Delayed onset of the tropical Asian summer monsoon in CMIP6 can be linked to the cold bias over the Tibetan Plateau

Most global circulation models (GCMs) from the Coupled Model Intercomparison Project Phase 6 (CMIP6) simulate a delayed onset of the tropical Asian summer monsoon of 3–6 pentads when compared with the observations. However, a clear explanation of this model bias has yet to be developed. This study indicates that 23 of the 31 of CMIP6 models generate both the Tibetan Plateau (TP) cold bias and the delayed monsoon onset across tropical Asia. The aloft TP cold air temperature associated with these models tends to reduce the land–sea thermal contrast and monsoon circulation, and hence it generates a delayed onset for the tropical summer monsoon. Two sensitivity experiments based on a coupled ocean–atmosphere–land GCM, together with additional data analysis, further confirm the underlying connection between monsoon onset and temperature anomaly over the TP. Therefore, it is of great importance that we attempt to reduce the model bias associated with the simulation of monsoon onset by improving the physical process parameterization scheme related to the TP temperatures.


Introduction
The Asian summer monsoon (ASM) is the most active monsoon system on the planet and has multiple and complex interactions with Earth' surface, ocean, atmosphere, hydrosphere, and biosphere (Wang 2006).The onset of the tropical ASM is characterized by the arrival of the rainy season and the abrupt transition of the large-scale atmospheric circulation (Wang andLinHo 2002, Mathison et al 2018).Every year, agriculture planting, food production, economic activity, and the day-to-day lives of the inhabitants across the Asian monsoon regions are profoundly affected by the evolution and variability of the tropical ASM (Biemans et al 2016, Bombardi et al 2019).Therefore, improving our understanding of the onset of the tropical ASM could have significant socioeconomic implications.
Climatologically, the onset of the tropical ASM comprises three main stages: the Bay of Bengal (BOB) summer monsoon (late April to early May); the South China Sea (SCS) summer monsoon (mid-May); and the Indian summer monsoon (late May to early June) (Wang et al 2004, Wang and Ding 2008, Zhan et al 2016).Climate and Earth System Models offer an effective approach to the investigation of monsoon dynamics (Zhou et al 2020).Despite the notable improvements in capturing the mean and variability of summer monsoon rainfall over Asia, an accurate simulation of ASM evolution remains an unresolved challenge for the current set of state-of-the-art global circulation models (GCMs) from the Coupled Model Intercomparison Project, Phases 3 (CMIP3) to 6 (CMIP6) (Gusain et al 2020, Khadka et al 2021).Common biases within the tropical ASM simulations (relative to the observations), including a weaker monsoon circulation, reduced land-sea thermal contrast, delayed monsoon onset, and shortened monsoon duration, persist in these state-of-the-art models (Dong et al 2015, Kong et al 2022).Among these biases, the depiction of the onset of the tropical ASM is the critical (Ha et al 2020).
The origins and consequences of these systematic model biases are one of the three major questions addressed by CMIP6, which has the Grand Science Challenges of the World Climate Research Programme as its scientific backdrop (Eyring et al 2016).However, a clear identification of the key reasons for the systematic delayed onset biases remains to be fully established.Previous studies show that an increase in resolution can help to overcome some of the difficulties experienced when attempting to simulating the onset of the summer monsoon (Zhang et al 2018), but a coarse resolution is not the primary driver behind the generation of the biases associated with the onset of the summer monsoon (Ashfaq et al 2016).Ashfaq et al (2016) suggested that the bias in diabatic heating over the slopes of the Himalayas and the Karakoram Range during pre-monsoon period is noticeably stronger (weaker) in CMIP5 models that generate a relatively poor (better) simulation of the onset of the South Asian summer monsoon.Those results imply a possible role for the simulated Tibetan Plateau (TP) heating bias in the creation of the summer monsoon onset bias over Asia.In fact, most GCMs simulations have a systematic cold bias over the TP, especially over the western TP during the cold seasons (i.e.boreal spring and winter) (Zhu and Yang 2020, Peng et al 2022, Hu et al 2022c).These studies inspired us to investigate the potential relationship between the systematic delayed onset bias of the tropical ASM and the cold bias over the TP from CMIP3 to CMIP6 models, as TP heating is closely related to ASM onset (Zhang et al 2015, Li and Xiao 2021, Hu et al 2022a).
In this study, we aim to address the following questions.(1) How do the CMIP6 models simulate the climate mean tropical ASM onset dates?(2) Is the systematic onset bias in the tropical ASM linked to the performance in simulating TP heating?(3) If so, what are the underlying mechanisms that cause the onset bias response over the tropical ASM to the TP heating bias?

Data
We used multi-source observational datasets to validate the ability of the model to reproduce the onset of the monsoon.Precipitation data were obtained from the Climate Prediction Center Merged Analysis of Precipitation (CMAP) with a horizontal resolution of 2.5 • × 2.5 • (Xie and Arkin 1997), and the Multi-Source Weighted-Ensemble Precipitation (MSWEP) with a horizontal resolution of 0.1 • × 0.1 • (Beck et al 2019).MSWEP is currently provides the best gridded precipitation datasets covering tropical Asia (Wati et al 2022).The observational atmospheric circulation variables were obtained from the latest European Centre for Medium-Range Weather Forecasts reanalysis product (ERA5) (Hersbach et al 2020) and the Japan Meteorological Agency Japanese 55 year Reanalysis (JRA55) (Kobayashi et al 2015).
Daily and monthly variables from 31 GCMs were obtained from the historical outputs of CMIP6 (table S1).The multi-level dataset has 19 vertical levels for the monthly data, and 8 vertical levels (1000, 850, 700, 500, 250, 100, 50, and 10 hPa) for the daily data.To facilitate a comparison between the GCMs and the observations, the model data and the observations were interpolated onto a 1 • × 1 • spatial resolution using bilinear interpolation.The period of focus for our analysis was 1979-2014.

Methods
Following the Intergovernmental Panel on Climate Change Summary for Policymakers (Allan et al 2021), we used gray (purple) hatches to indicate areas where more than 66% (90%) of the 31 models agreed with the sign of the anomalies.The 66% (90%) indicate assessed likelihood of an outcome is likely (very likely) (Allan et al 2021).This significance test, which is based on the consensus of the individual models, is an important measure for testing multi-model ensembles (MMEs) (Power et al 2012).
The onset of the climatological tropical ASM at each grid point was defined as the first pentad during which the relative pentad mean rainfall (i.e., the difference between the pentad mean rainfall and the January mean rainfall) exceeded 5 mm d −1 .To retain both the seasonal and sub-seasonal signals, the sum of the first 12 harmonics of the relative pentad mean rainfall were used in our calculations (Wang and LinHo 2002).This definition has been used in may previous studies (Zou andZhou 2015, Moon andHa 2020).

Model description
We used the Flexible Global Ocean-Atmosphere-Land System Model finite volume version 2 (FGOALS-f2) in this study.FGOALS-f2 participated in CMIP6 (He et al 2019a) and was developed by the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China.In this model, four components (atmosphere, ocean, land, and sea ice) are fully coupled using Coupler Version 7 (cpl7) from the Community Earth System Model Version 1.0 (CESM1.0)(Bao et al 2013).FGOALS-f2 performs well with respect to the basic patterns of atmospheric circulation and precipitation over the TP and Asian regions (Bao et al 2013, Zhou et al 2020).

Design of numerical experiments
We designed two experiments.The control run (CTRL) was freely integrated for 100 years using the coupled general circulation model (CGCM) experiment with FGOALS-f2, but only the last 20 years were analyzed.The sensitivity experiments conducted using the CGCM within FGOALS-f2-referred to as EXP_WarmTP-contained 20 members.Each EXP_WarmTP run used the initial fields from the last 20 years in the CTRL experiments and was integrated from 1 March to 30 September.To reduce the cold bias over the TP in CTRL (figure S1(a)), EXP_WarmTP was forced by the linearly increasing air temperature anomalies from the outermost ellipse (0 • C) to the center (30 • N, 80 • E; 6 • C) in the horizonal direction (figure S1(b)), and the warm center of the vertical direction was around 300-250 hPa (figure S1(c)).The correlation between the tropical ASM onset bias and the cold bias over the TP is significant from March and reaches its maximum in May (section 3.2).Consequently, the 20 EXP_WarmTP experiments were forced by linearly increasing the temperature anomalies from 1 March (0 • C) to 30 April (6 • C) and then retaining this maximum value between 1 and 31 May.

Simulations of climate mean tropical ASM onset under the current climate
Climatologically, the tropical ASM appears first in the southeastern BOB around the 24th-25th pentad, appears in the SCS around the 27th-28th pentad, and finally reaches the Indian Subcontinent around the 28th-29th pentad (figures 1(a) and (b)).The spatial evolution of the climatological mean tropical ASM in both the CMAP and MSWEP datasets is consistent with previous studies (Wang andLinHo 2002, Bombardi et al 2019).The onsets of the summer monsoon over the Arabian Sea (AS), Indian Subcontinent, BOB, Indo-China peninsula, and SCS exhibits a remarkably coherent variation; i.e. a synchronized advance and delayed onset (Hu et al 2022b, Liu andDuan 2023).Thus, we defined the area bounded by 5 • -20 • N, 60 • -120 • E as the tropical ASM region to be analyzed in this study.The conventional propagating features of the tropical ASM are accurately reproduced by the CMIP6 models.However, most of this generation of GCMs fails to capture the climatological mean tropical ASM onset dates (Ha et al 2020, Park et al 2020, García-Franco et al 2021).For instance, most CMIP6 GCMs tend to excessively delay the monsoon onset date over tropical Asia.In particular, more than 90% of the 31 GCMs simulate a systematically delayed monsoon onset over the western Indian subcontinent, BOB, and northern SCS.The tropical ASM onset is systematically delayed by about 3-6 pentads in the MME results (figures 1(d) and (e)).In addition, the meridional temperature gradient (MTG) in the mid-upper troposphere between 20 • N and 5 • N is another important indicator that can be used to determine the monsoon onset (Mao et al 2002, Dai et al 2013), and it is also delayed by about 3 pentads (figure 1(f)).Here, we used the 250 hPa temperature to represent the mid-upper troposphere because the daily data from CMIP6 contain only two levels (500 and 250 hPa) between 500 and 200 hPa.The delayed onset bias of the climatological mean tropical ASM in CMIP5 (Tung et al 2014, Dong et al 2015, García-Franco et al 2021) is also evident in the CMIP6 models and is independent of the selection of 'observations' and monsoon onset indicators.

Cause of the delayed onset bias for the tropical ASM
As the onset of the summer monsoon mainly occurs during late May, examining the simulated variables in May should assist our understanding of the delayed monsoon onset bias.The land-sea thermal contrast is the major driver of the formation of the monsoon; therefore, the thermal conditions in Asia will be examined first.A large cold bias from the surface to the mid-upper troposphere in the MME is evident over most parts of Asia and especially over the TP.The cold bias over the TP was generated by over 90% of the CMIP6 models.The maximum cold bias occurs between 300 and 200 hPa (figures 2(a) and (b)).The low-level cold bias is concentrated over the northcentral TP, but due to the strong advection, the cold temperature center in the mid-upper troposphere is shifted to the downwind side of the low-level cold bias center.(Wu and Zhang 1998, Wu et al 2015, 2018).
Pearson correlation analysis of the 31 inter-model between the averaged onset bias in the tropical ASM and the temperature bias over the TP shows significant negative correlations around those areas with a large cold bias.This significant correlation occurs from March at lower levels and reached a maximum in the mid-upper troposphere during May.This is consistent with the development of the cold bias over the TP.Therefore, cold air over the TP is accompanied by a delayed monsoon onset over the tropical ASM region (figure 2).Of the 31 CMIP6 models, 23 generated both the cold temperature bias over the TP and a uniform delayed onset in the tropical ASM region (figure S2).Furthermore, the intensity of the South From the above analysis, we note the significant contemporaneous relationships between the tropical ASM onset bias and the cold bias over the TP in the CMIP6 models.To further examine the precursor effect of the cold bias over the TP on the tropical ASM onset bias, we used the pentad data.As shown in figure 3(a), most of the CMIP6 models underestimate the temperature over the TP all year round, especially during the summer monsoon season (April-August).Air temperature over the TP peaks in July-August and falls in September in both the observations and models.The abrupt warming over the TP tends to occur before the onset of the summer monsoon and is closely linked to the summer monsoon onset (Wu and Zhang 1998).However, the climatological abrupt warming over the TP in the 31 MME occurs around the 24th pentad, about 6 pentads later than that in the observations (18th pentad).
Therefore, the delayed abrupt warming over the TP in the 31 MME corresponds to the delayed onset of the tropical ASM.Furthermore, pentad correlations shown in figure 3(a) show the precursor effect of the cold bias over the TP on the tropical ASM onset bias, which starts early in April and becomes significant after the warming of the TP.The air temperature over the ASM is much colder in the MME especially during May, leading to a weak land-sea thermal structure (figure 3(b)).Anomalous heating during the seasonal transition plays an important role in determining the anomalous state of the summer monsoon in the Northern Hemisphere (Yasunari 1991), thus it is to be expected that the largest correlation between the temperature bias over the TP and the tropical ASM onset bias occurs in May.Furthermore, the climatological zonal vertical wind shear accelerates with the growth of the temperature over the TP.Because of the delayed warming of the TP, U-shear significant weakens in MME during the reversal of the monsoon circulation from winter (negative U-shear) to summer (positive U-shear) (figure 3(c)), indicating a delayed reversal of summer monsoon circulation in MME (figure S4).The weak thermal forcing over the TP in the MME reduces the land-sea thermal contrast, impeding the reversal of the MTG with respect to the tropical ASM onset.Moreover, the weak thermal forcing over the TP decreases the monsoon circulation and weakens the intensity and northwesterly development of the SAH.These circulation conditions subdue the arrival of the rainy season.Therefore, most of the CMIP6 models tend to simulate a delayed onset of the tropical ASM.
Many previous studies have demonstrated the close relationship between atmospheric heating over the TP and ASM onset over interannual timescales (He et al 2019b, Lu et al 2021, Hu et al 2022a).The time series of TP temperature anomalies in May have significant negative correlations with the annual onset dates of the tropical ASM, especially over the AS, BOB, Indo-China Peninsula, and SCS for observation and models (figure S5).According to these relationships (figures S5(a1-31)), the tropical ASM onset bias related to the cold bias over the TP can be obtained by using a linear regression relationship in each individual model.The climatological tropical ASM onset bias is broadly consistent with the simulated TPcold-bias-related tropical ASM onset bias, and the inter-model correlation between them reaches 0.52 (figure S6).However, the values of TP-cold-biasrelated onset bias is smaller or larger than the climatological onset bias, implying that the delayed monsoon onset responses to the TP cold bias may be nonlinear, and other factors might also increase or offset the delayed onset of the tropical ASM onset bias in the CMIP6 models.

Verification analysis using FGOALS-f2
To investigate the underlying mechanisms associated with the systematic delayed onset response of the tropical ASM to the TP cold bias in the CMIP6 models, we performed two experiments using FGOALS-f2.The CTRL experiment accurately reproduces the conventional propagating features of the tropical ASM but also shows delayed onset dates for the tropical ASM (figure 4(a)).Moreover, the huge cold column over the TP is also evident in the CTRL run (figure S1(a)).In the EXP_WarmTP run, the land-sea thermal contrast, SAH, and summer monsoon circulation structure are stronger than those in the CTRL run (figure 4(b)) because of the warming-forcing over the TP (figure S1).The reversal of the MTG from the winter monsoon to the summer monsoon over the tropical ASM region is advanced by about a month (figure 4(d)).This advanced reversal of the MTG reflects a large range of the early onset of the tropical ASM (figure 4(c)).This is especially the case over the Indo-China Peninsula and SCS, where the confidence levels exceed 90%.These results illustrate that the land-sea thermal contrast, SAH, and summer monsoon circulation of the tropical ASM can be significantly enhanced by decreasing the cold bias over the TP.Thereby, the onset of the tropical ASM in the EXP_WarmTP run is earlier when compared with the CTRL experiment.

Conclusions
Accurately simulating tropical ASM onset dates using the current suite of state-of-the-art climate models remains a serious challenge.In this study using multisource reanalysis datasets as our 'observations' , we assessed the ability of the CMIP6 GCMs to simulate the climatological mean tropical ASM onset.Our results show that most of the CMIP6 GCMs tend to generate a delay of 3-6 pentads in the monsoon onset date across tropical Asia.Simultaneously, large cold temperature biases, extending from the surface to the mid-upper troposphere, are evident over Asia, and especially over the southern TP, in the CMIP6 models.The abrupt climatological warming over the TP in the CMIP6 models occurs around 6 pentads later than that in the observations.Of the 31 CMIP6 models, 23 simulated both the cold temperature bias over the TP and the uniform delayed onset of the tropical ASM.The tropical ASM onset bias is in negative proportion to the TP temperature bias during spring, which reflects the potential contributions of the TP cold bias to the tropical ASM delayed onset bias in the CMIP6 models.A large cold bias over the TP can reduce the land-sea thermal contrast and summer monsoon circulation over Asia and delay the onset of the tropical ASM.In addition, the cold bias over the TP leads to D Hu et al a weakened and southeastward SAH.This weak SAH impedes the upper tropospheric divergence pumping and suppresses precipitation.Therefore, the tropical ASM onset is delayed in these models.Moreover, two experiments carried out using FGOALS-f2 further confirmed the conclusion that reducing the cold bias over the TP help to overcome the delayed onset bias in the tropical ASM.

Discussions
Our results suggest that reducing the temperature bias over the TP can offset the errors associated with simulating the tropical ASM onset bias to a certain extent, but it is not the only factor responsible for the delayed monsoon onset.Most of the models present a negative relationship between the tropical ASM onset bias and the temperature bias over the TP, and the points of them fall in the second and fourth quadrants (figure 2(c)).A few models (represented by gray markers) fall in the first and third quadrants, and these models do not support the negative relationship.Among these models, some (AWI-ESM-1-1-LR, MPI-ESM-1-2-HAM, MPI-ESM1-2-LR, and NESM3) show advanced onset despite the cold bias over the TP.The advanced onset in these models may be related to the positive TCI (figure 2(f)).The cold bias in the tropical Indian Ocean exceeds the cold bias over the TP.The thermal conditions and surface sea temperature over the Indian Ocean (Zou and Zhou 2015) are also important factors with respect to determing the tropical ASM onset.The onset bias in the remaining models (including ACCESS-ESM1-5, INM-CM4-8 CAMS-CSM1-0, and CanESM5) may be attributed to the Somali cross-equational flow and the BOB cross-equational flow, as the development of the Somali and BOB cross-equational flows is vital to the formation of the ASM (Sun et al 2023).ACCESS-ESM1-5 and INM-CM4-8 (CAMS-CSM1-0 and CanESM5) show a delayed (advanced) onset accompanied by weaker (stronger) Somali and BOB cross-equational flows (figure S8) than seen in the observations.Furthermore, the summer monsoon onset dates over the AS and Indian Subcontinent are only slightly delayed in the EXP_WarmTP run when compared with the CTRL run.Inter-model correlation indicates that the delayed summer monsoon onset over the AS and Indian subcontinent may be more related to the AS cold surface sea temperature bias (Levine and Turner 2011, Li and Yang 2017, Wang et al 2017) rather than the cold bias over the TP (figure S7).Nevertheless, other factors that influence the simulation of the onset of the tropical ASM will require in-depth study in the future.
Given that the systematic tropical ASM delayed onset bias can be attributed to the contribution of the TP cold bias, a question is raised: what is the cause of the large cold bias over the TP?The cold bias in the mid-upper troposphere over the TP may be attributed to the low-level cold bias (figure 2(b)).The cold bias of the surface temperatures over the TP results in a weakened sensible heat flux over the TP (Wang et al 2022).The diabatic heating related to the surface heat plays a dominant role in the variability of the column temperature over the TP (Wu and Zhang 1998).Previous studies have suggested that improving land surface-cloud processes, snow cover parameterization, boundary layer, surface turbulent fluxes, land surface parameters (albedo and vegetation), and meteorological characteristic (cloud types and cloud cover) will be crucial for reducing the cold bias over the TP (Chen et al 2017, Lee et al 2019, Wu et al 2021).In addition to these local factors, the tropical SST bias can result in an apparent cold bias over the TP through an anomalous heat flux divergence over the TP (Wu et al 2022).The improvements in the performance of the circulation and land-atmosphere processes over the TP may be a crucial step toward overcoming the delayed monsoon onset biases in the CMIP6 models.

Figure 1 .
Figure 1.Onset dates for the tropical ASM based on the (a) CMAP and (b) MSWEP observational datasets and (c) the 31 MME from CMIP6 between 1979 and 2014.Also shown are the simulated onset biases between the 31 MME and the (d) CMAP and (e) MSWEP datasets (right panels; units: pentads).Gray and purple hatched areas in (d) and (e) indicate that more than 66% and 90%, respectively, of the individual models agree on the sign of the anomalies.(f) Zero isoline of the climatological MTG between 20 • N and 5 • N. Thick magenta, red, and blue lines indicate the JRA55, ERA5, and 31 MME data, respectively, and dashed lines indicate the 31 individual GCMs.

Figure 2 .
Figure 2. Simulation biases associated with the (a) climatological mean temperature for May at 250 hPa and (b) monthly air temperature over the TP (area-averaged over 20 • -40 • N, 60 • -100 • E) from 500 to 150 hPa (units: • C). Green contours in (a) and (b) indicate areas where the inter-model correlation between the onset bias of the tropical ASM and air temperature exceeds the 90% confidence level.Blue and red contours in (a) indicate the SAH at 100 hPa, denoted by 16 560 and 16 640 geopotential heights for the 31 MME and the observations (units: gpm).In (a) and (b), gray and purple hatched areas indicate that more than 66% and 90%, respectively, of the individual models agree on the sign of the anomalies.The orange curve in (a) shows the area of the Tibetan Plateau with an average altitude >2500 m.Also shown are scatter diagrams between the bias of climatological (c) tropical ASM onset (area-averaged over 5 • -20 • N, 60 • -120 • E), (d) U-shear (area-averaged over 5 • -20 • N, 60 • -120 • E), (e) SAHI (area-averaged over 20 • -35 • N, 65 • -95 • E), and (f) TCI difference between the average air temperature at 250 hPa over TP and that over the tropical Indian Ocean (10 • S-10 • N, 60 • -100 • E) versus the air temperature at 250 hPa over the TP (T250).The pink circle indicates the 31 MME.Black lines are linear least-squares regression fits to the data points, and values in the upper-right indicate the corresponding correlation coefficients.Superscripts * * * , * * , and * represent confidence levels p < 0.1, p < 0.05, and p < 0.01, respectively.Models from the same institutions are shown with the same markers.

Figure 3 .
Figure 3. Pentad evolution of the (a) air temperature at 250 hPa averaged over the TP (20 • -40 • N, 60-100 • E; units: K) and the inter-model correlation between the air temperature bias at 250 hPa over the TP with the tropical ASM onset bias (green line for CMAP data, dark-green line for MSWEP data), (b) zonally averaged (60 • -100 • E) air temperature at 250 hPa (units: K), and (c) zonally averaged (60 • -120 • E) zonal vertical wind shear bias (units: m s −1 ) for climatological mean in observation (purple contours) and simulated bias in MME (shading).Only bias with more than 66% of the individual models agreed on the sign of the anomalies are drawn for (b) and (c).In subplot (a), red and blue lines indicate observations and the 31MME, respectively, and light blue shading indicates the range of the 31 models.Gray and black dashed curves in (a) indicate the 90% and 95% confidence levels, respectively, for the correlation analysis.

Figure 4 .
Figure 4. Composite (a) differences for the onset dates of tropical ASM (units: pentad) between CTRL experiments and observations based on MSWEP data, (b) the ratio of the differences of several indexes (i.e.TP air temperature at 250 hPa, TCI, SAHI, U-shear in May and average onset date of the tropical ASM) between EXP_WarmTP and CTRL experiments to the standard deviation in CTRL experiments, (c) the differences for onset dates of tropical ASM (units: pentad), and (d) Zero isoline of the climatological MTG between the EXP_WarmTP and CTRL experiments.Gray and purple crossed areas denote values exceeding the 90% and 95% confidence levels, respectively.Orange dot in (b) indicate differences exceeding 95% confidence levels.