Assessment of Dynamical Downscaling Performance over CORDEX East Asia using MPAS-A Global Variable Resolution Model: Climatology, Seasonal Cycle, and Extreme Events

A 29-year variable resolution climate simulation is conducted from January 1988 to December 2016 using the Model for Prediction Across Scale-Atmosphere (MPAS-Atmosphere) with prescribed sea surface temperatures obtained from ERA-Interim reanalysis. The global variable resolution con�guration employed a mesh re�nement of 92-25km centered over East Asia (MPAS-A experiment, hereafter), which could preserve multi-scale features within the same model framework. The evaluation of MPAS-A is performed for precipitation, near-surface air temperature, and circulation features against observed climate using combined observational datasets. The MPAS-A experiment exhibited large-scale deviations due to the absence of observational constraints, especially for the northward displacement of rain belts, excessive rainfall over the tropical ocean, and stationary surface air temperature biases tied to speci�c regions. These deviations can be explained by the simulated circulation, moisture transports, and the relationship between precipitation and convective available potential energy (CAPE). While the simulated seasonal cycles and frequency are dominated by large-scale deviation patterns, certain climate indices demonstrated lower sensitivity in the re�ned regions, particularly regarding extreme rainfall. This �nding underscores the robustness and potential of the variable resolution (VR) approach in obtaining regional information within a single model framework.


Introduction
Dominated by the East Asia monsoon, the climate in East Asia demonstrates high variability and uncertainties across various spatiotemporal scales (Qian et al., 2002;Wang et al., 2001).Situated within the monsoon belt, East Asia regions are particularly vulnerable to the destructive effects of abnormal monsoon patterns, which can lead to a range of natural disasters (Chen et al., 2019).Simulating the complex and diverse climate of East Asia poses a signi cant challenge for climate models.Global climate models (GCMs) have emerged as a promising tool for climate processes across the world, including in East Asia (Jiang et al., 2020;Lin et al., 2008;Wang et al., 2005;Xin et al., 2020;Zhou & Li, 2002).Despite the recent advances in GCMs, they still exhibit certain de ciencies in capturing realistic monsoon systems with su cient spatial variability (Jiang et al., 2020;Sperber et al., 2013), which can be attributed to their low resolutions and shortcomings in physics parameterization.Due to relatively coarse grid resolution and limitations in computational resources and dataset storage, GCMs also have seldom been run at ne resolutions suitable for analyzing daily or sub-daily extreme events (Xu et al., 2022).
For regional details of climate, various downscaling methods are employed through statistical and dynamical approaches.The dynamical downscaling conducts a model with higher resolution to improve the explicit representation of the mesoscale processes, underlying topographies, and land-use heterogeneity.The use of regional climate models (RCMs) driven by coarse-resolution General Circulation Models (GCMs) is a common approach employed by many studies to acquire reliable information on a regional scale through nesting techniques (Giorgi, 2019).Considering the improvement of capturing daily extreme rainfall by higher resolution (Kim et al., 2019), RCMs demonstrate su cient capabilities for the extreme climate events over East Asia (Hui et al., 2022;Kim et al., 2021;Tang et al., 2022;Zou & Zhou, 2016, 2022).However, the driving models are crucial for the RCM's performance and dominate the largescale features of the nested models (Giorgi et al., 2001;Oliger & Sundström, 1978), which is one of the debatable issues in RCM studies (Filippo Giorgi, 2019).Systematic biases could be inherited from the forcing data (Bukovsky et al., 2013;Davies, 2014;Warner et al., 1997) and RCMs require su cient time for higher-resolution climate details produced or downscaled from coarse-resolution driving models (Denis et al., 2002).Some methods are applied over the LBCs and model interior to maintain the consistency between RCMs and driving models, such as two-way nesting (Lorenz & Jacob, 2005) and spectral nudging techniques (Storch et al., 2000).Traditional nesting techniques often exhibit abrupt transitions between GCMs and RCMs, which can be mitigated by using high-resolution GCMs instead.However, this approach can be computationally intensive for long-term period simulations.Variableresolution GCMs (VRGCMs) offer a cost-effective solution that combines the strengths of both GCMs and RCMs.By incorporating self-consistent interactions between global and regional scales, VRGCMs offer a seamless and e cient approach to capturing multiscale features.Prior research has indicated that the utilization of a stretch grid method can yield cost-effective and re ned regional simulations (Fox-Rabinovitz et al., 2006;Fox-Rabinovitz et al., 1997).However, the stretched grid modifying the resolution can interact with physics, resulting in the zonally nonuniform characteristics in simulations of aquaplanet.Despite this, stretched grid models have been employed in studies involving East Asia summer monsoon simulations (Zhou & Li, 2002).Recently, advanced unstructured grids such as the cubed sphere (Guba et al., 2014;Zarzycki et al., 2014) and Voronoi grids (Ringler et al., 2008) present tolerant grid imprints with large-scale error characterized by the coarsest resolution.The re ned region improves a solution at a local scale and has minimal in uence on global error (Ringler et al., 2011).
Focusing on the model physics response to mesh re nement, the testing and development of the Model for Prediction Across Scale-Atmosphere (MPAS-Atmosphere) using Voronoi grids showed that it produces acceptable asymmetric errors of heating and precipitation compared to quasi-uniform high-resolution con gurations (Hagos et al., 2013;Rauscher et al., 2013;Rauscher & Ringler, 2014;Zhao et al., 2016).
Following these veri cation studies on variable-resolution grids, many studies have emerged that utilize VRGCMs for regional detailed atmospheric features in the real world, notably including the Variable-Resolution Community Earth System Model (VR-CESM) and MPAS-A.The suitability of VR-CESM for longterm regional is demonstrated, such as climate modeling of California (Huang et al., 2016), Tibetan Plateau (Rahimi et al., 2019), and climate extremes (Xu et al., 2022).Focusing on the climate simulations using MPAS-A or Voronoi grids, regional details, and upscale effects (Sakaguchi et al., 2015(Sakaguchi et al., , 2016)), convection-permitting hindcast of extreme precipitation (Zhao et al., 2019), and resolving mesoscale convective systems (Feng et al., 2021) are included.
It should be noted that regional climate information for East Asia is primarily obtained from Regional Climate Models (RCMs).The performance of RCM is evaluated by using global reanalysis at the lateral boundary and comparing against observations.Multi-model intercomparison efforts are required for the accuracy and reliability of RCMs.The Coordinated Regional Downscaling Experiment (CORDEX) project, established by the World Climate Research Program (WCRP), is one such collaborative initiative that seeks to advance and apply regional climate downscaling through global partnerships.Giorgi et al. (Giorgi et al., 2009) documented this initiative, which has been further elaborated by Giorgi and Gutowski (Giorgi & Gutowski, 2015).The project comprises multiple domains worldwide, including a focus on East Asia (CORDEX-EA).Additionally, A notable example is the Weather Research and Forecasting (WRF) model and relevant studies acknowledge its capability for capturing reasonable climate characteristics including spectral nudging (Tang et al., 2018), the sensitivity of rainfall to cumulus parameterization (Niu et al., 2020), Future projection of extreme precipitation (Hui et al., 2022).Considering that MPAS-A has been demonstrated as a promising tool for regional climate simulations, it produces comparable climate details similar to some nesting regional models, especially when compared with WRF simulations using similar con gurations (Huang et al., 2022;Kramer et al., 2020;Liang et al., 2021;Tam et al., 2021).Almost all the RCMs involved in the CORDEX-EA project are forced with sea surface temperature (SST) obtained directly from the GCMs or reanalysis datasets.Undertaking a long-term downscaling experiment utilizing MPAS-A with the same prescribed SST and additional fractional area coverage of sea ice in East Asia would be a worthwhile endeavor.
The objectives of this study include the following: (1) to assess and evaluate the regional climate details through a global MPAS-A long-term hindcast, especially for the rainfall and near-surface air temperature.
(2) to investigate factors contributing to the simulated biases.We perform and evaluate 27-yr MPAS-A simulations over East Asia using variable resolution grids, with prescribed SST and fractional area coverage of sea-ice from ERA-Interim reanalysis (Dee et al., 2011).In Section 2, we outline the experimental design, reference datasets, and evaluation methods.Section 3 provides detailed analyses and evaluations of the performance of our simulations and partly explanations of simulated deviations.Finally, in Section 4, we summarize the results obtained from our study.

Model and experimental design
This study employs the MPAS-Atmosphere v7.0 (Skamarock et al., 2012), a fully compressible nonhydrostatic atmospheric model, to conduct variable resolution experiments by mesh spacing with unstructured centroidal Voronoi tessellations (SCVTs) over East Asia from 1988 to 2016.The global variable resolution grid is designed with four times grid re nement at the domain center (32.5°N, 105°E), where the grid spacing smoothly varies from ~ 25km to ~ 92km outside the re nement.Following the CORDEX-EA-framework which utilizes a horizontal resolution of approximately 25km, the nerresolution domain covers a large portion of East Asia and the western Paci c region.The vertical coordinate system used by MPAS-A is the geometric-height terrain-following coordinate system (Klemp, 2011), which has 55 vertical layers and a model top at 30km.The time step is set to 120 seconds, which is subject to the nest cell spacing over the SCVT mesh.Additionally, the coordinate surface is gradually relaxed to a constant height to eliminate small-scale terrain features.Table 1 provides a summary of the main con gurations employed in the MPAS-A experiments.The adopted convective parameterization scheme is the modi ed tiedtke scheme, which uses a mass ux approach and convective available potential energy closure to represent both deep and shallow convection.This scheme has been described in detail by Zhang et al. (Zhang et al., 2011).For the bulk microphysical parameterization, the WSM6 scheme developed by Hong and Lim (Hong & Lim, 2006) is implemented, which includes six prognostic water substance variables, namely, water vapor, cloud water, cloud ice, snow, rain, and graupel.The land surface is represented using the Noah scheme (Chen & Dudhia, 2001), which is a four-layer soil temperature and moisture model with canopy moisture and snow cover simulations.The planetary boundary layer and surface layer processes are represented using the Yonsei University (YSU) scheme (Hong, 2010;Hong et al., 2006) and the Monin-Obukhov scheme, respectively.The Rapid Radiative Transfer Model for GCMs (RRTMG) scheme (Iacono et al., 2000;Mlawer et al., 1997) is used to quantify the longwave and shortwave radiative transfer processes at 30-minute time steps.The SST and fractional area coverage of sea ice used in the simulations are prescribed at a 24-hour interval and obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis with a resolution of approximately 80 km.The MPAS simulations are conducted for 29 years from 1988 to 2016, which are initialized on 1st January 1988.The rst two years are adopted as the spinup time.(3) The GPCP V3.2 Monthly dataset offers a gridded (Level 3) dataset that contains globally homogeneous processed records of precipitation estimates.The spatial resolution is 0.5 degrees, and it provides monthly temporal resolution.The dataset covers the period from 1983 to 2020.The merged satellite-gauge precipitation estimate in this dataset is used in our study for the evaluation of the monthly rainfall.
(4) The GPCP Daily Version 1.3 Combined Precipitation Data Set (GPCP1DD, hereafter) is available for the period starting from October 1996 up to the present with a spatial resolution of 1.0 degrees.The dataset combines information from both satellite-based and in situ precipitation observations, ensuring comprehensive coverage and accuracy in capturing precipitation dynamics.The GPCP1DD dataset from 1997 to 2016 was used to evaluate daily precipitation.
(5) the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis with a spatial resolution of 0.75 degrees is utilized for evaluations against large-scale features, including the winds, moisture, geopotential height, and CAPE.Additionally, the near-surface air temperature from this reanalysis is also used for completing the combined observation.
To facilitate a combined observation dataset of the rainfall and air temperature, both the reference data and the MPAS-A simulation results are interpolated onto a common grid with a resolution of 0.25°×0.25°using the bilinear interpolation method.The combined datasets prioritize the utilization of more reliable data based on site observations (CN05.1,APHRO).To enhance the overall data coverage in regions where CN05.1 and APHRO data may have limitations, such as oceanic regions, satellite precipitation data (GPCP V3.2, GPCP1DD), and reanalysis air temperature data (ERA-Interim) complete rest regions.They are also utilized to extend the coverage beyond the limited temporal range of these stational datasets.By this means, we obtain combined monthly precipitation covering 1990-2016, as well as daily mean, maximum, and minimum near-surface air temperature.However, the combined daily precipitation data is limited to the period from 1997 to 2016 due to the utilization of GPCP1DD data.Similarly, the calculation of extreme precipitation indices based on daily precipitation is also restricted to this time range (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016).Despite these limitations, the provided datasets still offer valuable insights into long-term precipitation patterns and temperature variations within the speci ed time frame.
To quantitatively evaluate the performance of the simulations, some statistical metrics are chosen, including the temporal correlation coe cient (TCC), spatial correlation coe cient (SCC), and root mean square error (RMSE), which are calculated between the reference data and the model simulations.In addition, the vertically integrated moisture ux (Q ux) is computed across isobaric surfaces (1) to assess the impact of moisture transport on rainfall deviations. 1 where q is speci c humidity, p is pressure, is the horizontal wind vector and g is the gravitational acceleration.
Considering the moisture and energy, convective available potential energy (CAPE) is calculated in the reanalysis and simulation for comparisons as follows (2).
where Z_EL is the equilibrium level; Z_LFC is the level of free convection (LFC); T_vparcel is the virtual temperature of the air parcel; and T_venv represents the virtual temperature of the environment.
The skill of MPAS in downscaling temperature and precipitation extremes is assessed using the extreme indices recommended by the Expert Team on Climate Change Detection and Indices (ETCCDI) (Sillmann et al., 2013).For the rainfall, R95pTOT (Precipitation fraction due to very wet days (> 95th percentile)), R99pTOT (Precipitation fraction due to very wet days (> 99th percentile)), RX5day (Maximum 5-day precipitation), SDII (Average precipitation during Wet Days), CDD (Maximum consecutive dry days), and CWD (Maximum consecutive wet days) are included.For the near-surface air temperature, heat wave total length, heat wave max length, daily temperature range (DTR, difference between daily maximum and minimum temperature), daily temperature range variability (vDTR, the absolute day-to-day differences), CSDI (Cold-spell duration index), and WSDI (Warm-spell duration index).A heat wave event is characterized by a sequence of at least three consecutive days where the daily maximum temperature T vparcel − T venv T venv exceeds 35°C, as per the de nition by Wang et al. (Wang et al., 2019).The total duration of a heat wave is presented in terms of the number of days it persists.Six sub-regions shown in Fig. 1

Seasonal cycle
The 31-year averaged seasonal variations of regional averaged daily precipitation in six sub-regions are shown in Fig. 4. MPAS-A can reasonably capture the seasonal cycle of precipitation over sub-regions with the correlation coe cients all above 0.8, it exhibits wet biases in all sub-regions from January to April and has the largest RMSE in summer.In the monsoon-dominated regions including SC, YHR, JP, and KO, there is a prevalent tendency to underestimate the peak of the monsoon rainfall after May.In SC, MPAS-A fails to present su cient intensity for the rainy season from May to June, which is called the pre-ood season.It also underestimates the Meiyu season, which starts in mid-June, for YHR, KO, and JP, with the largest dry biases observed in June and July and the lowest TCCs presented.In YHR, this dry bias is the most prominent, which is maintained till December.Conversely, MPAS-A exhibits the best performance for the further north regions including NC and NE, with slight wet biases prevalent throughout the year.
The seasonal cycle of regional averaged T2m in all sub-regions is depicted in Fig. 5. MPAS-A can reproduce the seasonal variation of regional averaged T2m with high temporal correlations and low RMSEs.The observed T2m reaches the highest temperature during July and August, with signi cant simulated differences between coastal and inland areas.In the three coastal regions, SC, KO, and JP, MPAS-A exhibits cold biases which are alleviated during summer when simulated dry biases exist.These cold biases are more pronounced in autumn with the maximum bias reaching − 2℃ in SC.Conversely, MPAS-A shows annually averaged warm biases in the inland areas including NE and NC, with YHR and NC exhibiting summer drought and the highest warm biases exceeding 1℃ in June.
Figure 6 provides a detailed depiction of the spatiotemporal evolution of the rain belt over East Asia, which is presented by the time-latitude cross-section of the zonal mean (105°E ~ 125°E) daily precipitation amount from the combined observations and MPAS-A over the period 1997-2018.From early April to mid-June, a pre-ood season characterized by an enhanced rain belt with precipitation above 6mm/day centered around 25°N is observed in South China.Subsequently, the rain belt intensi es and gradually moves northward from mid-June to early July, coinciding with the Meiyu season in the Yangtze-huaihe River basin.After mid-July, two rain belts are observed in South and Northern China, respectively.Concerning the pre-ood season rain belt, MPAS-A can simulate a similar intensity and evolves about 15 days faster than the observations.However, it underestimates the intensity during the Meiyu season, with precipitation less than 10mm/day and a 15-day phase lag compared to the observations.In the northern regions, the simulated rain belt reaches a similar maximum latitude compared to the observations but presents a longer stationary period, leading to a slight wet bias in NC and NE.In contrast, MPAS-A underestimates precipitation in the Yangtze-huaihe River basin during the same period, leading to summer drought.After August, MPAS-A simulates a more rapid retreat of the rain belt than the observations.Additionally, the observed rainfall in South China is enhanced until mid-October, but the MPAS-A simulation maintains a stronger rain belt, with a maximum precipitation amount exceeding 18 mm/day until November.In summary, MPAS-A shows a bipolar bias distribution, with more precipitation concentrated in South and North China but less in the Yangtze River basin, due to its faster northward shift of the rain belt.

Frequency distribution
Figure 7 presents an analysis of the frequency-intensity distribution of daily precipitation in all subregions during 1997-2018.Sub-gures a-f contain the precipitation frequency distribution below 10mm/d, while sub-gures g-i contain parts above 10mm/d, including the distribution of extreme precipitation exceeding 50mm/d.For monsoon-in uenced sub-regions such as SC, KO, and JP, MPAS-A shifts its frequency distribution towards slight rainfall (0.1-2.0mm/day) and consistently underestimates the frequency of rainfall above 10mm/day.Although MPAS-A exhibits dry biases in YHR during summer and autumn, it reasonably captures the annual daily precipitation frequency, except for the slightly underestimated frequency of extreme precipitation exceeding 50mm/d.In contrast, for the northern subregions such as NE and NC, MPAS-A concentrates its frequency distribution on the moderate and extreme rainfall interval (10-100mm/day), leading to the lower frequency of slight rainfall (0.1-2.0mm/day).
Figure 8 presents the frequency distribution of daily T2m in all sub-regions during the period 1988-2018.MPAS-A has a relatively better performance in producing the distribution of T2m than that of precipitation.However, the frequency distribution of MPAS-A exhibits offsets, which are directly associated with the previous overall biases of T2m.MPAS-A reveals warm biases in the NE and NC, with a notable shift towards warmer T2m in the frequency distribution.It presents overall offsets in NC and overestimated frequency in the high tail in NE.In YHR, the simulated T2m also exhibits a warm bias, but the pattern shift towards higher T2m is primarily concentrated near the vicinity of the bimodal peaks.In contrast, the simulated T2m in SC, KO, and JP is underestimated, with the simulated frequency distribution leaning towards the colder side.MPAS-A displays a atter pattern with a lower peak in KO and JP and it fails to capture the bimodal pattern observed in KO.Despite the large cold biases in the SC region, MPAS-A adequately captures the unimodal pattern observed in SC with the offset toward lower T2m.

Extreme climate indices
Extreme climate events have severe impacts on agriculture, ecosystems, and society, so it is essential to evaluate the performance of MPAS-A in producing extreme climate events.Based on the de nition of extreme indices from ETCCTI, the extreme precipitation indices are calculated based on daily precipitation from 1997 to 2018, comprising R95pTOT, R99pTOT, RX5day, SDII, CDD, and CWD for both observation and simulation and the spatial distributions are presented in Fig. 9. MPAS-A generally reproduces the spatial characteristics of R95pTOT and R99pTOT (Figs. 9a-d), exhibiting only a minor northward deviation of approximately 5 degrees in the northwest Paci c Ocean.However, the simulated intensities for these two indices are overestimated in the arid regions of Northwest China and Central Asia.The overestimated rainfall in MPAS-A over tropical oceans contributes to the overestimation of RX5day and SDII (Figs. 9e-h), but this deviation is less pronounced over land, despite slightly overestimated RX5day and SDII over North China.The dry biases over the Yangtze-huaihe River basin have minimal impact on these climate indices tied to daily precipitation amounts.MPAS-A reproduces the realistic CDD distribution in arid regions but overestimates in Central Asia.In the Yangtze-huaihe River basin, simulated CDD is overestimated by about 15 days coinciding with the long summer drought there, suggesting the impact of the large-scale deviation.In contrast, the simulated CWD distribution in tropical oceans exceeds observations by more than 30 days, indicating MPAS-A's inclination to trigger convection in this region, which leads to overestimated wet biases.Additionally, in southwest China and the Tibetan Plateau, the simulated distribution of CWD is more strongly tied to the terrain compared with the observation.For instance, MPAS-A is more likely to trigger precipitation on the southern side of the Tibetan Plateau.
Overall, despite the biases in daily precipitation due to defects such as rain belt evolution, MPAS-A reasonably reproduces the extreme indices on land and shows robustness for certain indices tied to daily precipitation amount and extreme rainfall.
The extreme climate indices of the near-surface air temperature, including heat wave total length, heat wave max length, daily temperature range, daily temperature range variability, WSDI, and CSDI, are depicted in Fig. 10.MPAS-A captures the spatial distribution of the heat wave total length (Figs.10a-b) and heat wave max length (Figs.10c-d) on land but overestimates the total length by 10 days and 20 days in the eastern part of China and the northwest region of China, respectively.The simulated longer summer drought coincidences with higher heat wave total length.For the heat wave max length, MPAS-A well simulates the distribution patterns despite insu cient intensity over the Indian Peninsula.The simulated daily temperature range (Figs.10e-f) and its variability (Figs.10g-h

Possible reasons and mechanisms for the simulated results
In this section, possible reasons and mechanisms for the simulated near-surface temperature and precipitation biases are analyzed from the aspect of the large-scale circulation features, vertically integrated moisture ux, and relationship of the precipitation amount and CAPE values.

large-scale circulation features
Figure 11 shows the 27-year averaged spatial distributions of the seasonal mean geopotential height, air temperature, and winds at 500hPa from the ERA-Interim reanalysis and MPAS-A, and the biases between them.MPAS-A presents the reasonable climatology distribution of the circulation features compared with the ERA-Interim reanalysis, but nonnegligible deviations are presented in both summer and winter.The biases of simulated circulation in winter and spring are quite similar and exhibit dipolar bias patterns.
The simulated cyclonic deviations appear between 25°N and 40°N accompanied by the underestimated temperature at 500hPa, while the anticyclonic biases are presented north of 40°N with the overestimated temperature.These dipolar bias distributions are more pronounced during winter and the cyclonic deviations also correspond to the signi cant cold biases for simulated T2m on the Tibetan Plateau.The anticyclonic deviations, southeast wind biases, and warmer air in the Siberian plain indicate weaker activity of cold air transport for high-latitude regions, which induces warmer T2m around North China during winter.In summer, MPAS-A exhibits "negative-positive-negative" bias patterns across latitudes for the geopotential height and air temperature.Thus, it demonstrates a weaker but more northward Western Paci c Subtropical High (WPSH), which is crucial for the East Asia summer monsoon.Compared with the observation, this deviation induces the weaker northward evolution of the rain belt along the eastern coast of China suggesting longer summer drought and ampli ed heat wave total length around the Yangtze-huaihe River basin.During autumn, MPAS-A also exhibits a slower retreat of the WPSH compared with the ERA-Interim reanalysis associated with persistent drought in South China

vertically integrated moisture ux
Figure 12 shows the spatial patterns of the seasonal vertically integrated moisture ux (Q ux) from the reanalysis and the MPAS simulations.And the streamlines of simulated biases are also depicted to better present its distribution patterns.In winter, the Q ux distribution of ERA-Interim reanalysis exhibits a steady transport of moisture vapor by the tropical easterlies and mid-latitude westerlies.The observed moisture transports are well simulated in winter with the highest correlation and lowest Bias and RMSE except for the lower simulated Q ux intensity over the Northwest Paci c.The observed Q ux gradually strengthens in spring, forming two distinct moisture transport channels in monsoon regions over East Asia in summer.One channel is the southwesterly Q ux originating from the tropical Indian Ocean and passing through the Bay of Bengal and the Indo-China Peninsula.The other channel emerges as an anticyclonic air ow from the north side of the WPSH, extending towards the South China Sea.In spring, MPAS-A exhibits a similar distribution of Q ux biases.But MPAS-A overestimates QFlux exceeding 100 kg•m-1•s-1 in the Bay of Bengal contributing to the overall overestimated Q ux bias.MPAS-A also exhibits a larger eastward transport of Q ux in the tropics.In summer and autumn, compared to spring, MPAS-A exhibits larger positive biases with the maximum overestimation of Q ux occurring over the tropical ocean, which corresponds to the stronger wet biases.It consistently exhibits cyclonic Q ux bias patterns over the northwest Paci c, which is associated with a weaker and northward-shifted WPSH, especially for summer.These cyclonic bias patterns enhance the zonal component of Q ux over its southern side, resulting in more water vapor transport toward the South China Sea and less toward monsoon regions.
On the northern side of the cyclonic bias patterns, more water is transported northward toward North China and Northeast China.Thus, the weaker simulated Q ux around the east side of the cyclonic biases induces a summer drought from the middle and lower reaches of the Yangtze River to the southern coast of Japan.Additionally, in autumn, the cyclonic bias strengthens, and MPAS-A underestimates the Q ux over South China with the lowest SCC for its two components.

relationship of the precipitation and CAPE
Figure 13 depicts the frequency distributions of the relationship between daily mean CAPE and precipitation on each grid during 1997-2016, including the observed and simulated results on the continent and ocean.The frequency in the logarithmic scale is averaged into evenly spaced bins of CAPE and daily precipitation.For a bin of speci c CAPE, lines presenting multiple quantiles (0.25, 0.5, 0.75, 0.95, 0.99) for the corresponding precipitation frequency are also plotted with con dence intervals (shaded) with a con dence level of 0.05.In the observation, the frequency distribution of CAPE-Precipitation presents an unimodal structure.For the multiple quantiles lines, the precipitation is positively correlated with low CAPE (CAPE < 500 KJ/kg), but after the peak, precipitation gradually reaches its maximum and presents a slower increment.The observed frequency shows narrower peaks on the continent compared with the ocean.MPAS-A exhibits systematically higher CAPE values and bimodal structure of the frequency, which leads to more widespread distribution patterns and frequency peaks offsetting the higher CAPE values compared with the observation.This overestimation suggests that MPAS-A requires a higher CAPE to trigger convection.For the quantile's lines, MPAS-A presents comparable precipitation before the peaks, but underestimation exists for the rainfall with higher CAPE values especially for the precipitation with frequency quantiles of 0.95 and 0.99.For the extreme CAPE interval (CAPE > 2000 KJ/kg), The simulated results reach a lower maximum precipitation and present suppression of the extreme rainfall.This underestimation indicates that the conversion of CAPE into vertical velocities of convective air parcels is less e cient in MPAS-A for extremely high CAPE.The underestimation of the simulated frequency of extreme rainfall in SC, YHR, KO, and JP may be associated with this de ciency.

Conclusions
In this study, we performed an MPAS-A experiment using a variable resolution mesh with a spatial resolution of approximately 25km in the re ned region, allowing for interactions between multiscale features.To facilitate comparison with existing RCMs, the SSTs and sea ice used in the experiment are constrained by the ERA-Interim reanalysis for the period from 1988 to 2016.We validated the model results against combined observed datasets and conducted evaluations of rainfall and near-surface air temperature.The evaluations encompass various aspects, including mean climatology, seasonal cycle, frequency distribution, and extreme indices.
Although MPAS-A reasonably captures the climatology of rainfall and climate extreme indices on land, it fails to accurately depict the temporal evolution of precipitation patterns.Instead, it demonstrates a faster advancement of the rain band compared to combined observed data, resulting in inadequate monsoon precipitation and intensi ed rainfall further north.The consistently underestimated summer rainfall and frequency in SC, YHR, KO, and JP re ect prolonged summer drought there, while overestimated rainfall and frequency in NC and NE suggest further northern rain belts.Notable discrepancies arise in the overestimation of precipitation over the tropical ocean and signi cantly affect the simulated distributions of RX5day and SDII, which are closely tied to the daily rainfall amount.However, the large-scale wet biases have a minor impact on simulated R95pTOT and R99pTOT, which demonstrates the robustness of capturing extreme rainfall with an improved spatial resolution.Furthermore, the biases presented in the simulated CWD exceeding 50 days suggest that MPAS-A tends to trigger more convection in the tropical ocean.
The simulated results for near-surface air temperature demonstrate a high level of reliability, with predominantly stationary deviations observed across most regions.Speci cally, there is a warm bias in the inland areas of China, while most coastal regions exhibit slight cold biases, despite the seasonal transitions of biases in the Tibetan Plateau.These biases are also re ected in the region-averaged seasonal cycles with signi cant differences in simulation biases observed between the south and north.The simulated frequency distribution presents a systematic shift in the overall distribution, indicating homogeneous biases for most grid points across the whole sub-region.Accurate temperature simulation enhances the performance of extreme indices, while some biases are also evident.For the heat wave indices and WSDI, an overestimation in the middle and lower reaches of the Yangtze River is simulated coinciding with summer drought.MPAS-A presents smoother variability such as underestimated daily temperature range and its mean absolute day-to-day variation, especially in North China associated with stationary warm biases.Generally, compared to reanalysis-based RCMs, the MPAS-A simulation exhibits more prominent systematic biases in the large-scale circulation due to the absence of observational constraints, resulting in de ciencies in simulating the monsoon system and erroneous phase of rainbands and impacting the rainfall climatology and seasonal cycle.However, MPAS-A still demonstrates its robustness, as it captures reasonable near-surface temperature distributions despite stationary bias patterns.The simulated rainfall on land also demonstrates lower sensitivity to the large-scale circulation's biases in re ned regions for the extreme indices, especially for the R95pTOT and R99pTOT.This highlights the robustness of the VR approach in obtaining regional information within the same framework of the single GCM model.
) show an overall underestimation north of 25°N, especially for the simulated variability which misses the high-value band across North and Northeast China.The simulated daily temperature range over the Tibetan plateau and Northwest China shows strong topography signals similar to the warm bias distributions.MPAS-A demonstrates a commendable performance in capturing the WSDI (Figs.10i-j) characteristics over most land areas in East Asia.MPAS-A clearly underestimates WSDI over southern India and the Indo-China Peninsula, while an evident overestimation of WSDI is observed over northwestern India and Pakistani.Although MPAS-A can simulate the distribution of CSDI (Figs.10k-l) over coastal regions in China, India, and the Indo-China Peninsula, it clearly overestimates CSDI in central China and over the Tibetan Plateau with the overestimation at about 4-5 days, which coincides with notable cold biases that are closely linked to high topography during winter.The warm deviation of T2m in North China appears to have minimal impact on the climate indices of extremely high temperatures, such as WSDI and heat wave index.Despite an overall overestimation of temperature frequency and similar warming deviations observed in the NC and NE regions, MPAS-A overestimates the extremely low-temperature index (CSDI).

DeclarationsAcknowledgments:
This work is jointly funded by the National Natural Science Foundation of China (U2242204, 42130602), and the Jiangsu Collaborative Innovation Center for Climate Change.The numerical calculations in this paper have been done on the computing facilities in the High Performance Computing Center (HPCC) of Nanjing University.The authors acknowledge with thanks to the ECMWF for providing the ERA-interim reanalysis data as driving elds in the simulations, NOAA for providing the GPCP data, and NOAA's Climate Precipitation Center for providing the CMORPH observational data.We also thank Dr. Jia Wu from the National Climate Center (NCC) of China Meteorological Administration

Figures
Figures

Figure 1 The
Figure 1

Figure 4 Monthly
Figure 4

Figure 5 Monthly
Figure 5

Figure 7 Frequency
Figure 7

Figure 13 the
Figure 13

Table 1
referred to as APHRO, as described by Yatagai et al. (Yatagai et al., 2012), covers the Asian monsoon region and provides a spatial resolution of 0.25 degrees for land areas.The daily precipitation of APHRO-V1901 covering 1998-2015 and the near-surface air temperature of APHRO-V1808 covering 1990-2015 are utilized.
The capability of the MPAS-A model in producing the spatial distribution of the climatology is evaluated against the combined reference precipitation datasets during 1988-2018.For the daily precipitation, the spatial distribution in East Asia is dominated by the East Asian Summer Monsoon (EASM) and the Westerlies.Figure2depicts the spatial distribution of the 27-year averaged seasonal mean precipitation from the observation, simulation, and the biases between MPAS-A and observation.The observed distribution exhibits a southeast-northeast decrease gradient of the precipitation amount on land in every season, which is more prominent in summer.The rainband extends from the South China Sea to Japan, In summer, signi cant dry biases appear in the south side of the Tibetan Plateau and the middle and lower reaches of the Yangtze River, which extends to the East China Sea and the southern side of Japan.These dry biases suggest a rainband located further north during the wet season, which also leads to excessive rainfall in North China.Over the tropical ocean, MPAS-A simulates large wet biases exceeding 3.5mm/d.These biases rst appear in the spring over the Bay of Bengal and the South China Sea and reach the maximum during the summer over the western Paci c and the Indian Ocean, with a simulated zonal rain belt exceeding 16mm/d.terrain-induced signals over Northwest China.MPAS-A has warm biases in central and northwestern China, with pronounced biases exceeding 1.5℃ during both summer and winter seasons.Over Southern Asia and Southeastern Asia, such as the Indochinese and Indian Peninsulas, cold biases exist.In winter and spring, MPAS-A shows the most signi cant cold biases below − 5°C over the Tibetan Plateau, resulting in the largest RMSE of 3.3℃.However, those simulated biases turn warm during the summer and autumn.
passing through southern China and the Korean Peninsula.In the Northwest Paci c, the observed precipitation is characterized by two distinct rain belts.In the Bay of Bengal and the South China Sea, observed precipitation exhibits more pronounced seasonal variations, reaching maximum values exceeding 16mm/day during the summer and autumn seasons.MPAS-A reasonably captures the gradient of precipitation amount with the spatial correlations coe cient (SCC) larger than 0.84 with simulated biases ranging from − 2.5mm/d to 2.5mm/d.In arid regions such as Northwest China and Central Asia, MPAS-A consistently exhibits a dry bias of approximately − 0.5mm/day throughout all seasons.In the monsoon-in uenced regions during winter and spring, MPAS-A generally exhibits wet biases except for dry biases in East China.Those wet biases are particularly pronounced in the downstream areas of the Plateau in Southwest China with the maximum exceeding 2.5mm/day.Figure3shows the spatial distribution of the 31-year averaged seasonal mean near-surface air temperature (T2m) from observation and simulation, and the simulated biases.The observed T2m generally decreases from south to north, with the highest values over India and Indo-China Peninsula, and southeastern coastal areas.The coldest T2m is observed in the northern regions, such as Siberia, Mongolia, and northern China.The simulated results demonstrate agreement with the observed distributions, with SCC surpassing 0.96.MPAS-A predominantly displays stationary spatial distributions with signi cant (Tang et al., 2018)13) a widespread overestimation in the northwest region of China and the Tibetan Plateau tied to severely cold biases during winter.contrasttotheobservedunimodaldistribution.This wider simulated distribution is more prominent over the ocean, with most precipitation triggered around a higher peak of CAPE leading to an excessive rainfall there.However, at extremely high CAPE values, MPAS-A suppresses precipitation, suggesting an inadequate conversion e ciency of CAPE to vertical momentum in the ascending air masses, resulting in a de ciency of extreme precipitation.Some studies of GCMs also indicate that this oceanic wet bias and missed rainfall in South China contribute to erroneous moisture transport which causes excessive water vapor accumulation over the South China Sea(Sperber et al., 2013), uncertainties on the simulated large-scale circulations or differentclass precipitation (D.-Q.Huang et al., 2013).In this long-term simulation, the simulated biases are primarily dominated by deviated large-scale circulation.Introducing more realistic driving models through spectral nudging(Tang et al., 2018)or nesting methods can signi cantly improve these large-scale features' biases but cannot eliminate them.Similar large-scale biases are captured in other RCMs' studies, such as excessive moisture transport in the lower layer (Niu et al., 2020; Tang et al., 2022), and stronger but more northward WPSH (G.Kim et al., 2021; X.-Z.Liang et al., 2019, p. 2).The convective parameterization schemes in RCMs are found to trigger similar excessive oceanic rainfall (Niu et al., 2020; Yu et al., 2019; Zou et al., 2014).Additionally, the convective parameterization used in MPAS-A, in the Siberian plain suggest reduced cold air transport in high-latitude regions, inducing warmer temperatures around North China.The simulated Q ux over the tropical ocean exhibits a more zonal orientation compared to the reanalysis, leading to a surplus of moisture over the tropical ocean and a de ciency of moisture in the Yangtze-huaihe River Basin.The analysis of CAPE-precipitation frequency distributions reveals that MPAS-A systematically overestimates CAPE and leads to a widespread bimodal distribution, in whether it is the Grell-Freitas or the Tiedtke scheme, tends to produce more precipitation over tropical oceans, which may be related to the interaction between convective parameterization schemes and other physical parameterization schemes (Fowler et al., 2016).It may be necessary to apply some convection suppression criteria(Zou et al., 2014)to alleviate these wet biases.