Assessing the Regional Climate Response to Different Hengduan Mountains Geometries With a High‐Resolution Regional Climate Model

The Hengduan Mountains (HM) are located on the southeastern edge of the Tibetan Plateau and feature high mountain ridges (>6,000 m MSL) separated by deep valleys. The HM region also features an exceptionally high biodiversity, believed to have emerged from the topography interacting with the climate. To investigate the role of the HM topography on regional climate, we conduct simulations with the regional climate model COSMO at high horizontal resolutions (at ∼12 km and a convection‐permitting scale of ∼4.4 km) for the period 2001–2005. We conduct one control simulation with modern topography and two idealized experiments with modified topography, inspired by past geological processes that shaped the mountain range. In the first experiment, we reduce the HM's elevation by applying a spatially non‐uniform scaling to the topography. The results show that, following the uplift of the HM, the local rainy season precipitation increases by ∼25%. Precipitation in Indochina and the Bay of Bengal (BoB) also intensifies. Additionally, the cyclonic circulation in the BoB extends eastward, indicating an intensification of the East Asian summer monsoon. In the second experiment, we remove deep valleys by applying an envelope topography to quantify the effects of terrain undulation with high amplitude and frequency on climate. On the western flanks of the HM, precipitation slightly increases, while the remaining fraction of the mountain range experiences ∼20% less precipitation. Simulations suggest an overall positive feedback between precipitation, erosion, and valley deepening for this region, which could have influenced the diversification of local organisms.


Introduction
The Hengduan Mountains (HM) are located on the southeastern edge of the Tibetan Plateau (TP).Covering an area of over 600,000 km 2 and featuring an average elevation of more than 4,000 m MSL, the HM represents the longest and widest north-south mountain range system in China (Z.Li et al., 2011;Ning et al., 2012;K. Zhang et al., 2014).The contemporary topography is shaped by plate tectonics, which has led to the formation of folded mountains and a series of faulted basins, as well as by spatially heterogeneous erosion, responsible for the creation of deep river valleys.These valleys possess high topographic complexity and exhibit active geomorphic processes at the kilometer scale (Clark et al., 2005; L. Ding et al., 2022;Royden et al., 2008;Tian et al., 2015;E. Wang et al., 2012;Yang et al., 2016).Despite being located at higher latitudes, the HM hosts exceptionally high biodiversity, comparable to tropical regions (Mutke & Barthlott, 2005).This feature is believed to be linked to past complex interactions between plate tectonics, land surface dynamics, and atmospheric circulation in this region (Antonelli et al., 2018).Understanding the complex interaction between topography and climate is key to comprehending the features that make this region climatically and biologically unique.
Situated at the convergence of the Indian, East Asian, and western North Pacific summer monsoon systems (ISM, EASM, and WNPSM), the climate of HM exhibits a typical monsoon dynamic with distinct rainy and dry seasons (B.Wang & LinHo, 2002).The rainy season, which spans from May to September, sees the South Asian monsoon strike the mountain range, bringing substantial moisture and resulting in high rates of precipitation, particularly in the southwestern part of the HM (Z.Zhang et al., 2004).The influence of the north-south orientation of the HM is evident in the heterogeneous spatial distribution of local precipitation-the southwestern part of the HM receives relatively high precipitation, while the central and northeastern parts experience relatively low precipitation (Yu et al., 2018).Moreover, the complex topography with a profoundly dissected landscape generates a heterogeneous distribution of precipitation with a contrast between moist and dry valleys.Both the mean precipitation and precipitation extremes decrease from southwest to northeast across the HM (Z.Li et al., 2011;Ning et al., 2012;K. Zhang et al., 2014).Precipitation over the HM plays a significant role in shaping local ecological productivity through its impacts on glacier growth, surface runoff, and river flow (Dong et al., 2016;Qi et al., 2022).
The topography of the TP and the HM are known to significantly influence the Asian monsoon through both dynamic and thermal effects.The topography acts as a barrier, preventing the intrusion of cold, dry extratropical air into the warm, moist regions affected by the Asian monsoon (Boos & Kuang, 2010).Additionally, the landmass releases energy into the atmosphere in summer, inducing air pumping, deflecting mid-latitude westerlies, and generating cyclonic circulation in the lower troposphere in the Bay of Bengal (BoB) (Wu et al., 2012).However, the relative importance of these effects-i.e., the blocking versus air pumping-for monsoon formation remains a matter of debate (Acosta & Huber, 2020;G.-S. Chen et al., 2014;Molnar et al., 2010;Park et al., 2012;Xu et al., 2019).Both data diagnosis and numerical experiments have exhibited that the topography affects the downstream EASM through mid-latitude Rossby wave propagation and air-sea interaction (KOSEKI et al., 2008;Duan et al., 2011;Y. Liu et al., 2020;M. Lu et al., 2023;Y. Zhang et al., 2004;Zhao & Chen, 2001).B. Wang et al. (2008) argued that the warming TP enhances summer frontal rainfall in the East Asia region by strengthening the anticyclonic circulation at upper levels and the cyclonic circulation at lower levels.This facilitates the eastward propagation of Rossby wave energy and fortifies the anticyclonic ridge over eastern China, strengthening moisture transport toward the East Asia subtropical front.According to Wu et al. (2017), under global warming, the sensible heat of the TP experienced a reduction from the mid-1970s to the end of the 20th century due to decreased surface wind speed.This reduction has resulted in a weakened near-surface cyclonic circulation and, consequently, a weakened EASM.Hence, the rain belt remains situated over South China, intensifying the precipitation in the region.The discrepancy between the findings of these studies may be ascribed to the different sources and uncertainties in data quality.A more reliable modeling study is required to tackle the physical processes by which the status of the TP affects the regional climate.
Numerical simulations have been widely employed to investigate the impact of mountain uplift on local and largescale climate in interaction with the Asian monsoon system.Early studies focusing on the surface uplift effects of the TP treated the region as a single, vast feature, using low-resolution climate models with just two scenarios: with and without mountains (Manabe & Terpstra, 1974).Subsequent research used 'phased uplift' scenarios, assuming a linear increase in elevation based on the premise that past TP states can be approximated by spatially homogeneous scaling of contemporary topography (Botsyun et al., 2016;D. Jiang et al., 2008;X. Liu & Yin, 2002;Paeth et al., 2019).However, geological evidence suggests that the TP has experienced regional uplift, rather than a uniform rising process (Tapponnier et al., 2001).More realistic regional uplift scenarios are now being considered, and the role of the HM is being examined.H. Tang et al. (2013) found that the EASM enhancement is primarily driven by the surface sensible heating of the central and northern TP and HM.R. Zhang et al. (2015) underscored the role of the HM in modifying the low-level cyclonic circulation in the BoB, leading to substantial precipitation in this area.Yu et al. (2018) proposed that the uplift of the HM primarily causes local, rather than large-scale, changes.The topography in the HM is characterized by both the high average elevation and its local variance and both should be evaluated to understand the complex climate of the region.
The complex topography of the TP and HM regions poses a significant challenge to accurately modeling monsoon-influenced mountain climate.Yet, many previous terrain modification studies have relied on coarseresolution global climate models, typically featuring a grid spacing of 100-200 km (Botsyun et al., 2016;D. Jiang et al., 2008;X. Liu & Yin, 2002;H. Tang et al., 2013;R. Zhang et al., 2015) or intermediate-resolution regional climate models with a grid spacing of 20-50 km (Paeth et al., 2019;Yu et al., 2018), which are unable to capture small-scale topography and its influence on the HM climate.Previous studies have demonstrated that high-resolution convection-permitting model (CPM) can offer a more accurate representation of climate, particularly in terms of capturing extreme events such as heavy precipitation and the water cycle in areas of complex terrain, compared to global climate simulations (Ban et al., 2015;Giorgi & Mearns, 1999;Kotlarski et al., 2014;Prein et al., 2016;Schiemann et al., 2014).Over the TP and the HM, Lin et al. (2018) found that CPM more effectively resolves orographic drag in complex terrains, enhancing the representation of water vapor transport and precipitation.Gao et al. (2020) emphasized the added value of CPM in accurately simulating the spatial distribution of precipitation and downstream snow simulation.P. Li et al. (2021) demonstrated that CPM more accurately depicts both the frequency and intensity of summer precipitation.Ma et al. (2022) observed that CPM yields better simulations of precipitation and temperature over the TP, notably improving the representation of the location and intensity of the heavy Meiyu precipitation.Z. Liu et al. (2022) showed that CPM more effectively captures the diurnal cycle of precipitation over the northern and eastern TP.However, due to high computational costs, previous studies utilizing CPMs over the TP and the HM have been constrained by short simulation periods or limited to small simulation domains.
In this study, we evaluate the impact of the HM geometry on both regional and local climates, with a focus on extreme precipitation events.We employ the regional climate model COSMO (Rockel et al., 2008) to conduct numerical experiments for the period from 2001 to 2005 with both contemporary and two idealized topographies that are linked to the formation of the HM.In the first experiment, we produce a topography with a lower average elevation in a spatially non-uniform way, which reflects a past potential state of the HM.In a second experiment, we eliminate deep valleys, formed by uplift and river incision, by applying an envelope topography to quantify their impact on climate.This experiment with smaller-scale terrain modifications focuses more on local-scale terrain influences on the atmosphere.For each topographic scenario, two distinct simulations are performed.The first simulation is performed with a grid spacing of 12 km.Subsequently, a higher-resolution simulation with a convection-permitting grid spacing of 4.4 km is nested within the first simulation.
The structure of the manuscript is as follows: Section 2 introduces the climate model used in this study and its configuration, the derivation of the idealized topographies, and the reference data employed in this study.Sect. 3 presents an evaluation of COSMO's capability to reproduce the control climate.Section 4 discusses the experiments with modified topography.Section 5 provides a summary of the main findings of this study and concluding remarks.

Model Simulations
In this study, we apply the non-hydrostatic COSMO model (Rockel et al., 2008) in climate mode within a twostep, one-way nesting framework.The COSMO version used here takes advantage of a heterogeneous hardware architecture with Graphics Processing Units (GPUs), enabling more efficient exploitation of available hardware, and energy resources, and achieving higher computational performance (Fuhrer et al., 2014;Leutwyler et al., 2016).The model uses the generalized terrain-following height coordinate (Gal-Chen & Somerville, 1975) with rotated latitude-longitude coordinates and applies a split-explicit third-order Runge-Kutta scheme in time (Wicker & Skamarock, 2002).For convective parameterization, COSMO employs the Tiedtke Mass flux scheme with equilibrium closure based on moisture convergence (Tiedtke, 1989).The multi-layer soil model TER-RA_ML, coupled with the groundwater-runoff scheme described by Schlemmer et al. (2018), is used for the representation of land surface processes (Heise et al., 2006).The radiation parameterization scheme is based on a δ-two-stream version of the general equation for radiative transfer (Ritter & Geleyn, 1992).A turbulent-kineticenergy-based parameterization is used for vertical turbulent diffusion and surface fluxes (Raschendorfer, 2001).Cloud microphysics is represented by a single-moment scheme that considers five species: cloud water, cloud ice, rain, snow, and graupel (Reinhardt & Seifert, 2006).
We use COSMO in the following framework: We define a large-scale model (LSM) domain (Figure 1a) with a grid spacing of 0.11°(∼12 km) and 1,058 × 610 grid cells.This domain approximately corresponds to the CORDEX East Asia domain (Giorgi & Gutowski, 2015) but extends eastward to allow an unconstrained imprint of the modified topography on the large-scale climate downstream of the typical westerly flow.We perform LSM simulations with parameterized deep convection.Within the LSM domain, we nest a CPM with a grid spacing of 0.04°(∼4.4km) and 650 × 650 grid cells.The CPM domain, centered over the HM, covers Southwest China and parts of Indochina (Figure 1b).The CPM simulations explicitly resolve deep convection and are initialized from the LSM experiments.In the vertical direction, all simulations are run with 57 model levels ranging from the surface to the model top at approximately 30 km.We use a sponge layer with Rayleigh damping in the uppermost levels of the model domain.All simulations (control and two experiments with modified topography; see Section 2.2) span a 5-year period from 2001 to 2005.LSM simulations are initialized and laterally driven by the European Center for Medium-Range Weather Forecast (ECMWF) operational reanalysis ERA5 (Hersbach et al., 2020) at 6-hourly increments.Previous regional climate model experiments have shown that model performance can be improved with the application of spectral nudging (von Storch et al., 2000;Cha & Lee, 2009)also for the East Asian region (D.-K. Lee & Cha, 2020;D. Lee et al., 2016;J. Tang et al., 2017).In this setup, forcings are stipulated not only at the lateral boundaries but also in large-scale flow conditions inside the model integration domain.However, we opt not to apply spectral nudging because modified topography is expected to impact climate on both local and larger scales.Spectral nudging would adjust large-scale atmospheric flow at upper levels toward the reanalysis state, which is derived from unmodified modern topography.To avoid this inconsistency and to allow for more unconstrained imprints of modified topography on large-scale flow, this technique is not used.

Modification of the Hengduan Mountains' Topography
The modern control topography (Figure 2a), as well as the two modified topographies, are derived from the highresolution digital elevation model (DEM) MERIT (Yamazaki et al., 2017).This DEM demonstrates very good performance in terms of data quality and general statistics compared to similar available DEM products for the High-Mountain Asia (HMA) region (K.Liu et al., 2019).For consistency, we apply the topographic changes to both the coarse-(0.11°/∼12km) and high-resolution (0.04°/∼4.4 km) model topography.We refer to the coarse and high-resolution control simulations as CTRL11 and CTRL04, respectively.Before running COSMO simulations, we use COSMO's pre-processing tool EXTPAR (Asensio et al., 2021) to generate static external fields such as surface elevation, land-sea mask, and background albedo.Some of these fields, such as the orographic sub-grid parameters, depend on the raw input topography.To ensure consistency among all topography-based fields, we modify the MERIT data fed into EXTPAR, rather than altering the output topography from EXTPAR.

Reduced Topography
To study the impact of regional surface uplift, we generate a topography representing a possible past stage of the HM with a lower average surface elevation.Detailed regional information on the past stages of the geological evolution of the Southeastern TP is uncertain (Royden et al., 2008).This hypothetical stage is inspired by the topographic configuration before the onset of the eastward extension in the central TP (Hoke et al., 2014).In this scenario, topographic changes are confined to the Southeastern TP and part of the Indochina Peninsula (Figure 2b).The east-west extension of the TP is represented in the model by a geographically-based modification of the modern HM topography, with elevation reductions ranging from 0% to 90%.A more detailed description of the topography modification scheme is presented in Supporting Information S1.We refer to the coarse-resolution simulation with reduced topography as TRED11 and the high-resolution simulation as TRED04.

Envelope Topography
In this topography modification experiment, we investigate the role of deep valleys, which have formed through river incision and erosion, on the local climate.To remove river incisions from the modern topography, we compute an envelope topography.This concept has been applied in other studies (Li & Zhu, 1990;Damseaux et al., 2020), though driven by different research questions.We derive an envelope topography by computing a three-dimensional convex hull from the MERIT DEM, whose curvature is enhanced by a certain factor.The triangle mesh from the convex hull is subsequently rasterized back to the regular MERIT grid.This raw envelope topography is then embedded into the unmodified MERIT data with a 100 km wide transition zone to ensure smooth and continuous terrain between the raw envelope and the unmodified topography (see Figure S4c in Supporting Information S1).However, this embedded raw envelope topography represents an unrealistic scenario because the additional weight of the material used to fill the valleys would lead to an isostatic adjustment and, thus, a general lowering of the terrain.We account for this effect by estimating plate deflection using a twodimensional model (Jha et al., 2017;Wickert, 2016).The final envelope topography that we apply is displayed in Figure 2c.A more detailed description of the topography modification scheme is presented in Text S2 in Supporting Information S1.We refer to the coarse-resolution simulation with envelope topography as TENV11 and the high-resolution simulation as TENV04.

Adjustment of Land Cover to Elevation Changes
Changes in the surface elevation of grid cells induce modifications in climate, such as temperature changes according to the local lapse rate.In turn, the local land cover would adjust to the new climate.A land cover type that is particularly sensitive to elevation is permanent ice (i.e., glacier coverage).Ice-covered grid cells exhibit distinctive surface properties (e.g., in terms of albedo) compared to unglaciated grid cells and should thus be adjusted in response to elevation changes.We perform a brief analysis of the regional line, above which permanent snow and ice prevail, based on GlobCover 2009 data (Arino et al., 2012).Based on these results, we adjust the glaciation of grid cells with changed elevation using a conservative approach (see Text S3 in Supporting Information S1).Additionally, in the case of a grid cell changing from ice-free to glaciated, there is a form of 'selfadjustment' in COSMO as such grid cells will accumulate permanent snow and will thus behave similarly to cells that are predefined as ice-covered.For other land cover classes, while their dependencies on elevation are recognized (Chang et al., 2023), they are complex and not yet fully understood in our study region.Adjusting Note.For the applied variables: 2 m temperature (T), precipitation (P), wind (W) and specific humidity (QV) at 850 hPa. a Ground in situ data was used for calibration.b Inferred from reanalysis and ground in situ precipitation data, gridded evaporation data sets and observed runoff.
these classes without solid scientific grounding could introduce further uncertainties and biases into the model.Moreover, the differences between vegetation classes (e.g., in terms of albedo) are generally less pronounced than those between ice-covered and non-glaciated grid cells.Therefore, we have opted to retain the original land cover classes, with the exception of the ice class.

Reference Data
To evaluate the model's performance, we employ a combination of in situ observations, satellite products, and reanalysis data (see Table 1 for an overview and product references).ERA5 reanalysis data are used to evaluate the large-scale circulation simulated by COSMO, as well as 2 m air temperature and precipitation.In evaluating precipitation, we additionally consider the following observation-based products: Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG), the Asian Precipitation-Highly-Resolved Observational Data Integration Toward Evaluation (APHRODITE), and the Global Precipitation Climatology Center (GPCC) data set.The first product is derived from remote sensing information and calibrated with ground in situ data, while the latter two data sets are inferred from precipitation gauge measurements only.Gauge-derived or calibrated gridded precipitation data sets tend to underestimate actual precipitation (Prein & Gobiet, 2017;Singh & Kumar, 1997), particularly in areas with complex terrain and at higher latitudes (Beck et al., 2020).Such biases are also quantified for our study region (Y.Jiang et al., 2022) and are primarily caused by two factors: first, rain gauges undercatch precipitation, particularly in wind-exposed and snow-dominated environments (Kirschbaum et al., 2017;Schneider et al., 2014).Second, precipitation gauge networks are disproportionately located in valley floors, which typically receive less precipitation than valley flanks and ridges (Rasmussen et al., 2012;Sevruk et al., 2009).GPCC is corrected for precipitation undercatch (Schneider et al., 2014) but not for the second issue mentioned above.Therefore, we consider another precipitation reference product (called PBCOR) from Beck et al. (2020).This product accounts for both undercatch and the spatial non-representativeness of gauge stations by estimating precipitation as a residual from modeled/observed evaporation and runoff.The output from this study has been applied in Prein et al. (2022) to evaluate modeled precipitation in the HMA region.Moreover, we consider hourly precipitation measurements from 62 ground-based meteorological stations of the China Meteorological Administration (CMA; see Figure 1b for station locations) to compare the impact of parameterized versus explicitly represented deep convection on modeled precipitation.We use the method outlined by Kaufmann (2008) to compare modeled precipitation with station data.For CTRL11, the station data are compared with values from the closest model grid cell.For CTRL04, we select the grid cell closest to the station's altitude within a 6 km radius.This method has previously been utilized by Ban et al. (2015) and S. Li et al. (2023) in their validation of simulated precipitation against station data.To further assess the 2 m air temperature, we consider two station-derived products: the APHRODITE daily mean temperature data set (AphroTemp), and the surface observation time-series data set from the University of East Anglia Climatic Research Unit (CRU).

Precipitation Indices and Spatiotemporal Evaluation
We use multiple statistical indices outlined in Table 2 to study the characteristics and variations of precipitation and its extremes in both observational data and model simulations.Following Ban et al. (2021), a wet day is defined as daily precipitation greater than or equal to 1 mm d 1 , and a wet hour is defined as hourly precipitation greater than or equal to 0.1 mm hr 1 .
For the majority of our analyses, we consider the rainy (MJJAS) and dry (NDJFM) seasons, which are common periods for studying Asian monsoon climate (B.Wang, 2006;B. Wang & LinHo, 2002).We mostly focus on the summer monsoon (MJJAS), because the majority of the yearly accumulated precipitation occurs in this period in the HM and the surrounding area.In the validation part (Section 3) however, we also carry out model evaluations on a seasonal basis, that is, for winter (DJF), spring (MAM), summer (JJA), and autumn (SON) over 5 years, to allow for a direct comparison with previous modeling studies (e.g., B. Huang et al. (2015); W. Zhou et al. (2016)).
For spatial analysis, we define multiple domains, which are displayed in Figures 1b and 1c.The largest domain, ET, encompasses the majority of the land area of the CPM domain and all CMA precipitation gauge stations (see a Note that all percentile indices are expressed relative to all (wet and dry) days/hours (Schär et al., 2016).
Figure 1b).The HM domain contains the majority of the area that is affected by the topographic modification scenarios (see Section 2.2).We further split this domain according to the national boundaries between China and India/Myanmar into an upstream and a center region (HMU and HMC, respectively).HMU represents the HM area that is located upstream of the prevailing atmospheric flow during the summer monsoon (see Figure 1c).For model evaluation (see Section 3.2), this domain is divided again into a northern part (HMUN), which experiences very large precipitation amounts, and a southern part (HMUS) which features a dryer climate.

Evaluation of Simulated Climate Over 2001-2005
In this section, we first validate the ability of the coarser-scale, CTRL11 simulation to reproduce the characteristics of the East Asian summer climate.We conduct an evaluation of this simulation for each season independently.To keep this section concise, we present only the results for the summer season, with those for winter, spring, and autumn available in Figures S6-S11 in Supporting Information S1 for a more comprehensive view.Subsequently, we evaluate the convection-permitting control simulation CTRL04.This evaluation places a focus on extreme precipitation indices, for which we use an extended set of rain gauge precipitation stations in China that operate at an hourly resolution.

East Asian Climate
The performance of CTRL11 in simulating the mean characteristics of the East Asian summer climate is presented in Figure 3.We remap the model outputs to the corresponding observation or reanalysis grids using bilinear interpolation for continuous variables like temperature and wind speed.Precipitation is remapped using the first-order conservative method to maintain the water budgets (P.W.Jones, 1999).Figures 3a-3c display the mean precipitation from June to August during 2001-2005 in CTRL11, IMERG, and their difference.The spatial distribution of summer precipitation over East Asia shows significant variation, and CTRL11 simulation reproduces these variations quite well with a pattern correlation of 0.77 and a mean bias of 0.17 mm d 1 .However, it is important to note the presence of compensation effects.During the summer season, areas near the southern coast of the continent, including the northeastern BoB, the northeastern Arabian Sea, the Philippine Sea, and the South China Sea (SCS), experience the highest precipitation amounts in both the simulation and the observation.
The southern flanks of the Himalayas also receive heavy rainfall due to the monsoon winds bringing moisture from the Indian Ocean and the BoB-a process effectively captured by our model.However, the summer precipitation over India and the SCS is underestimated in CTRL11 by 3-5 mm d 1 (Figure 3c).In contrast, in the mid-latitude regions of the West Pacific Ocean and the low-latitude region of the BoB, the precipitation is overestimated by approximately 5 mm d 1 .The precipitation bias pattern over the lower latitudes in CTRL11 resembles that found in previous modeling studies over this area (B.Huang et al., 2015;W. Zhou et al., 2016).Unlike previous modeling efforts (Bucchignani et al., 2014;D. Wang et al., 2013), our simulations feature lower precipitation biases over the TP, indicating potential benefits from employing a higher spatial resolution.
Figures 3d-3f illustrate the simulated and observed mean summer 2 m air temperature and the difference between the simulation and the CRU data set.CTRL11 reproduces the observed spatial pattern of surface air temperature very accurately, with a pattern correlation of 0.97.A weak cold bias is present over Siberia, while central Asia exhibits an evident warm bias.W. Zhou et al. (2016) reported a similar warm bias during the summer season in their COSMO simulations.In CTRL11, the simulated surface air temperature aligns better with observations over India, the Indochina peninsula, TP, and southeastern China compared with previous simulations (Bucchignani et al., 2014;Meng et al., 2018;W. Zhou et al., 2016).
To understand the biases in surface climatology, we compare the low-level atmospheric flow and specific humidity between CTRL11 and the ERA5 reanalysis data.Figures 3g-3i depict the spatial patterns of the wind and specific humidity at 850 hPa.The specific humidity reveals excellent spatial agreement with the reanalysis, demonstrating a pattern correlation of 0.98 and a bias of 0.01 g kg 1 .The most significant negative biases in specific humidity occur over Central Asia and Pakistan.CTRL11 simulates a stronger northerly flow over Afghanistan and Pakistan.This flow correlates with the transportation of drier continental air toward the coastal regions, which then advects over India, potentially causing the precipitation bias there.
The region of Asia experiencing the monsoon weather pattern exhibits the most distinct annual variations in precipitation, characterized by alternating dry and wet seasons synchronized with the seasonal reversal of the monsoon circulation features (Webster et al., 1998).The monsoon circulation patterns in India and East Asia have unique characteristics (Y.Ding & Chan, 2005).Figure 4 presents a Hovmöller diagram of the observed and simulated annual cycle of meridional precipitation (from 5°N to 50°N, and zonally averaged over 70-80°E and 110-120°E).The ISM's and EASM's spatiotemporal characteristics are very well captured in this representation.
It shows a generally good alignment between CTRL11 and IMERG, particularly in terms of the temporal and latitudinal progression of monsoon precipitation.CTRL11 effectively captures the gradual onset of the monsoon over India, but it does underestimate rainfall during the summer season (Figure 4a).As shown in Figure 4b, before mid-May, the main rain belt in the SCS longitudes is located south of 10°N, while a second rain belt is found in South China between 20 and 30°N.Around mid-May, the tropical rain belt suddenly shifts northward, resulting in the merging of the two rain belts.CTRL11 accurately captures this rapid onset process, which has also been documented by previous monsoon studies (Y.Ding & Chan, 2005;Matsumoto, 1997;B. Wang & LinHo, 2002).

Eastern Tibet Climate
We evaluate the accuracy of the simulated ET and HM climate by comparing it with several observational data sets.Figure 5a displays the ET-averaged seasonal precipitation cycle based on observational data, reanalysis, and model simulations.The seasonal cycle of precipitation over ET typically features a dry winter and a prolonged rainy season from May to September, with a precipitation peak in July, according to the reference data.In terms of precipitation magnitudes, both CTRL11 and CTRL04 closely match or fall within the upper bound of the reference data sets.However, it's important to note that the APHRODITE data set does not correct for any orographic effects discussed in Section 2.3.The GPCC data set, which is partially corrected, aligns better with the simulated precipitation values.The closest agreement is with PBCOR, which takes into account undercatch effects, and ERA5, a model-based data set that does not have the limitations stated in Section 2.3.A study by Y. Jiang et al. (2022) conducted for a sub-region of the ET domain, found that simulation-based precipitation data sets (e.g., ERA5) perform better than IMERG in terms of precipitation intensity.The seasonal precipitation cycle is well captured by both CTRL11 and CTRL04, although both simulations show an earlier onset of monsoon precipitation, with the annual maximum precipitation occurring in June.This bias likely stems from an early development of the summer monsoon circulation, represented by a lower-level westerly atmospheric flow, in our simulations.A study by D. Lee et al. (2016), in which COSMO was applied for East Asia, also identified an unseasonably early precipitation peak, demonstrating that improved alignment could be achieved through spectral nudging.Our analyses of the seasonal precipitation cycles for the sub-regions of ET yielded similar results to those shown in Figure 5a, so we present only the condensed results for the rainy/dry seasons and the annual averages in Figure 5b.Our simulations effectively capture the spatially different precipitation magnitudes, such as the very high summer monsoon precipitation in the HMUN region, aligning well with ERA5 and PBCOR.Both CTRL11 and CTRL04 generally overestimate precipitation in the dry season, which is likely due to the premature onset of the summer monsoon in our simulations.
Figure 5c presents our analysis of the mean seasonal cycle of 2 m temperature.Compared to the station-derived data sets and ERA5, CTRL11 exhibits a weak warm bias, while CTRL04 aligns better with the reference data sets.The model's performance across both the rainy and dry seasons shows high consistency.The HM region, as well as the ET domain, feature complex terrain that ranges from sea level to approximately 7,000 m. Figure 5d shows how well 2 m temperatures, as a function of elevation, are represented in our control experiments.The agreement with AphroTemp and CRU is excellent for both seasons but seems to deteriorate slightly at higher elevations.This might be due to the typically larger uncertainty of the reference products at higher elevations, given the sparser station coverage.Notably, CTRL04 and CTRL11 align much better with AphroTemp and CRU at higher elevations in the dry season compared to ERA5, which tends to underestimate temperature.This underestimation in ERA5 relates to the overestimation of snow coverage in ERA5 in the HMA region (Orsolini et al., 2019).In contrast, snow coverage in our simulations aligns well with observational data sets (not shown).
To further explore the impact of explicitly resolved convection on simulated precipitation, we perform a validation using data from 62 rain gauge stations across the ET that recorded hourly measurements during the simulation period.Figure 6a illustrates the comparison of observed and modeled wet-day frequency.We found that CTRL11 tends to over-represent drizzle events, with a bias of 6.86%.In contrast, CTRL04 aligns more closely with the observed data, with a bias of 0.23%.Regarding wet-day intensity, CTRL04 tends to overestimate daily precipitation, presenting a bias of 3.35 mm d 1 (Figure 6b).However, it's important to note that rain gauges are subject to precipitation undercatch issues, likely leading to observed intensities that are too small.Conversely, CTRL11 tends to underestimate daily precipitation intensity, a tendency also noted in other geographical regions (e.g., Ban et al. (2021)).Figure 6c demonstrates that CTRL04 slightly underestimates the wet-hour frequency (bias = 0.45%), while CTRL11 tends to overestimate it (bias = 4.74%).This issue of excessive drizzle in CTRL11 is a common challenge in many climate models (D.Chen et al., 2021;Stephens et al., 2010;Trenberth & Zhang, 2018).However, the enhanced convection representation in CTRL04 effectively alleviates this drizzling problem, which is consistent with a previous study by P. Li et al. (2021).In terms of simulating hourly precipitation, CTRL04 provides a more accurate representation of intensity than CTRL11, as shown in Figure 6d.CTRL11 tends to significantly underestimate wet-hour intensity, particularly at stations where heavy hourly precipitation occurs, consistent with previous studies (S.Li et al., 2023;Schär et al., 2020;Zeman et al., 2021).For locations with high hourly intensities, CTRL11 underestimates precipitation intensity by up to a factor of 3 (R 2 = 0.25)-a difference that can be essential for erosion and river runoff (Nearing et al., 2005).Overall, the model evaluation with in situ rain gauge station data suggests that high-resolution convection-permitting simulations deliver better performance in reproducing precipitation indices in this region.Consequently, the explicit representation of convection and the finer spatial grid at 4.4 km appear beneficial for simulating precipitation characteristics in our domain, which features complex terrain and a monsoondominated climate.

Results
Here we discuss the climate effects of changing the HM geometry (see Figures 1 and 2).In the first two subsections 4.1 and 4.2, we will address the impacts upon the large-scale climate (near and beyond the vicinity of the topographic modifications), and the effects upon the onset of the monsoon.As remote effects are much more pronounced when reducing the height of the HM, we will restrict discussion to TRED11 in these sections.In Section 4.3, we will discuss the effects on the regional climate in the vicinity of the HM and will address both TRED and TENV experiments.

Imprints on Large-Scale Climate
Figures 7a-7c display precipitation and low-level wind averaged over the rainy season.In CTRL11, heavy precipitation is located in the northeastern BoB, southeastern SCS and western North Pacific (WNP) (Figure 7a).In TRED11, precipitation intensity over the HM, northern BoB and northern Myanmar decreases compared to CTRL11, while precipitation increases in the northeastern TP and SCS (Figure 7c).The large-scale imprint of the topography change can be found along a southwest-northeast-oriented belt over WNP (Figure 7c).Changes in East Asian precipitation patterns agree well with a study by Yu et al. (2018), in which a similar topographic modification experiment was performed with a regional climate model nested in a global climate model.
Water vapor transport plays a pivotal role in the Asian summer monsoon system (T.Zhou & Yu, 2005).In CTRL11, the Indian monsoon transports vast amounts of moisture from the Arabian Sea and the BoB toward the HM and the Indochina Peninsula (Figure 7d).The onshore flow is compelled to rise upon reaching the coastal region of Myanmar, which is characterized by a narrow plain bordered by a mountain range.As the monsoon moves inland, it brings significant rainfall to the HM.The Indian monsoon travels across the Indochina Peninsula and the SCS then converges with the Southeast Asian monsoon, which carries moisture from the SCS and the WNP into eastern China (R. Huang et al., 1998;Simmonds et al., 1999;Renhe, 2001;T. Zhou & Yu, 2005).In contrast, the reduction of the HM in TRED11 weakens the large-scale monsoon circulation, leading to decreased eastward water vapor flux transport in the coastal region of Myanmar and upstream of the HM region (Figure 7f).This finding aligns well with Yu et al. (2018), where adding the southeastern TP strengthens the monsoon circulation and increases precipitation over the BoB.The orographically triggered precipitation in the southwestern HM also significantly decreases due to the topographic modification and the overall weaker monsoon circulation.Without the HM serving as a barrier, the warm tropical water vapor from the BoB flows northeastwards into northern China before encountering the Qilian Mountains, resulting in increased precipitation there.Furthermore, there is a reduction in moisture transport from the SCS to southeastern China, leading to increased local precipitation over the SCS region.More distantly, strong convergence of the subtropical and extratropical water vapor flux anomalies is found at approximately 30°N between 140 and 170°E, favoring strengthened precipitation over the WNP (Figure 7f).
The change in water vapor transport is closely tied to the alteration in monsoon circulation, which is in turn influenced by topography (Huber & Goldner, 2012;B. Wang et al., 2008;R. Zhang et al., 2015;Z. Zhang et al., 2004).To scrutinize the circulation changes governing water vapor transport, we examine how thermodynamic structure alters in response to topographic modifications (Figures 7g-7l).In CTRL11 featuring modern topography, the Asian landmass-including the Indian subcontinent-undergoes more rapid heating during the summer months than the surrounding ocean.This leads to the formation of a low-pressure system over the land and a persistent high-pressure system over the ocean (Figure 7j).As observed in previous studies (Boos & Kuang, 2010), the upper-tropospheric temperature displays a maximum located south of the Himalayas.Thermal forcing from continental India and the TP triggers the formation of an anticyclone in the upper troposphere (not shown).Driven by the pressure gradient, the thermal effect of land-sea contrast propels the South Asian summer monsoon circulation.In the lower troposphere, the monsoon's westerlies travel from the Indian Ocean and converge with the southwesterly trades at the low-level North Pacific subtropical anticyclonic ridge, forming the southwesterlies (Figure 7a) (Z.Zhang et al., 2004).
In TRED11, the reduced diabatic heating induces a significant cooling of the upper troposphere over the southern HM (Figure 7i).The reduction in diabatic heating leads to an anticyclonic change at lower levels and a cyclonic change at upper levels.In the upper troposphere, a barotropic cyclone is found over the WNP, originating in the TP and moving along the upper-level westerly jet stream (Figure 7i).At lower levels, the weakened India westerlies give rise to decreased water vapor transport.Additionally, cooling of the lower atmosphere over the SCS suppresses the Walker circulation over the Indian Ocean, resulting in an overall weakening of the monsoon circulation (Figure 7l).Remotely, the atmospheric response propagates northeastward along the monsoon winds and favors the cyclonic change pattern to the east of Japan (Figure 7f).This circulation pattern curtails the water supply along the northwestern flank of the western Pacific subtropical high, causing decreased precipitation over the coastal region of northeastern China, the Korean Peninsula and Japan.
The effects of the envelope topography on precipitation are more localized and less pronounced due to the smaller relative change in mountain volume.The influences of both the envelope and reduced topography on the local HM climate, with particular emphasis on (extreme) precipitation indices, will be discussed in Sect.4.3.

Effect of Topographic Changes on Monsoon Precipitation Onset
Figure 8 shows the Hovmöller diagrams that illustrate the seasonal precipitation cycle, which is zonally averaged over the BoB, HM and eastern China.The shift from the dry season to the rainy season is vividly depicted in the latitude-time cross-sections of mean precipitation.We first discuss the situation in the CTRL11 climate (left-hand panels in Figure 8).The transition from the dry to rainy season upwind of the HM happens quite suddenly around the latitude of approximately 25°N, typically occurring around mid-March (Figure 8a).Before this transition, the rainfall belt remains relatively stable over the southern BoB, located south of 10°N (Figure 8a).However, after mid-March, there's a noticeable northward shift in the near-equatorial rainfall belt.This belt gradually moves northwards, merging with the HM rainfall belt by mid-May.This gradual migration is in contrast to the abrupt transition observed in Myanmar (Figure 8b).There, a substantial increase in rainfall occurs early in May, which signifies the onset of the monsoon over the Indochina peninsula.This onset process aligns with observations documented in previous studies (Y.Ding & Chan, 2005;B. Wang & LinHo, 2002).Over the SCS, the rainy season typically commences around mid-May, as shown in Figure 8c.This occurrence is a result of the eastward expansion of the southwesterly monsoon into the SCS region, accompanied by the eastward retreat of the western Pacific subtropical high (Z.Zhang et al., 2004).
After reducing the HM's elevation (TRED11, middle panels in Figure 8), both the shift from the dry season to the rainy season and the precipitation intensity experience notable changes.However, the effects vary across different regions.Over Bangladesh and northeasternmost India, the onset of the rainy season is delayed by approximately one month, starting around mid-April.Additionally, precipitation intensity throughout the rainy season typically decreases by approximately 10 mm d 1 (Figure 8a).In the northern BoB, while the start of the rainy season remains consistent, there is a noticeable decrease in precipitation intensity.Over the HM, the precipitation intensity during the rainy season also declines, but not as significantly as it does upwind, underscoring the role of the mountains in orographic rainfall (Figure 8b).Over the SCS, we observe an increase in rainfall in July and August, which is consistent with our previous discussion.The mountains affect the surrounding circulation, reducing the amount of water transported to mainland China, and subsequently increasing local rainfall in the SCS (Figure 8c).Nonetheless, the Hovmöller diagram reveals that the thermal forcing of the HM, which impacts the circulation, begins to exert its influence at a later stage during the advance of the Asian summer monsoon.This observation aligns with previous research by Z. Zhang et al. (2004).

Effects on Regional Climate
The evaluation presented in Section 3.2 reveals that the ET/HM climate, particularly mean rainy season precipitation in terms of patterns and magnitudes, is overall very similar between CTRL11 and CTRL04.However, when considering precipitation indices investigated in this section, CTRL04 generally outperforms CTRL11 (see Figure 6).For these reasons, we have opted to discuss the results of the CPM simulations exclusively in this section.Figure 9 shows the maps of vertically integrated water vapor flux, precipitation indices and convective available potential energy (CAPE) over the HM.Statistics over the HM and its sub-regions are computed over the rainy season and presented in Table 3.
Figure 9a depicts the water vapor transport in the ET region during the rainy season in CTRL04.The atmospheric water flux is approximately parallel to the elevation gradient on the southwestern side of the HM.This causes the distinctive spatial distribution of climatological rainy-season precipitation, which leads to pronounced orographic precipitation in easternmost India and northernmost Myanmar, as shown in Figure 9d.A secondary peak is visible at the western side of the Sichuan Basins (WSSB).The average daily precipitation during the rainy season and simulation period upwind of the HM amounts to 12.7 mm d 1 .Over the HM, high precipitation amounts often coincide with local topographic peaks, whereas the valleys often receive smaller precipitation amounts due to rain-shadow effects.On average, the daily precipitation over the central HM is 7.2 mm d 1 .Figures 9g and 9j show the extreme daily precipitation p99D and extreme hourly precipitation p99.9H in CTRL04.For both extreme precipitation indices, maxima are found southwest of the HM, along the Indian/Myanmar border, and over the BoB and its adjacent land area.In the area upwind of the HM, p99D averages to 97.0 mm d 1 , while p99.9H reaches 29.4 mm hr 1 .In contrast to mean precipitation, the distinct signature of the eastern HM is not evident, with p99D and p99.9H reaching 56.5 mm d 1 and 17.3 mm hr 1 in HMC, respectively.Central China experiences more intense extreme precipitation compared to the central and eastern HM.This pattern reflects the distribution of the convective available potential energy (CAPE) and is consistent with the fact that daily/hourly precipitation extremes are more related to convective-triggered precipitation events (i.e., thunderstorms) than to orographically induced or stratiform precipitation (Figure 9m).
In TRED04, the absence of a topographic barrier that alters atmospheric circulation leads to a shift in the direction of water vapor flux to the northeast (Figure 9b).This change results in a 33% decrease in mean precipitation upwind of the HM and an 18% reduction over the central HM.Conversely, precipitation increases in the northern HM (Figure 9e).Figures 9h-9k display the changes in extreme daily precipitation p99D and extreme hourly precipitation p99.9H between CTRL04 and TRED04.Over the HM region, where topographic changes exceed 500 m, the spatial patterns of different precipitation indices exhibit substantial variation.The distribution of changes in extreme daily precipitation displays a distinct pattern (Figure 9h), as the northern part of HM experiences an increase in extreme daily precipitation after elevation reduction, while the rest remains almost unchanged (Figure 9h).On average, the HMC region sees an increase of 8%, while the upwind region experiences a decrease of 12%.Moreover, changes in extreme hourly precipitation contrast with that of mean precipitation, with nearly the entire region with modified topography experiencing an increase in extreme hourly precipitation, averaging to an increase of 20% (Figure 9k).We assume that this more uniform change in hourly extreme precipitation is caused by a combined effect of higher surface temperatures and a deeper atmosphere, which favors convection.This hypothesis is confirmed by the change in simulated CAPE as seen in Figure 9n.Specifically, the increase in CAPE is most prominent in the central and southern HM in TRED04.In addition to changes in precipitation, there is a notable decrease in net water flux at the surface (i.e., runoff) across the entire HM region, amounting to a 40% decrease.This includes a substantial decrease of 51% in runoff upwind of the mountains and a more moderate reduction of 35% over the HMC region.
The summer mean precipitation in TENV04 exhibits two peaks, similar to the CTRL04 simulation, with one located over the western HM and the other over the WSSB (not shown).Figure 9c shows the spatial distribution and magnitude of differences between CTRL04 and TENV04 for integrated water vapor flux.The topographic change in TENV04 results in less moisture transport from the ocean.However, the western HM experiences a Journal of Geophysical Research: Atmospheres 10.1029/2023JD040208 XIANG ET AL.
very small increase in precipitation (see Figure 9f) probably due to enhanced orographic precipitation caused by the larger mountain volume (Imamovic et al., 2019).A few dry valleys in the north, such as the Three Parallel Rivers (i.e., Salween, Mekong, and Yangtze), experience increased precipitation in the TENV scenario due to the vanished rain shadowing effect.However, in the majority of the central and eastern HM region, mean precipitation during the rainy season decreases substantially ( 19%), amounting to a very similar reduction as in TRED.On the WSSB, the upward motions play a crucial role in the changes in precipitation (Tao et al., 2020).A smoother terrain over the HM in TENV04 leads to a more streamlined atmospheric flow, with less turbulence and mixing, which inhibits the formation of clouds and precipitation.This result is explained through differences in vapor transport and stability between CTRL04 and TENV04 in the following section.Figure 9i shows changes in extreme daily precipitation in TENV04, which largely mirror the spatial pattern of changes in mean precipitation.These changes include an increase in heavy daily precipitation over the western HM and a decrease in the northeastern HM. Figure 9l reveals that the spatially coherent decrease in precipitation indices for the northeastern HM is not apparent for hourly extreme precipitation, which is consistent with the change in CAPE, as shown in Figure 9o.Compared to CTRL04, the simulated CAPE over the HM in TENV04 decreases, although the change is very small compared to changes in TRED04.This is reflected in the modest and less consistent changes observed in extreme hourly precipitation.Notably, the envelope topography resulted in a 26% reduction in surface net water flux over the HMC.This reduction suggests a positive precipitation-erosion feedback for this region, where highrelief topography favors conditions for increased mean precipitation, which accelerates erosion and the further formation of a more pronounced terrain relief.
To further analyze thermodynamic and dynamic processes during the rainy season, we examine how the alongsection wind, moisture, vertical velocity, total diabatic heating, and equivalent potential temperature (θ e ) change at different atmospheric heights with modified HM geometries.Figure 10 shows a transect that crosses the HM and is approximately parallel to the prevailing (lower-level) wind direction (see top left of Figures 10a and 1b for location).
By examining the distribution of precipitation depicted in Figure 9a, it is evident that the western boundaries of HM, facing the windward direction, receive a larger proportion of rainfall compared to other orographic features (e.g., WSSB at ∼105°E) located further downwind.The reduction in precipitation observed in areas downwind can be attributed to variations in specific humidity (Figure 10a).The vertical transect of total diabatic heating across the HM reveals two distinct maxima of upward motions (Figure 10d), one at the southern flanks of the Himalayas at ∼92°E and another over the eastern HM, where the significant upward motion can reach up to the 200 hPa pressure level.On the southern flanks of the Himalayas, the surface fluxes from the non-elevated part of northern India play an important role in the large-scale South Asian monsoon by changing the meridional temperature gradient between northern India and the equator (Boos & Kuang, 2013).The precipitation on the WSSB is mainly caused by the vertical moisture flux convergence (Tao et al., 2020) and is related to the vertical distribution of upward motions (Figure 10d).In the southwestern HM, upward motions and diabatic heating are centered near the surface of the windward slopes.This suggests that mechanical lifting due to orographic forcing is a contributing factor.The topography of the HM acts as a barrier to the southwest winds, leading to the generation of lower-level convergence, which contributes to horizontal moisture flux convergence and upward motions.Figure 10b displays the moisture availability and along-section wind in the reduced topography experiment, which reveals an intensification of south-westerly winds and a decrease in moisture supply compared to CTRL04.
Comparing the diabatic heating over the HM between CTRL04 and TRED04 (Figures 10d and 10e), it is apparent that the reduction of the mountain range significantly weakens the diabatic heating and the upward movement over the mountains, especially over the eastern HM where the moisture flux convergence is an important factor for local precipitation.Moreover, the reduction of the mountain range has a significant impact on diabatic heating to the west of the mountain range at ∼92°E (Figure 10e).Additionally, the vertical transects of θ e across the HM (Figures 10g and 10h) reveal decreased values in TRED04 at intermediate heights relative to CTRL04, indicating a less stable atmosphere in TRED04, favoring higher convective activities (i.e., heavy hourly precipitation).These findings suggest that the HM affect the Asian monsoon through both orographic insulation and plateau heating.
The general patterns of moisture and along-section winds are very similar in CTRL04 and TENV04 (Figures 10a  and 10c).However, differences in the strength of winds and the availability of moisture do exist.In TENV04, southwesterly winds are stronger over the mountains, which contributes to the intensified precipitation on the windward slopes (Figure 9f).The presence of filled valleys in TENV04 leads to an overall increase in surface elevation, which results in a reduction of near-surface specific humidity over the HM.This reduction can be attributed to lower temperatures and saturation vapor pressure at higher elevations.Apart from the direct changes in elevation, the filled valleys also create a more effective barrier to moisture flow, increasing the depletion of water vapour due to orographic precipitation.This, in turn, limits the amount of moisture that can be transported further into the interior of the region.Figure 10f shows the vertical transect of vertical velocity and total diabatic heating in TENV04.Comparing these results with CTRL04 reveals a reduction in diabatic heating and upward movement over the eastern HM.Inspection of θ e shows decreased near-surface values in TENV04 relative to CTRL04 (Figure 10i).The modified topography obstructs the transport of moisture to the eastern HM and the WSSB, resulting in a more stable atmosphere.To complement the more localized perspective of the envelope experiment, the diurnal cycle of valley winds across the Three Parallel Rivers, averaged over the rainy season, is shown in Figure 11 for both the CTRL04 and TENV04 scenarios.At the end of the night (04:00), CTRL04 simulates an almost quiescent atmosphere at the surface of the river valleys (Figure 11a).By late morning (10:00), the up-valley flow begins to develop, becoming more pronounced in the southern Mekong and Yangtze river valleys (Figure 11b).In the late afternoon (16:00), the up-valley wind intensifies in the Yangtze and Dadu river valleys, predominantly flowing in a south-north direction.However, in the Salween and Mekong valleys, the up-valley wind is noticeable only in the southern part of the domain, while in the north, southeastern winds prevail (Figure 11k).Finally, during the evening (22:00), the Yangtze and Dadu exhibit a light up-valley flow (Figure 11l).In the Salween and Mekong, the flow is dominated by weaker southeasterlies compared to those in the afternoon (Figure 11d).Compared to CTRL04, TENV04 exhibits much stronger surface winds throughout the day, attributed to decreased surface roughness (Figures 11e-11h).However, the along-valley surface wind component, perpendicular to the cross-section, is significantly weaker than in CTRL04 (Figure 11o).This difference is particularly evident in the eastern region of the HM, where the predominant surface wind runs from south to north.This configuration leads to a more stable atmosphere, consequently resulting in a reduction in precipitation over the HM.

Discussion and Conclusions
In this study, we applied the limited-area model COSMO with a large-scale simulation at a horizontal resolution of 12 km, covering an extended CORDEX East Asia domain, and a nested convection-permitting simulation at a horizontal resolution of 4.4 km, covering the Hengduan Mountains (HM), including parts of southwestern China and Indochina.We first evaluated the model's ability to simulate the control climate for the period 2001-2005 (CTRL).We then proceeded with two sensitivity experiments involving modified HM topography scenariosa first scenario with a spatially heterogeneous reduction of the HM (TRED) and a second scenario with an envelope topography, in which the deep valleys were filled (TENV).To our knowledge, this is the first terrain modification experiment conducted over the Tibetan Plateau (TP) using a convection-permitting model (CPM).The improved representation of precipitation frequency and intensity in CPMs allowed us to study the sensitivity of these processes on the mountain geometry.This approach also allowed us to investigate the complex interplay of HM topography, large-scale processes and localized convective systems.The main findings of these experiments are summarized below, followed by a section, in which we embed the results in a broader context, and an outlook.
1. Validation results demonstrate the ability of the control simulations (using 12 and 4.4 km grid spacings) to simulate the climate over East Asia and the HM region for the period 2001-2005.The simulated precipitation reproduces the spatial variations well, albeit with a slight underestimation over India and the South China Sea (SCS).Moreover, our simulation features lower precipitation biases over the TP compared to previous modeling efforts owing to a higher spatial resolution (B.Huang et al., 2015;D. Wang et al., 2013;W. Zhou et al., 2016).The simulated monsoon reproduces the temporal and latitudinal progression of both the Indian and East Asian monsoon precipitation.Over the HM, both CTRL11 and CTRL04 capture the seasonal precipitation cycle well, but reveal an onset of the summer monsoon that is seasonally too early.An additional validation against in situ rain gauge station data reveals that the explicit representation of convection at finer spatial resolution is beneficial for reproducing accurate magnitudes of wet day frequencies and the spatial range of precipitation intensities on a daily/hourly scale.2. TRED results show that the HM acts as a topographic barrier, resulting in pronounced orographic precipitation in easternmost India and northernmost Myanmar.The study also reveals an increase in diabatic heating over the uplifted HM, which triggers circulation changes around the uplifted region and strengthens the westerly wind from the ocean in South Asia.Consequently, there is a marked intensification of precipitation in Indochina and southwestern China, along with decreased precipitation in the SCS.Additionally, the strengthened cyclonic circulation in the Bay of Bengal extends eastward, indicating an intensification of the East Asian summer monsoon upon the uplift of the HM.However, the uplift of the HM causes a shallower and more stable atmosphere locally, leading to less convective activity and thus decreased extreme hourly precipitation.3.In contrast to TRED, the TENV's remote effects on climate are negligible.TENV results indicate that the removal of valleys is associated with an overall reduction in precipitation and runoff.In the HM upstream region, spatially integrated precipitation slightly increases, but the central and eastern HM experience a marked drying.This finding suggests a positive feedback mechanism between precipitation and erosion-at least for this region with its specific terrain configuration and flow regime during monsoon.
Geological evidence shows that the southern two-thirds of the HM have grown higher in the latest Miocene or Pliocene (Hoke et al., 2014).Additionally, geological studies indicate that northeastern India experienced a more humid climate between the Late Miocene to Pliocene (Hoorn et al., 2000).Thus, both the geological evidence and the simulations conducted in this study support the notion that the uplift of the HM contributes to the intensification of the Asian monsoon.However, some relations remain uncertain.Molnar and Rajagopalan (2012) linked the more arid northwestern Indian subcontinent between 11 and 7 million years ago to the growth of the eastern margin of the TP.While in our study, the reduction in topography does not result in a significant change in precipitation in northwestern India.Therefore, if the uplift of the eastern TP is not the primary cause, the arid climate in northwestern India may be more closely related to the global climatic cooling (H.Lu & Guo, 2014).
The HM's complex interaction with monsoon systems has created a complex regional and local climate, where dissected topography from erosion further enhances precipitation.This unique feedback between topography and climate has likely shaped the complex topographic and climatic heterogeneity of the region, providing a wide diversity of habitats for species (Antonelli et al., 2018).Therefore the unique combination of tectonic uplift and the monsoon system has created unique conditions for biodiversity (W.-N.Ding et al., 2020).
Further studies are needed to assess the influence of different HM geometries on both regional and large-scale climates under different climate conditions.Specifically, it would be intriguing to explore whether the observed climate response to reduced HM topography is consistent across different paleoclimates, such as the Last Glacial Maximum with globally colder temperatures or periods of warmer temperatures.Another compelling area for investigation involves examining if imprints of topography on large-scale circulation depend on atmospheric oscillations or modes, such as the El Niño-Southern Oscillation and Indian Ocean Dipole, which are both thought to influence the interannual variability of the Asian summer monsoon (Pothapakula et al., 2020).Addressing this question would necessitate longer simulation periods; however, the substantial computational costs of fine-scale, convection-permitting simulations currently pose a significant challenge.With a resolution of 4.4 km, we are able to resolve the main valleys of the HM (see Figure 2a)-however, local wind systems that could influence precipitation are still not fully resolved.Running simulations with even finer grid spacings would therefore shed more light on the complex influence of (small-scale) terrain relief on precipitation formation.
Regarding the envelope topography experiment, we noted that lower-level atmospheric flow is predominantly perpendicular to the main valleys and obtained results might therefore be limited to this specific configuration.
Additional experiments with more valley-aligned flow would thus nicely complement the findings of this study.

Figure 1 .
Figure 1.Overview of the COSMO domains used in this study.We apply (a) a large-scale domain at 12 km grid spacing (LSM) and (b) a nested domain at 4.4 km grid spacing (CPM).Black circles in (b) denote 62 precipitation stations in China considered for model evaluation.Additionally, the dashed outlines highlight the region of eastern Tibet (ET) and Hengduan Mountains (HM).In (b), the blue line represents a transect used in Section 4, which crosses the HM and is approximately parallel to the prevailing wind direction.Panel (c) shows the precipitation (units: mm d 1 ) and vertically integrated water vapor transport (units: kg m 1 s 1 ) during the rainy season averaged over the year 2001-2005 from IMERG and ERA5, respectively.We further divide the HM into three subregions, including two upstream regions (HMUN, HMUS) with relatively high and low precipitation amounts, respectively, and one downstream region (HMC).

Figure 3 .
Figure 3. Spatial distributions of JJA (a-c) precipitation (units: mm d 1 ), (d-f) 2 m air temperature (units: °C) and (g-i) 850-hPa wind (vector; units: m s 1 ) and specific humidity (shading; units: g kg 1 ).All quantities are averaged over the period 2001-2005.The first column displays the CTRL11 model, the second one observations and the third one their differences.Correlation coefficients and spatially integrated biases are indicated in the upper-right part of panels (c), (f) and (i) for precipitation, 2 m air temperature and specific humidity, respectively.Missing values in IMERG and CRU are represented by gray areas.

Figure 4 .
Figure 4. Hovmöller diagrams of the seasonal precipitation cycle zonally averaged over (a) 70-80°E and (b) 110-120°E (units: mm d 1 ).A 5-day moving average has been applied to the 5-year climatology to remove high-frequency variability.

Figure 5 .
Figure 5. Seasonal cycles of (a) precipitation and (c) 2 m temperature of control simulations and the reference data sets averaged over the Eastern Tibet domain.Temporally integrated quantities over the rainy (MJJAS) and dry (NDJFM) seasons (and the entire year) are displayed on the right.Panel (b) shows precipitation for the rainy/day season and averaged over the year for the Hengduan mountains sub-regions.Note the different y-axis range.The brown boxes in panel (a) and panel (b) specify the uncertainty range of PBCOR for the annual values.Panel (d) displays the 2 m temperature as a function of elevation for the rainy and dry seasons integrated over the HM region.

Figure 6 .
Figure 6.Validation of JJA precipitation for ERA5-driven simulation with 12 km (CTRL11, green) and 4.4 km (CTRL04, blue) grid spacing with in situ precipitation data from 64 stations in China: (a) wet day frequency (units: %), (b) wet day intensity (units: mm d 1 ), (c) wet hour frequency (units: %), and (d) wet hour intensity (units: mm h 1 ).R 2 denotes the square of the correlation coefficient between the models and observations.

Figure 7 .
Figure7.Maps of (a-c) Precipitation (contour; units: mm d 1 ) and 850-hPa wind (vector; units: m s 1 ), (d-f) vertically integrated water vapor transport (units; kg m 1 s 1 ), (g-i) 200-hPa temperature (shading; units: K) and geopotential height (contour; units: meters) and (j-l) 500-hPa temperature (shading; units: K), geopotential height (contour; units: meters) averaged over the rainy season (MJJAS) from the year 2001-2005.From left to right are the results from CTRL11, TRED11 and their differences, respectively.The green line in the difference maps indicates regions with topographic changes greater than 500 m.

Figure 9 .
Figure 9. (a-c) Vertically integrated water vapor flux, (d-f) mean precipitation, (g-i) the 99th percentile of daily precipitation (p99D), (j-l) the 99th percentile of hourly precipitation (p99.9H) and (m-o) convective available potential energy (CAPE) during the rainy season.From left to right are the results from CTRL04 and the differences between TRED04 and TENV04 with respect to CTRL04.Regions with topographic changes greater than 500 m are delineated by the green line in the differences maps.

Figure 11 .
Figure 11.Mean diurnal cycle of the surface winds during the rainy season (MJJAS).Left-hand panels show the wind speed (shading) and direction (vectors) at 10 m above ground for CTRL04 (a-d) and TENV04 (e-h).Gray shading indicates the terrain height (1,000 m contour interval) and the bold black line indicates the 4,000 m contour.Right-hand panels show a vertical cross-section through the Three Parallel Rivers (see green line in panel (a)) for CTRL04 (i-l) and TENV04 (m-p).It depicts the wind component normal to the cross-section (shading; positive for northerly flow and negative for southerly flow) and potential temperature (gray contours; units: K).Local time is indicated in the bottom right corner.

Table 1
Overview of the Applied Reference Data in This Study

Table 2
Precipitation Indices Applied in This Study a

Table 3
Changes in Precipitation in the Hengduan Mountains and Its Sub-Regions (Figure1c) for the Topographic Modification Experiments With Reduced Topography (TRED04) and Envelope Topography (TENV04) Statistics are computed over the rainy season (MJJAS) and the years 2001-2005.P refers to mean precipitation, p99D to the daily 99th percentile, p99.9H to the hourly 99.9th percentile and P Q to precipitation minus evaporation (i.e., the net water flux at the surface).