Convection-Permitting Simulations of Current and Future Climates over the Tibetan Plateau

The Tibetan Plateau (TP) region, also known as the “Asian Water Tower” or the “Third Pole”, contains the largest ice mass outside the polar regions. The glaciers and regular snow melt from this region form the headwaters of over 10 of Asia’s largest rivers, providing water to over 2 billion people (Immerzeel et al., 2010; Yao et al., 2012, 2022). Therefore, projections of future climate change in this region are crucial for climate change adaptation activities in this vulnerable and sensitive region.

Climate models have been widely used as tools for understanding climate processes and projecting climate changes. Due to complex topography and landscapes, climate models with coarse spatial resolution generally have difficulty reproducing the climate characteristics over the TP region (Su et al., 2013). The global climate system models included in phase 6 of the Coupled Model Intercomparison Project (CMIP6), with a horizontal resolution of 100–200 km, exhibit very large wet and cold biases over the TP region and no evident improvements were found from CMIP5 to CMIP6 (Zhu and Yang, 2020). Atmospheric general circulation models or regional climate models, with resolutions of 20–50 km, partly alleviate the too-wet and too-cold biases, but most of the biases still remain (e.g., Feng et al., 2011; Ji and Kang, 2013; Gao et al., 2014, 2015; Niu et al., 2021a).

Convection-permitting modeling (CPM), with a horizontal resolution of less than 4 km, which can explicitly deep convection and does not need deep convection parameterizations (Prein et al., 2015; Stevens et al., 2020), exhibits great advantages in reproducing the climate over the TP region. For example, kilometer-scale modeling outperforms the latest global reanalysis and the High Asia Refined regional reanalysis in simulating the summer 10-m wind speed and precipitation over the central and eastern TP region (Zhou et al., 2021). This scale of modeling reduces the water vapor transport through the central Himalaya toward the TP and thus the wet biases over the TP in coarse-resolution models (Lin et al., 2018). The reduction in wet biases in CPM over the central and eastern TP is also linked to a better simulated diurnal cycle of rainfall, with the removal of spurious afternoon rainfall (Li et al., 2021; Liu et al., 2022), which is associated with an overly strong dependence on convective available potential energy (Li et al., 2021). Furthermore, CPM tends to improve the simulation of surface evaporation over the TP region, resulting in added value in simulating the precipitation recycling ratio (Zhao et al., 2021).

Given the importance of CPM for the TP region, an internationally coordinated project, CPTP (Convection-Permitting Third Pole), was endorsed by WCRP-CORDEX as a Flagship Pilot Study (FPS) in 2019, with the aim of enhancing our understanding of the water cycle over the TP region (Prein et al., 2023). Multimodel intercomparisons of case studies indicate that kilometer-scale modeling over the TP region shows high performance in simulating precipitation and near-surface temperature across a range of meteorological situations (Prein et al., 2023).

To date, projections of climate changes over the TP region under warming scenarios have mainly been based on CMIP models with resolutions of 100–200 km (e.g., Zhou and Zhang, 2021; Zhao et al., 2022; Jiang et al., 2023) or dynamical downscaling of regional climate models (RCMs) with resolutions of 10–40 km (e.g., Ji and Kang, 2013; Gao et al., 2018; Zhang et al., 2019; Niu et al., 2021b; Fu and Gao, 2023). Comparisons of climate changes over the TP region projected by RCMs and the driving GCMs indicate that the projected climate change signal is resolution-dependent. The driving GCM with coarse resolution projects an overall uniform wetting over the TP region under warming scenarios, while a diverse pattern with wetting in the north and drying in the south is found in dynamical downscaling (Gao et al., 2018). Meanwhile, the future precipitation-minus-evaporation changes over the TP region under the warming scenario in driving GCMs are dominated by the thermodynamic component, whereas in RCM projections they are mainly driven by the dynamic component (Zhang et al., 2019). Ma et al. (2023) conducted a gray-zone resolution (9 km) climate change projection with the WRF model over the TP region, in which deep convection parameterization was turned off, and suggested that substantial warming and a decrease in snow are likely over the western TP by the end of the 21st century.

Convection-permitting (resolution: < 4 km) projections over the TP region have not been reported. In this study, with a focus on the TP region, we use the Icosahedral Nonhydrostatic Weather and Climate Model (ICON) with a convection-permitting resolution (3.3 km) to conduct decade-long present-day climate simulations and future climate projections using a pseudo-global warming approach under the SSP5-8.5 scenario. The objectives of the study include the following:

(1) To assess the skill of the kilometer-scale convection-permitting simulation in capturing the seasonal mean climate over the TP region.

(2) To assess the projected changes in seasonal mean precipitation and surface air temperature over the TP region and the associated elevation-dependent characteristics from ICON and the CMIP6 models.

The remainder of the paper is organized as follows. The model, the experimental design, and the datasets are described in section 2. The general performance of ICON in the simulation of precipitation and surface air temperature is evaluated in section 3. The projected changes in seasonal mean precipitation and surface air temperature and the associated elevation-dependent characteristics are given in section 4. A summary and conclusion are provided in section 5.

The convection-permitting simulations are performed with ICON, version 2.6.1, in its limited area mode. ICON was developed by the German Weather Service and the Max Planck Institute for Meteorology (Zängl et al., 2015). The model is formulated on an icosahedral-triangular grid, with a vertical height-based coordinate system. The following model physics are employed: the Tiled TERRA land surface scheme (Schulz et al., 2016), one-moment cloud microphysical scheme with graupel (Baldauf et al., 2011), RRTM radiation scheme (Mlawer et al., 1997), and Prognostic TKE for planetary boundary layer parameterization (Raschendorfer, 2001). Both the deep cumulus parameterization and the shallow convection scheme were switched off. This model configuration has been included in CORDEX CPTP phase I for case studies and performed one of the best among the models in simulating precipitation (Prein et al., 2023).

The regional simulations were performed over the domain of 17°–45°N and 64°–112°E (Fig. 1a). The model domain covered the entire TP and could be divided into 12 major river basins (Fig. 1b). The total number of grid cells within the model domain was 1 442 636 (the horizontal resolution was approximately 3.3 km). There were 75 layers in the vertical direction, with the model top at 20 km. The buffer zone was set to 8 grid cells.

Figure 1.  (a) ICON model domain and topographic elevation (units: m). The red frame indicates the domain of (b). (b) Divisions of 12 river basins over the TP and locations of the 81 gauge stations (blue dots) over the TP used in this study. The TP is defined as regions with elevation higher than 2400 m.

Two sets of 12-year continuous numerical simulations were conducted. The first set of experiments was a control simulation to reproduce the current climate. This control simulation was forced by the fifth major global reanalysis produced by ECMWF (ERA5; Hersbach et al., 2020), which has a horizontal resolution of 0.25° × 0.25° and is updated every 3 hours. The soil moisture and temperature were initialized with ERA5 data. The simulation extended from September 2008 to September 2020, and no spectral nudging was applied during the integration.

The second simulation was the climate change experiment, following the pseudo-global warming (PGW) approach (e.g. Rasmussen et al., 2011). During the integration (September 2008 to September 2020), the model was driven by the 3-hour ERA5 data plus projected climate change information under the SSP5-8.5 scenario:

The climate change perturbations $\Delta $CMIP6_SSP5-8.5 were the CMIP6 multimodel ensemble-mean monthly climate changes under the SSP5-8.5 scenario, which were obtained from

The perturbed climate variables included horizontal wind, vertical velocity, temperature, specific humidity, sea surface temperature, sea level pressure, cloud liquid water content, and cloud liquid ice content. The PGW approach assumes that the year-to-year variability at the boundaries does not change in the future climate. Previous studies have suggested that this approach is suitable for assessing the climate change signal related to the large-scale thermodynamic effect and lapse-rate effect (Kröner et al., 2017). This method has been employed to assess climate changes over North America (e.g., Rasmussen et al., 2011; Liu et al., 2017) and East China (Tang et al., 2023). In the PGW simulation, the greenhouse gas concentrations were set to the values of the SSP5-8.5 scenario in 2091. In both the control and PGW simulations, the first year (September 2008 to September 2009) was regarded as the “spin-up” time, and the results are excluded in the following analysis.

In our study, the ensemble-mean monthly climate change signal was created from 18 CMIP6 models (Table 1), which provided all the necessary climate variables. The perturbed climate changes are briefly described below. Figure 2 shows the projected changes at 850 hPa derived from 18 CMIP6 model ensemble means under the SSP5-8.5 scenario between the future climate period (2081–2100) and the present climate period (1995–2014). The warming (2.3°C–7.8°C) in the lower troposphere is significant around the TP, with the largest warming occurring over the northern TP in the summer season (Fig. 2c). Accompanied by low-level warming, the specific humidity is increased by 11%–70% following the Clausius–Clapeyron relationship. The low-level zonal and meridional wind perturbations are less than 1 m s⁻¹ in the winter and spring seasons, while in the summer and fall seasons the wind anomalies are slightly larger (−1.9 to 1.5 m s⁻¹), especially at lower latitudes.

Figure 2.  Spatial distributions of the 18 CMIP6 model ensemble mean seasonal differences in horizontal winds (vectors; units: m s⁻¹), air temperature (shading; units: °C), and the percentage changes in specific humidity (contours; %) at 850 hPa between the future (2081–2100) under the SSP5-8.5 scenario and the past (1995–2014) for (a) DJF, (b) MAM, (c) JJA, and (d) SON. The white area represents the TP, where the surface pressure is lower than 850 hPa.

The thermodynamic perturbations in the middle troposphere (Fig. 3) are dominated by significant warming (4.2°C–6.3°C) and a large increase in specific humidity (36%–92%), especially in summer and autumn at higher latitudes. For the wind perturbations, easterly anomalies are evident in the middle troposphere, with the largest anomalies being found in summer (−2.18 m s⁻¹) and winter (−2.35 m s⁻¹). In the upper troposphere (Fig. 4), the greater warming at lower latitudes (up to 8°C) than at higher latitudes strengthens the meridional temperature gradient. This enhancement increases the westerlies in the upper troposphere (1.9–4.4 m s⁻¹), with the largest increase in the spring season (Fig. 4b).

Figure 3.  As in Fig. 2 but for 500 hPa.

Figure 4.  Spatial distributions of the 18 CMIP6 model ensemble mean seasonal differences in horizontal winds (vectors; units: m s⁻¹) and air temperature (shading; units: °C) at 200 hPa between the future (2081–2100) under the SSP5-8.5 scenario and the past (1995–2014) for (a) DJF, (b) MAM, (c) JJA, and (d) SON.

Due to the harsh environment, in situ observations are quite limited over the TP region. To evaluate the performances of ICON over the TP, the following datasets were employed:

(1) The daily precipitation amount and near-surface air temperature at 81 national ground stations (blue dots in Fig. 1b) mainly over the central and eastern TP from the China Meteorological Administration (CMA; downloaded from http://data.cma.cn/en) for the period from October 2009 to September 2020. Quality control of this dataset has been performed by the CMA, including several consistency checks and suspected and erroneous data corrections (Liu and Ren, 2005). This dataset is nearly complete, with less than 5% missing data each year.

(2) The daily precipitation amount derived from NASA’s Global Precipitation Measurement (GPM) Integrated MultisatellitE Retrievals for GPM (IMERG) product (IMERG, hereafter) (Huffman et al., 2019) for the period from October 2009 to September 2020. The IMERG product, with a horizontal resolution of 0.1°, merges the available satellite microwave and infrared satellite precipitation estimates and was calibrated using precipitation gauge observations.

(3) The daily near-surface air temperature (T2m) with a horizontal resolution of 0.25° derived from the Global Land Data Assimilation System (GLDAS) (Rodell et al., 2004) for the period from October 2009 to September 2020. With advanced land surface modeling techniques, GLDAS assimilated ground-based and satellite observational data.

Note that both the IMERG rainfall and GLDAS T2m have also been employed as the reference datasets in the CORDEX CPTP FPS case study for multi-RCM intercomparisons (Prein et al., 2023). When comparing to meteorological station data, the model output and the gridded reference datasets at the grid points nearest to the stations were selected. The dates with missing values in the station data were also set to missing values in the model output to match the observations. In addition, the model output was interpolated to the grids of the reference datasets (precipitation from IMERG or T2m from GLDAS) using areal conservative remapping to calculate the mean biases, spatial pattern correlation coefficient (SCC), and root-mean-square error (RMSE).

In this section, the simulated seasonal mean and annual cycle of precipitation and T2m over the TP are assessed against the station data and the reference datasets.

Figure 5 shows the annual cycle of precipitation and T2m averaged over 81 stations and the corresponding results derived from ICON for the period October 2008 to September 2020, along with the results from the gridded reference datasets. Compared to the station data, the IMERG data tend to overestimate the precipitation in the wet season (June to September), but underestimate the precipitation in the cold season (December–January–February, DJF) (Fig. 5a). ICON exhibits a reasonably good performance in simulating the annual cycles of precipitation over the central and eastern TP (Fig. 5a). Compared to the station data, an overall dry bias is seen in ICON, with the largest dry bias of 0.34 mm d⁻¹ in May–June (20% of the observed climatological precipitation in May, but 11% in June).

Figure 5.  The 11-year (October 2009 to September 2020) average monthly (a) precipitation (units: mm d⁻¹) and (b) T2m (units: °C) averaged over 81 stations over the TP (see blue dots in Fig. 1b), derived from rain gauge observations (blue bars) and the corresponding ICON results (red bars). Black bars in (a) represent the corresponding precipitation derived from IMERG, and in (b) the T2m derived from GLDAS. The gridded precipitation and T2m data were interpolated to the locations of the rain gauge stations using nearest neighbor interpolation.

For T2m, compared to the station data, GLDAS tends to produce colder T2m over the central and eastern TP (Fig. 5b). The annual cycles of T2m revealed by the station data over the TP are reproduced well in ICON (Fig. 5b). Cold biases are found in the model, with the largest cold biases of −1.1°C in August (8.3% of the observed climatology) in comparison to the station data. The performance of ICON is much better than any individual CMIP6 model, which generally exhibit very large wet biases and severe cold biases over the central and eastern TP (e.g. Zhu and Yang, 2020).

The spatial patterns of the simulated seasonal mean precipitation averaged for the period October 2009 to September 2020 are compared against both the station data and the IMERG data in Fig. 6. Precipitation over the TP exhibits significant seasonal variations. In DJF, affected by westerlies, the precipitation center in the IMERG data is located in the Amu Dayra, Indus, and Ganges basins (Fig. 6a). In the summer (June–July–August, JJA) season, affected by the Indian summer monsoon, the precipitation center is observed over the southern slopes of the Himalayan Mountains and major river source regions (Yarlung Zangbo, Nu, Lancang, Yangtze, and Yellow basins) (Fig. 6e). In the transition seasons [March–April–May (MAM) and September–October–November (SON)], precipitation is found in the eastern Yangtze and Yellow basins and over the windward slopes of the Hengduan Mountains located in the southeastern TP (Figs. 6c and g). These spatial patterns and the seasonal variations in precipitation revealed by the IMERG data are reasonably reproduced by ICON (second column of Fig. 6), as demonstrated by SCCs of 0.77, 0.72, 0.74, and 0.58 for DJF, MAM, JJA, and SON (Table 2), respectively. Compared to the IMERG data, finer spatial details are evident in the ICON model because of the higher horizontal resolution (second column of Fig. 6).

Figure 6.  Spatial distributions of 11-year seasonal mean precipitation (units: mm d⁻¹) for (a, b) DJF, (c, d) MAM, (e, f) JJA, and (g, h) SON derived from IMERG (first column) and ICON (second column). The third column shows the differences in the seasonal mean precipitation between IMERG and the ICON control simulation. Dashed areas are statistically significant at the 5% level, according to Student’s t-test. The results from ICON were interpolated to the 0.1° × 0.1° resolution of the IMERG data using areal conservative remapping to calculate the differences. The fourth column shows the differences (units: mm d⁻¹) in the seasonal mean precipitation between the station data and the ICON control simulation. The ICON model output at the points nearest to the stations was selected to calculate the differences. The TP is defined as regions with elevation higher than 2400 m.

The differences in the seasonal mean precipitation over the TP between the ICON simulation and the IMERG data are shown in the third column of Fig. 6. The RMSEs of the precipitation over the TP between the ICON and IMERG data are 1.66, 1.74, 1.98, and 1.24 mm d⁻¹ for DJF, MAM, JJA, and SON (Table 2), respectively. Spatially, compared to the IMERG data, ICON shows more precipitation over the Amu Dayra and western Indus basin across the seasons. Since in situ observations are rare and quite limited there, it is difficult to identify which is more accurate. In DJF, while the whole TP sees a wet bias (Fig. 6i), in JJA, the Himalaya, central TP, and Qaidam basin witness a dry bias (Fig. 6k). Further sensitivity experiments are required to improve the simulation of JJA precipitation in ICON. The patterns of simulated precipitation biases in MAM (Fig. 6j) and SON (Fig. 6l) are similar to those in JJA, but the magnitudes are much smaller. The comparisons with IMERG data are quantitatively consistent with comparisons to the station data over the eastern TP (fourth column of Fig. 6).

The simulated seasonal mean T2m by ICON are shown in Fig. 7 with comparisons to the GLDAS T2m and the station data. In GLDAS, high T2m is observed over the Qaidam Basin and eastern TP, while low T2m is found over the western TP. This zonal contrast due to topographic effects is seen across the seasons, with the weakest being found in JJA (Fig. 7e). These general features are reproduced well by ICON (second column of Fig. 7). The simulation skill for T2m is higher than that of precipitation, as evidenced by much larger SCCs of 0.94, 0.95, 0.93, and 0.95 for DJF, MAM, JJA, and SON (Table 2), respectively. Again, ICON exhibits finer spatial details because of the much higher horizontal resolution employed (3.3 km vs 0.25°).

Figure 7.  As in Fig. 6 but for the surface air temperature (T2m; units: °C). The T2m derived from the GLDAS dataset is employed as the reference data in the third column.

As shown in the third column of Fig. 7, compared to the T2m from GLDAS, in JJA, ICON exhibits overall colder T2m over the entire TP (Fig. 7k), while in other seasons colder T2m is mainly found over the inner TP and warmer T2m is found in the eastern TP, including the Hexi Corridor and major river source basins. The largest warm biases over the southeastern TP are found in DJF (Fig. 7i). The RMSEs of the T2m over the TP between the ICON and GLDAS data are 2.45°C, 1.98°C, 2.06°C, and 1.85°C for DJF, MAM, JJA, and SON (Table 2), respectively. The comparisons of simulated T2m with the GLDAS T2m are supported by those with the station data over the eastern TP (fourth column of Fig. 7).

The vertical distributions of differences in seasonal mean precipitation and T2m over the TP between the ICON simulation and the gridded reference datasets (IMERG for precipitation, GLDAS for T2m) are shown in Fig. 8. The differences, which can be roughly regarded as the biases of the model, are binned with a 500 m size. For precipitation (Fig. 8a), below the elevation of 5000 m, the median value of biases is approximately 0.5–1.5 mm d⁻¹, which is generally uniform across different elevations and different seasons. Above the elevation of 5000 m, the characteristics of precipitation biases are season-dependent. In cold seasons (DJF and MAM), which are primarily affected by westerlies over the TP, larger wet biases are found at higher altitudes. This phenomenon is not pronounced in warm seasons (JJA and SON).

Figure 8.  Differences in seasonal mean (a) precipitation (units: mm d⁻¹) and (b) T2m (units: °C) between the ICON simulation and the observations (IMERG for precipitation, GLDAS for T2m) shown as a function of height over the TP. The results from ICON were interpolated to the resolutions of the reference datasets (0.1° × 0.1° of IMERG, 0.25° × 0.25° of GLDAS) using areal conservative remapping to calculate the differences. Differences were calculated in bins of 500 m in size, and the values are shown in the middle of each bin. The number of samples in each bin is shown on the right-hand y-axis. The dashed lines represent one standard deviation of the spread of the differences in each bin.

For T2m, below the elevation of 4000 m, warm biases are dominant across different seasons, with the largest (lowest) in DJF (JJA). With increasing elevation, the warm biases tend to be reduced, and cold biases are found above 5500 m across different seasons. The largest cold biases (the median value of biases is up to −2.2°C) are found in JJA above 5500 m. Large cold biases at high elevation over the TP are also seen in other kilometer-scale modeling (Prein et al., 2023).

Note that the gridded reference datasets over the TP region have large uncertainties (e.g., Prein et al., 2023). GLDAS assimilates ground-based and satellite observational data, while the precipitation derived from IMERG needs calibration using station observations. Usually, stations over the TP are located in valleys, and thus can underestimate the precipitation over the nearby mountains and lead to cold biases for temperature. The lack of gauge undercatch correction over the high mountains may also lead to uncertainties (e.g. Fu et al., 2023).

In summary, these preliminary evaluations of the seasonal mean precipitation and T2m suggest that ICON, with its convection-permitting resolution, exhibits overall good performance in simulating the present-day climate over the TP region, albeit with some biases still existing, i.e., cold biases in JJA and wet biases in DJF at high elevation. The simulated biases in ICON are much less than those in the CMIP6 models (Zhu and Yang, 2020). In addition, ICON’s km-scale resolution provides more spatial detail of the surface climate over the TP than the observational datasets. These results gave us confidence in the projected changes in surface climate over the TP with ICON as reported in the following section.

In this section, focusing on the seasonal mean precipitation and T2m over the TP, the projected spatial changes by ICON PGW and the associated elevation-dependent characteristics are presented with comparison to those from a selected 18 CMIP6 model ensemble mean.

The 11-year mean seasonal precipitation differences over the TP between the ICON control and PGW simulations are shown in Fig. 9, along with the projected precipitation changes under the SSP5-8.5 scenario derived from the ensemble mean of the 18 CMIP6 models. Widespread wetting over the TP is projected across different seasons in the ICON PGW simulation (left column of Fig. 9), with relatively greater wetting being found in JJA (Fig. 9c) and SON (Fig. 9d). In JJA and SON, the average wetting over the TP is 0.35 mm d⁻¹, corresponding to a 10% and 26% increase of the present-day climatology, respectively.

Figure 9.  (a–d) Differences in seasonal mean precipitation (units: mm d⁻¹) over the TP between the ICON PGW simulation and the ICON control simulation in (a) DJF, (b) MAM, (c) JJA, and (d) SON. (e–h) Spatial patterns of the projected changes in seasonal mean precipitation (units: mm d⁻¹) over the TP derived from the 18 CMIP6 multimodel ensemble under the SSP5-8.5 scenario for the period of 2081–2100 relative to 1995–2014 for (e) DJF, (f) MAM, (g) JJA, and (h) SON. Dashed areas are statistically significant at the 5% level, according to Student’s t-test.

These general features of projected precipitation changes by ICON PGW are comparable to those from the CMIP6 ensemble mean (right column of Fig. 9). However, significant differences between them are evident: (1) In DJF (Fig. 9a) and MAM (Fig. 9b), large increases (up to 4 mm d⁻¹) in precipitation are projected over the Karakoram Mountains and the surrounding regions in ICON PGW, while the corresponding projected increases are quite uniform in the CMIP6 ensemble (Fig. 9e and Fig. 9f). This may be an indication of the added value of the kilometer-scale resolution over high elevations. (2) In JJA, the projected precipitation by CMIP6 increases over most areas of the TP (Fig. 9g), with the maximum increase over the southern TP, except over the northwestern TP, where slight drying is found (Fig. 9g). Compared to CIMP6, the projected increases in precipitation by ICON PGW over the southern TP are much weaker (Fig. 9c). The greater increase in CMIP6 is because the Himalaya are not high enough in the GCMs, and thus more water vapor is transported to the northern slopes of the Himalaya, which leads to a larger increase of summer precipitation (e.g., Gao et al., 2008). Over the eastern Himalaya and eastern marginal region of the TP, mixed precipitation change signals are observed in ICON PGW (Fig. 9c). This pattern is consistent with previous studies based on dynamical downscaling with RCMs over the TP or South Asia (Choudhary and Dimri, 2018; Gao et al., 2018; Ma et al., 2023), which has been attributed to the well-resolved land surface heterogeneity that is missing in CMIP6 with coarse resolution (Gao et al., 2018).

Figure 10 shows the spatial patterns of projected seasonal mean percentage changes in precipitation from ICON PGW and the CMIP6 models. As seen in the left-hand column of Fig. 10, ICON PGW projects a more than 100% increase in precipitation, particularly over the inner TP and the Qaidam basin during the cold and dry seasons (DJF, SON) (Fig. 10a and Fig. 10d). While the broad spatial features of projected percentage changes in precipitation in ICON PGW are similar to those from the CMIP6 ensembles (right-hand column of Fig. 10), particularly during JJA and SON, significant differences are evident between them: (1) ICON PGW exhibits more regional detail due to its kilometer-scale resolution. In contrast, the CMIP6 ensemble projects more uniform percentage changes in precipitation, with median increases of approximately 20%–30% across the different seasons. (2) Compared to CMIP6, ICON PGW exhibits larger percentage increases in projected precipitation changes, particularly during cold and dry seasons (Fig. 10). Note that the CMIP6 models significantly overestimate the precipitation over the TP throughout the year (Zhu and Yang, 2020). Conversely, ICON, with it kilometer-scale resolution, tends to mitigate this overestimation. This suggests that any projected increase in precipitation by ICON PGW will have a larger impact in percentage terms compared to CMIP6. Nevertheless, further in-depth investigations are necessary to understand the underlying mechanisms responsible for the differences in projected precipitation changes between ICON PGW and the CMIP6 models.

Figure 10.  (a–d) Percentage differences in the seasonal mean precipitation over the TP between the ICON PGW simulation and the ICON control simulation in (a) DJF, (b) MAM, (c) JJA, and (d) SON. (e–h) Spatial patterns of the projected seasonal mean percentage changes in precipitation over the TP derived from the 18 CMIP6 multimodel ensembles under the SSP5-8.5 scenario for the period of 2081–2100 relative to 1995–2014 for (e) DJF, (f) MAM, (g) JJA, and (h) SON. Dashed areas are statistically significant at the 5% level, according to Student’s t-test. Note the difference in the color scale between the left and right column.

The 11-year mean seasonal T2m differences over the TP between the ICON control and PGW simulations are shown in the left-hand column of Fig. 11. Substantial warming over the TP is projected across different seasons with geographical variations (Fig. 11). The average projected warming over the TP is largest in SON (5.48°C), while it is relatively low in MAM and JJA (4.89°C–5.03°C). Spatially, comparatively weaker warming is found over the southeastern TP in each season, while stronger warming is found in high-latitude regions, particularly the Qaidam Basin, and in high-elevation regions over the central-western TP. These spatial patterns in projected T2m changes over the TP are generally consistent with those projected by RCMs with 25 km resolution driven by CMIP5 models under the RCP 8.5 scenario (Niu et al., 2021b).

Figure 11.  (a–d) Spatial patterns of the projected differences (units: °C) in the 11-year mean (October 2009 to September 2020) T2m over the TP between the ICON PGW simulation and the ICON control simulation for (a) DJF, (b) MAM, (c) JJA, and (d) SON. (e–h) Spatial patterns of the projected changes in T2m (units: °C) over the TP derived from the 18 CMIP6 multimodel ensemble mean under the SSP5-8.5 scenario for the period of 2081–2100 relative to 1995–2014 for (e) DJF, (f) MAM, (g) JJA, and (h) SON. All the results are statistically significant at the 5% level, according to Student’s t-test. The regional mean T2m differences (units: °C) over the TP are shown in the bottom-left corner of each plot.

The projected seasonal mean T2m changes by ICON are compared to those derived from the CMIP6 ensemble mean (right-hand column of Fig. 11). Both the magnitude and spatial pattern of projected T2m changes in ICON are comparable to the CMIP6 ensemble temperature changes, showing overall greater warming in high-latitude and high-elevation regions, in all seasons except DJF. The differences between them include the following: (1) ICON PGW produces much finer regional details due to the kilometer-scale resolution. For example, ICON tends to project stronger warming in the Qaidam Basin than in its surrounding regions, which is missing in CMIP6. (2) ICON produces overall weaker warming over the TP than the CMIP6 models, especially in the DJF season. In DJF, up to 7°C of projected warming is found over the central TP in CMIP6 models (Fig. 11e), while in most areas of the central TP, less than 6°C of warming is projected in ICON PGW (Fig. 11a). Lower projected surface warming in RCMs than those from the driving global models were also reported by studies in China (e.g., Gao et al., 2012). The mechanisms are complex, which needs further in-depth analysis.

Elevation-dependent warming (EDW) is an important feature in the high mountain areas of the world (e.g., Pepin et al., 2015). Previous studies based on CMIP models (Zhu and Fan, 2022) and RCMs with coarse resolutions (Niu et al., 2021b) suggest that EDW over the TP will persist in the future. Here, we compare the height dependence of the seasonal mean projected T2m warming over the TP between ICON PGW and CMIP6 in Fig. 12. To fairly compare to the CMIP6 model results, the projected changes from ICON PGW were interpolated into the 1° × 1° grids of the CMIP6 multimodel ensemble mean using areal conservative remapping. There were 297 grids in total over the TP at a resolution of 1° × 1°.

Figure 12.  Projected changes (units: °C) in T2m derived from the ICON PGW simulation and the 18 CIMP6 multimodel ensemble mean shown as a function of height over the TP. The projected changes from the ICON PGW simulation were interpolated into the 1° × 1° grids of the CMIP6 multimodel ensemble mean using areal conservative remapping. Projected changes were calculated in bins of 500 m in size, and the values are shown in the middle of each bin. The number of samples in each bin is shown on the right-hand y-axis. The dashed lines represent one standard deviation of the spread of the changes in each bin.

As shown in Fig. 12, an evident EDW signal over the TP is projected under the SSP5-8.5 scenario in both ICON PGW and the CMIP6 ensemble in all seasons except JJA. The projected EDW is rather weak in JJA (Fig. 12c), which is also seen in observational evidence for the present-day climate (Guo et al., 2019) due to a stronger vertical mixing in the warm season than other seasons. The differences between ICON PGW and the CMIP6 ensemble mean include the following: (1) The projected EDW signal is stronger in the CMIP6 ensemble than in the ICON PGW simulation, which is mainly contributed by the enhanced projected warming at high elevations above 5000 m. The differences may be associated with the snow-albedo feedback. (2) ICON PGW tends to project minimum warming at approximately 3500–4000 m, while in the CMIP6 models the projected warming monotonically increases from low elevation to high elevation. The non-monotonic projected warming from low elevation to high elevation is supported by the observed features of EDW over the TP from 1961 to 2012 (Pepin et al., 2015), highlighting the added value of kilometer-scale projection.

In this study, two 12-year convection-permitting (3.3 km) simulations, with a focus on the TP region, were conducted using ICON in the limited area mode. The simulations included a retrospective simulation, which dynamically downscaled the ERA5 reanalysis data for the period September 2008 to September 2020, and a PGW simulation for the same period driven by the modified ERA5 reanalysis data at the boundaries by adding the monthly CMIP6 ensemble-mean climate change under the SSP5-8.5 scenario. This overview paper introduces the experimental design of these decades-long simulations and presents a preliminary analysis of the evaluation and projection of seasonal mean precipitation and surface air temperature. The major results can be summarized as follows:

(1) ICON captures well the spatial pattern of seasonal precipitation and T2m over the TP and the corresponding seasonal variations in the present-day climate. Compared to the data from 81 meteorological stations mainly located over the central and eastern TP, the average biases in precipitation (temperature) are less than −0.34 mm d⁻¹ (less than −1.1°C) throughout the year.

(2) The simulated biases over the TP by ICON are season- and height-dependent. Wet biases are dominant in DJF, while dry biases are found over the Himalaya, central TP, and Qaidam Basin in JJA. More precipitation is simulated over the northwestern TP across the seasons. For T2m, overall cold T2m biases are dominant over the TP in JJA, while warmer biases are significant in the eastern TP in other seasons. Above 5500 m, cold biases (up to a median value of −2.2°C) in JJA and wet biases (up to a median value of +1.7 mm d⁻¹) in DJF are evident.

(3) ICON PGW projects substantial warming over the TP, with the average warming ranging from 4.89°C to 5.48°C. Comparatively weaker warming is found over the southeastern TP, while stronger warming is found in high-latitude regions, particularly the Qaidam Basin, and in high-elevation regions over the central-western TP. Both the magnitude and spatial pattern of projected T2m changes in ICON are comparable to the CMIP6 ensemble temperature changes, but overall greater warming is projected in CMIP6, especially in the high-elevation region, which leads to a stronger EDW signal.

(4) Widespread wetting is projected over the TP in ICON PGW, with relatively larger wetting being found in JJA and SON. The general features are comparable to those from the CMIP6 ensemble mean. However, the added value from kilometer-scale modeling is evident in the precipitation projection. For example, in DJF and MAM, large absolute increases in precipitation are projected over the Karakoram Mountains and the surrounding regions in ICON PGW. In contrast, with the large and uniform increase in precipitation over the southern TP in CMIP6, weaker and mixed precipitation change signals are observed in ICON PGW over the eastern Himalaya and eastern marginal region of the TP. Compared to CMIP6, larger percentage increases in projected precipitation changes are found in ICON PGW, particularly over the inner TP and the Qaidam basin during cold and dry seasons.

In this overview paper, we present a preliminary analysis of present-day and future climates using these kilometer-scale simulations with a focus on the TP region. The in-depth analysis is ongoing and includes the added value in simulating the water cycle, the diurnal cycle of precipitation, precipitation extremes over the TP, the PGW-projected water cycle changes, extreme precipitation changes, and mesoscale convective system changes. The differences in the projected changes in surface climate over the TP between ICON PGW and CMIP6 ensembles deserve further investigation.

In addition, the CORDEX CPTP FPS is now advancing to Phase II, which will coordinate decades-long convection-permitting simulations over the TP using multiple RCMs. The ICON-downscaled high-resolution simulation datasets provide an opportunity for the CORDEX CPTP FPS community to investigate one possible scenario of local climate change and impacts over the TP.

Acknowledgements. This work was jointly supported by the National Key Research and Development Program of China (Grant No. 2022YFF0802004), the National Natural Science Foundation of China (Grant Nos. 41988101 and 42275182), the K.C. Wang Education Foundation (Grant No. GJTD-2019-05), the Jiangsu Collaborative Innovation Center for Climate Change, and the National Key Scientific and Technological Infrastructure project “Earth System Science Numerical Simulator Facility” (EarthLab). This is a contribution no 18 to CORDEX-FPS-CPTP. Special thanks to the German Weather Service and the Max Planck Institute for Meteorology, Germany, for developing and providing ICON. The CMIP6 multimodel simulations are publicly available from the Earth System Grid Federation (https://esgf-node.llnl.gov/search/cmip6/). The GPM IMERG data can be obtained from NASA (https:// gpm.nasa.gov/data/imerg). GLDAS T2m data can be downloaded from https://ldas.gsfc.nasa.gov/gldas/model-output. Precipitation and surface air temperature data from national ground stations over the TP can be downloaded from http://data.cma.cn/en with registration. The monthly mean precipitation and surface air temperature data over the TP derived from the ICON present-day simulations and PGW projections used in this paper have been archived in figshare (https://doi.org/10.6084/m9.figshare.24316621). The total output volume for the ICON simulations is about 150 TB, and is stored at the local cluster of LASG.