonged aridity over MP is usually attributed to anomalous atmospheric circulation, creating favorable drought conditions.Numerous evidence shows that the persistent anticyclonic anomalies (high-pressure system) diminish cloud cover and precipitation and increase the PET, facilitating the formation of aridity (e.g.Liu 2012, Mueller and Seneviratne 2012, Hoerling et al 2014, Trenberth et al 2014, Berg et al 2016, Dai et al 2018).However, the causes and physical processes responsible for the MP drought are still unclear.In particular, to what extent do the external forcing and internal variability, particularly the internal dynamics, contribute to recent aridity trends of MP?
In this study, we employ the observational data, outputs from large-ensemble simulations of a climate model, and dynamical adjustment method to address these fundamental questions.It is crucial to improve our understanding of present and future climate change in MP.The structure of this paper is organized as follows: section 2 describes the data and methods used in this study.Section 3 quantitatively assesses the contributions of external forcing and internal variability to recent aridity trends in MP and unveils the physical mechanisms responsible for the contributions.A summary and discussion are displayed in section 4.

Data
The observational monthly precipitation and potential evaporation are obtained from the Climatic Research Unit (CRU) of the University of East Anglia (CRU_TS version 4.05) with a horizontal resolution of 0.5 • × 0.5 • covering the period from 1901 to 2020 (Harris et al 2021).We use monthly surface air temperature, sea level pressure (SLP), zonal and meridional winds (U, V), and geopotential height (Z) data at a 2.5 • × 2.5 • global grid from the fifth generation reanalysis provided by European Centre for Medium-Range Weather Forecast (ECMWF) (ERA5, Hersbach et al 2020).The sea surface temperature (SST) data are taken from the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST, version 5 dataset (Huang et al 2017).

Model simulations
Employing a large model-ensemble mean is a practical approach to separate the contributions of internal variability and external forcing (e.g.Deser et al 2014, 2016, Guo et al 2019, Gong et al 2021).In this study, we make full use of a 100member ensemble of simulations with the Max Planck Institute Earth System Model (MPI-ESM), which is a fully coupled atmosphere-ocean-land-sea ice climate model with a horizontal resolution of about 1.9 • × 1.9 • in the longitude-latitude grids.The MPI-ESM performs well in reproducing the observed climate in the Northern Hemisphere (e.g.Yu et al 2020, Gong et al 2021).The observed summer AI and precipitation trends are within the range of inter-member spread among 100 ensemble members of MPI-ESM (figure S1).All MPI-ESM members have the same external forcing (historical for 1920-2005 and RCP8.5 emissions scenario for 2006-2100) following phase 5 of the Coupled Model Intercomparison Project (CMIP5) protocol and only differ in their atmospheric initial conditions (Maher et al 2019).By averaging across ensemble members, the random sequences of internally generated variability in the individual realizations can be sufficiently muted to reveal the model's response to anthropogenic climate change.Therefore, the multiplemember mean (MME) of climate anomalies across 100 ensemble members is used to represent the forced response.The internal components are then deduced by subtracting these forced components from the observational data.

Methods
AI is the ratio of precipitation (P) to PET.The PET in CRU_TS version 4.05 is calculated using the Penman-Monteith formula (Allen et al 1998).Penman-Monteith's (PM) formulation incorporates the effects of wind and humidity, plus solar and longwave radiation.However, most of these data are not readily available in the MPI-ESM model.By comparing the climate state and trend distribution of AI, it is found that the AI index calculated based on the Thornthwaite method is consistent with that using the PM method in MP (figure S1).Therefore, we use the 'Thornthwaite method' to account for PET effects as an alternative in calculating the AI in the MPI-ESM model (Trenberth et al 2014).The 'Thornthwaite method' emerges as a globally optimal candidate due to its estimation of PET solely relying on air temperature and photoperiod (maximum sunshine duration) (Thornthwaite 1948, Pereira and Camargo 1989, Ahmadi and Fooladmand 2008, Li et al 2022).The North Atlantic SST (NAS) index is defined as the area-averaged SST anomalies in the North Atlantic (50 • -70 • N, 65 • -10 • W), which is most closely associated with MP summer drought (Iwao and Takahashi 2006, Piao et al 2017, Li et al 2023).This analysis focuses on the boreal summer, defined as the mean of June, July, and August (hereafter JJA).All the results are smoothed using 3 year running averages before computing the trends to eliminate the noise.Prior to the analysis, all the observed and model data are bilinearly interpolated to an ordinary resolution of 1 • × 1 • .The wave activity flux is employed to delineate the propagation of the planetary waves (Takaya and Nakamura 2001).The zonal and meridional components of the T-N wave activity flux can be written as: where ψ is the geostrophic stream function, which is defined as Φ⁄f (Φ and f are the seasonal mean geopotential height anomaly and Coriolis parameter, respectively); U and V are the mean zonal and meridional climatological winds, respectively; and |U| is the magnitude of the mean horizontal winds.The overbars represent the basic states, and primes represent perturbations.In this analysis, we use the 42 year monthly climatology means as the basic states and monthly anomalies derived from the regression states as the perturbations.
To illustrate the contributions of the dynamical and thermodynamical components to the aridity trends in MP, the dynamical adjustment method is employed in this study (Deser et al 2016).Here, we briefly describe our method.For the observed summer of 1979 as an example, we first randomly subsample 30 summer SLP patterns over the MP (37 • -54 • N, 87 • -123 • E) in ERA5 during 1979-2020 summers and compute their optimal linear combination to form a constructed SLP analog that resembles most closely the observed SLP in the summer of 1979.Next, we apply the coefficients derived from the optimal linear combination of SLP patterns to the corresponding precipitation and AI fields and obtain the 'dynamically induced' portion of precipitation and AI patterns.Random selection and optimal reconstruction are repeated 100 times to avoid overfitting.Therefore, dynamical components of precipitation and AI are obtained by the average of the 100 dynamical-related precipitation and AI fields to provide the best estimates of these fields over those 100 analogs (figure S2).The thermodynamically related changes in precipitation and AI are obtained by removing the dynamically induced changes from the original precipitation and AI data.
Furthermore, separating the dynamical contribution of internal atmospheric circulation, internal SSTdriven process, and forced SST-induced process on MP drought.The NAS-related AI trend is calculated as follows, in relation to the quantitative contributions of NAS anomaly to the observed drought trends in MP.First, we reconstruct the three-dimensional AI anomalies related to the NAS index based on the NAS-related AI anomaly pattern and the time series of the NAS index.Then, we calculate the AI trends in the reconstructed data to obtain the NAS-related AI trend.For the member i in the period τ (here τ = 1979-2020) can be expressed as: where r AI,NAS (i ) is the regression coefficient of dynamically-induces AI (AI) regressed onto the NAS of member i for 1979-2020 at each grid.∂NAS(i )   ∂t is the trend of the NAS of member i. ∂AI NAS(i ) ∂t is the NAS-related dynamically-induced AI trend of member i over the same period and varies among the ensemble members.The internal atmospheric circulation-induced AI trend in MP is obtained by subtracting the NAS-induced AI trend from the dynamically-induced AI trend.In addition, since the NAS trend is contributed both from internally generated interdecadal variations and externally forced warming, the MME of SST anomalies in the North Atlantic during 1979-2020 among 100ensemble members in MPI-ESM is used to estimate the forced NAS.We subtract the forced NAS trend from the observed NAS trend to obtain the internally generated NAS trend.

Dynamical and thermodynamical contribution to PRE trend
During the 1979-2020 summer, widespread decline trends of precipitation were observed over the MP, with the most pronounced drying in the northern parts of MP (figure 1(a)).These drought trends are attributed to an anomalous atmospheric circulation trend.The SLP is increasing over the whole MP with an intense positive anomaly center in eastern MP, extending into southern Inner Mongolia.We perform the dynamics adjustment method in the observed precipitation trend over the MP to demonstrate the influence of atmospheric circulation, also named dynamically induced variability, on precipitation trends.The spatial pattern of dynamicallyinduced precipitation trend is similar to the observed precipitation trends, but for more drought trend over the majority of MP (figure 1(b)).The location of the dynamically-induced precipitation trends in MP corresponds very well with the center of the highpressure system (figure 1(b)).After subtracting the dynamically induced component of observed precipitation trends, the residual thermodynamicallyinduced precipitation trend features widespread wetting trend in MP (figure 1(c)).It is noted that the thermodynamics and external forcings both contribute to a widespread increasing trend in precipitation across the entire MP, though they do not completely correspond in terms of spatial patterns.In fact, the thermodynamical contributions decomposed by the dynamic adjustment method can be also influences from localized internal thermal processes which may explain the spatial inconsistency between the thermodynamical and external forcings contributions.The forced SLP trends in MPI-ESM are very weak compared to the observed SLP trends, emphasizing that radiatively forced wetting in MP in the MPI-ESM model is mainly attributable to the thermodynamical contributions (figure 1(d)).AI is an essential indicator for assessing the adequacy of water supply to meet demand.To verify the role of precipitation changes in AI trends over MP, the observed summer AI trends from 1979 to 2020 are displayed in figure S3.The spatial pattern of AI trends agrees well with the precipitation trends in total, dynamically-induced, thermodynamically-induced, and MME (figure S3).Under global warming, widespread increase trends of PET were observed over the MP, with the pattern correlations between PET and AI patterns being notably weaker, falling below 0.7 (figure S4), compared to the stronger correlation of 0.92 between precipitation and AI.This suggests that precipitation change dominates the aridity trends over MP.

Time-varying PRE and AI trends
To quantitively demonstrate how the AI and precipitation trends have evolved since 1979 over the MP, the temporal evolutions of total, dynamic, thermodynamic, and forced-induced AI and precipitation trends in MP are displayed in figure 2. The xaxis indicates the ending year of the calculated trend starting from 1979.The total AI exhibited a pronounced drought trend from 1979 to 2011.With the extension of the period, the total drying trend of  AI is gradually decreasing, and dynamically induced trends coincide with the total trends, but with much stronger drought trends (figure 2(a) and (b)).The thermodynamically induced and externally forced AI trends in MP are relatively stable with the extension of the period, both showing wetting trends in MP (figure 2(a)).The magnitudes of thermodynamically induced and externally forced AI trends are similar during all the periods.The evolution of summer precipitation trends in MP is very similar to AI trends (figure 2(b)).Also, there is high consistency in the temporal evolutions of thermodynamically induced and externally forced precipitation trends in MP.It suggests that the thermodynamically-induced AI and precipitation trends are primarily attributed to the thermodynamical response to the external forcing.

Physical processes responsible for the PRE and AI changes
The above results indicate that the observed drought trends in MP during recent decades are primarily attributed to the persistent atmospheric circulation anomalies.Here a question naturally arises: what are the crucial physical processes responsible for drying trends in MP? figure 3(a) shows the regression map of the 200 hPa geopotential height (HGT200) and corresponding wave activity fluxes (WAF200) anomalies onto the normalized reversed dynamicallyinduced AI in MP during 1979-2020.There is an evident zonal stationary wave train emanating from the North Atlantic Ocean, propagating eastward to the Eurasian continent and eventually to the MP, with pronounced positive geopotential height anomalies in MP (figure 3(a)).Similar spatial structures are also found in the 700 hPa geopotential height (figure 3(b)), implying the quasi-barotropic characteristics of the zonal wave train.The persistent anticyclonic anomalies in MP reduce the cloud cover and precipitation and also increase solar radiation and PET, contributing to the formation of drought trends there.Previous studies have indicated that the NAS anomalies can trigger atmospheric Rossby wave train and influence the climate downstream (e.g.Piao et al 2017, Liu et al 2019).To demonstrate the possible linkage between the drying in MP and SST anomalies in the North Atlantic, figures 3(c) displays the regression map of summer SST in the North Atlantic onto the normalized reversed dynamically-induced AI in MP during 1979-2020.Remarkable positive SST anomalies are observed over the northern part of the North Atlantic.To further verify the effect of North Atlantic warming on the drought trend of MP, the NAS index is defined as the area-averaged summer SST anomalies in the northern North Atlantic (50 • -70 • N, 65 • -10 • W).The correlation coefficient between the dynamically induced AI index in MP and the NAS index is −0.74, exceeding the 99% confidence level.Figure 3(d) shows the regression maps of HGT200 and WAF200 onto the normalized NAS index.The spatial structure of the stationary wave train from the North Atlantic to MP is quite similar to that linked to the dynamically induced AI index.The detrended results further confirm the impact of NAS anomalies on the atmospheric circulation and resultant AI changes in MP (figure S5).The result suggests that North Atlantic warming has been a crucial factor contributing to the drought trend in MP in recent decades.
It should be noted that the observed SST warming in the North Atlantic is the combined result of the internally generated interdecadal variations and external forcing related to global warming (figure S6).This prompts the inquiry into the precise quantitative contributions of these factors to the observed drought trends in MP (see detail in Methods).The spatial pattern of the NAS-induced AI trend is highly consistent with the dynamically-induced AI trend, contributing to the 143% drought trend of observed AI in MP (figures 4(a) and S3(b)).The internal atmospheric circulation-induced AI trend pattern exhibits a north-south dipole distribution, accounting for about 14% of the observed AI trend in MP (figure 4(b)).Forced and internally generated NAS-induced AI trends in MP contribute about 86% and 57% of observed AI in MP, respectively (figures 4(c) and (d), table 1).It is noted that although the direct thermodynamical response of AI to external forcing exhibits a wetting trend in the majority of MP, the atmospheric circulation anomalies associated with the forced NAS warming not only offset the wetting effect of direct thermodynamic impact due to external forcing but also further exacerbate the aridification.External force contributes about 43% to the aridification of MP based on its combined effect during recent decades.These results suggest that the observed drought trends of AI in MP during recent decades are caused by a combination of internal variability and external forcing.

Summary and discussion
Based on observational data, large-ensemble simulations of the MPI-ESM model, and the dynamical adjustment method, we explore the causes behind the drought trend in MP and underlying physical mechanisms.First, we show that the dynamicallyinduced drought trend associated with the changes in atmospheric circulation in MP is the primary cause of the decrease in precipitation and the intensification of aridification in MP during 1979-2020.Then, the thermodynamical process induces a wetting trend in MP.The AI trend in MME of 100-ensemble members from the MPI-ESM model is highly consistent with the thermodynamically induced part, implying a direct influence of external forcing on AI.The spatial pattern of AI trends agrees well with the precipitation trends, suggesting a dominant role of precipitation changes on AI trends over MP.Furthermore, we find that an atmospheric wave train, triggered by internally interdecadal warming of the NAS, accounts for approximately 57% of the drought trend in MP.Despite some regions in the North Pacific also exhibiting positive SST anomalies associated with AI, theses warm SST anomalies over the North Pacific is mainly forced by the atmospheric circulation and exerts a minor influence on the summer precipitation changes in MP (figure not shown).
Although external forcing can lead to a wetting trend in MP through direct thermodynamic processes, the atmospheric wave train associated with forced North Atlantic warming not only offsets the wetting effect but also further induces the drying, together contributing about 43% aridification of MP during 1979-2020.Our findings indicate that the observed drought trend in MP is caused by a combination of internal variability and external forcing, and internal variability seems to play a more important role in the drought trend of MP than external forcing in recent decades.Our results provide an insight into understanding the present and future climate changes in MP.
Several issues remain to be explored in the future.For example, although this study employed MPI-ESM model owing 100 ensemble members to estimate the external forcing contribution and try to average out most of the noise from internal variability, the results from a single large-ensemble model may have certain limitations.The model uncertainty should be further examined using other large-ensemble models in the future.Despite these caveats, this paper suggests that when considering the impact of global warming on regional climate, it is essential not only to account for their direct effects but also to consider their indirect influences to fully understand the climatic impacts of global warming.Moreover, it is noted that the present drought trend of MP is largely influenced by internal variability, particularly the interdecadal changes in NAS.Therefore, to better predict and project future summer climate changes in MP, it is crucial not only to consider the potential changes in both the direct and indirect impacts of external forcing under scenarios of global warming in the future but also to take into account the possible effects of phase changes in NAS pattern.

Figure 1 .
Figure 1.(a) The summer (June-August) precipitation (shading) and SLP trends (contour) during 1979-2020 based on CRU data sets.(b) Dynamical contribution to (a).(c) Thermodynamic residual trends (a) minus (b).(d) The MME of summer precipitation and SLP trends during 1979-2020 in all 100 ensemble members from the MPI-ESM.Stippling regions indicate the significance exceeding 90% confidence level.Contours are sea level pressure trends in 0.2 hPa 42 years −1 increments starting at ±0.2 hPa.

Figure 3 .
Figure 3. (a) Regression maps of 200 hPa geopotential height (shading; unit: gpm) and corresponding wave activity fluxes (vectors; unit: m 2 s −2 ) anomalies onto the normalized dynamically-induced AI in MP during 1979-2020.(b) As in (a), but for 700hPa geopotential height (shading; unit: gpm), winds (vectors; unit: m s −1 ), (c) Regression map of SST (units: K) anomalies onto the normalized dynamically-induced AI in MP during 1979-2020.(d) As in (a) but for the NAS index.Rectangle in (c) denotes the region defined by the NAS index.Stippling regions indicate the significance exceeding 90% confidence level.