The future extreme precipitation systems of orographically locked diurnal convection: the benefits of using large-eddy simulation ensembles

The precipitation hotspot of the orographically locked convection highly depends on the interactions among physical processes governing local energetics and cloud dynamics. Accurately estimating the future change of these hotspots will require a model with sufficient spatial resolution as well as an appropriate representation of the critical physical processes. In this study, ensembles of TaiwanVVM large-eddy simulations (Δx = 500 m) were designed to capture the summertime diurnal convection in Taiwan when local circulation dominates. The precipitation hotspots identified by long-term observations are well represented by the present-day ensemble simulations with appropriate environment variabilities. A pseudo global warming experiment is carried out to identify changes in convective structures, which results in local rainfall changes. Under the scenario of 3 K uniform warming with conserved relative humidity, the changes in the thermodynamic environment feature an overall higher convective available potential energy and a small decrease in convective inhibition (CIN), owing to the marked increase in low-level water vapor in the marine boundary layer. The results show that mean precipitation and the occurrence of extreme convective systems (ECSs) increase, with hotspots over mountains expanding toward the foothills and plains. The response in cloud dynamics leads to more short-duration, intense rainfall events. The tracking of ECSs with maximum rainfall exceeding 100 mm h−1 reveals more numerous short-lived ECSs (lifetime <6 h) and the enhancements in maximum updrafts by ∼10 m s−1, in cloud top heights by ∼1 km, and in the volume of cloud objects by ∼1.5 folds. These sets of high-resolution simulations under the specific weather regime offer critical information for assessing the potential impacts of the future changes of extreme rainfall contributed by the orographically locked diurnal convection on natural disasters and water resources.


Introduction
Diurnal convection is the dominant and fundamental mode of precipitation variability over islands and lands in the tropics and summer monsoon regions (Yang andSlingo 2001, Nesbitt andZipser 2003).It significantly contributes to the climatology of convection, cloudiness, and surface temperature, which are integral components of tropical weather systems.These convective processes typically peak during the late afternoon and evening over land, indicating sensitivity to the diurnal heating of the atmosphere and land surfaces, and can extend their influence over adjacent ocean areas through gravity waves and other dynamic interactions such as mountain blocking effects (Yang and Slingo 2001).Furthermore, organized and intense convective systems, with their capacity to generate extreme rainfall events, can be regularly formed in the diurnal cycle, particularly over the mountains and coastlines where topographic forcing facilitates the organization processes (e.g.Romatschke and Houze 2011, Chen et al 2015, Song and Zhang 2020, Krishna et al 2021).The intensity and organization of tropical diurnal convection are influenced by large-scale atmospheric dynamics, This study introduces the first application of a framework with ensembles of large eddy simulations (LESs) to assess how global warming affects the summertime orographically locked diurnal convection in Taiwan, a subtropical island with strenuous terrain located in the East Asian Monsoon.While the orographic locking effects constrain the variability of diurnal convection, the critical tool for this framework is the TaiwanVVM (Vector Vorticity Equation Model with realistic topography of Taiwan Island; Wu et al 2019), a model capable of appropriately resolving and representing the physical feedback of convection development on the complex terrain with a 500 m horizontal resolution (detailed in section 2.1).Previous studies have demonstrated that this TaiwanVVM framework can effectively disentangle aerosol-cloud interactions (Chang et al 2021) and the deep inflow structure of orographically locked convection (Chang et al 2023).Adopting the storyline approach, the large-scale regime of focus here is when the island is surrounded by the subtropical high-pressure system.The condition allows the local circulation to fully develop and dominate the evolution of diurnal convection, while the moist marine boundary layer results in convective instability that promotes deep and intense convection (detailed in section 2.2).The ensemble simulations of the current condition are designed to span the realistic variability of the diurnal convection under the chosen large-scale regime.The major objective here is to identify how the local circulation and cloud dynamics will respond to the warming climate in a model with well-resolved physics.The PGW strategy is hence applied to obtain the future ensemble simulations by applying thermodynamic forcing simplified from a GCM future projection.The manuscript is organized as follows.Section 2 describes the TaiwanVVM and the design of the ensemble experiments.Section 3 presents the responses of mean precipitation and the extreme convective systems (ECSs) to warming based on statistics of the convection-tracking algorithm.The discussion and conclusion are provided in sections 4 and 5, respectively.

Description of TaiwanVVM
The model used in this research is the vector vorticity equation cloud-resolving model (VVM; Jung and Arakawa 2008, Wu and Arakawa 2011, Chien and Wu 2016) with realistic Taiwan topography and land-use types at 500 m horizontal resolution (TaiwanVVM; Wu et al 2019).In the vector vorticity formulation in height coordinate, the vorticity field can directly respond to horizontal buoyancy gradients, and the wind fields are diagnosed by solving the elliptic equation of vertical velocity.The model top is a rigid lid with gravity wave damping.The lower boundary over the terrain is treated as block mountains (Wu et al 2011, Chien andWu 2016) with imposed vorticity at the corner to satisfy kinematic boundary conditions.The vorticity formulation, together with the coupling to the Noah land surface model (Noah LSM v 3.4.1;Chen et al 1996, Chen andDudhia 2001), makes the model particularly advantageous for capturing the local circulation generated by differential surface heating, especially over complex, steep topography, as compared to the model with terrain-following vertical coordinate.Previous studies have applied the VVM and TaiwanVVM to investigate the phenomena that are dominated by local circulation and boundary layer processes, such as diurnal convection over complex topography (Kuo and Wu 2019, Wu et al 2019, Chang et al 2021, 2023), and air pollutant transport (Hsieh et al 2022, Hsu et al 2023).In this study, the model operates at a horizontal resolution of 500 m, covering a domain size of 512 km by 512 km with doubly periodic lateral boundaries.The default timestep is 10 s but can be dynamically adapted during integration to satisfy the Courant-Friedrichs-Lewy criterion.The topography of Taiwan Island is placed at the center of the domain, surrounded by ocean.The vertical layers extend from sea level to 3900 m with intervals of 100 m and have stretched grids reaching approximately 19 260 m.With the predicted particle properties microphysics scheme (P3; Morrison and Milbrandt 2015), the large eddies generated by the convective-scale dynamics and the turbulent flows of the local circulation developed over the strenuous terrain are explicitly simulated and well resolved, as demonstrated in figure 1 by the example of a deep convective system over the southwestern slope of Taiwan in one of the TaiwanVVM ensemble members in this study.The sub-grid turbulence parameterization (Shutts and Gray 1994) only involves the stability and deformation-dependent mixing treatment to the eddy viscosity and diffusivity coefficients, allowing the resolved vorticities to interact with other physical processes.

Experiment design: current and future ensembles 2.2.1. Current ensemble
The storyline of this study focuses on the weather regime under which the development of local circulation dominates the evolution of afternoon thunderstorms during summertime in Taiwan.As shown in the composite in figure 2 based on 198 cases selected from May to September 2005-2015, the synoptic configuration generally features that the western Pacific subtropical high extends westward over Taiwan, which is usually signified by the 5880 m contour of the 500 hPa geopotential height.The region surrounding  Taiwan exhibits weak low-level horizontal winds, mid-level subsidence, and weak vertical wind shear conditions.The column-integrated column water vapor varies around 45-55 mm, as the oceanic boundary layer remains highly moist, while the mid-level moisture can vary with the intensity of the subsidence.In such an environment, the temperature, moisture, and vertical wind shear changes are the leading factors modulating the convective variability of the precipitation hotspots, while the background flow pattern plays a secondary role.
Therefore, to capture both the key features and the range of variability of the thermodynamic structure, we have selected 30 observational soundings from the targeted weather regime from May to September 2005-2015 at Banqiao Station in northern Taiwan (Chang et al 2021) to carry out the LESs of the 'Current' ensemble under the present-day scenario, as shown in figure 3(a).Vertical profiles of temperature, moisture, and horizontal winds are extracted from each of these soundings and applied uniformly over the entire domain as initial conditions of the TaiwanVVM simulations.This simulation configuration is commonly employed in LES to emphasize the interactions among physical processes governing the evolution of local circulation and convection and to explore the sensitivity of these processes to critical environmental factors (e.g.Grabowski et al 2006).The blue bars in figure 3(b) demonstrate that the sounding profiles share similar features of the convective instability environment: moderate to high convective available potential energy (CAPE), the presence of convective inhibition (CIN), and high column-integrated water vapor (CWV).Owing to the strong evaporation of the surrounding ocean, the moisture conditions between the surface and 850 hPa are high and less variable (24-26 mm), while the mid-level moisture, modulated by the subsidence of the sub-tropical high, can be more variable (20-27 mm) and determines the overall variability of the total CWV.Noah LSM is initialized by assigning the daily averaged soil moisture over the island of Taiwan from the Global Land Data Assimilation System (GLDAS; Rodell et al 2004) version 2.0 to the topsoil layers for all land grids in the model.
The 24 h simulations are started from midnight (00LST) to generate the temperature structure corresponding to the nocturnal cooling processes while preserving the large-scale thermodynamic structure from the sounding.After sunrise, the land-sea temperature contrast is explicitly generated by the local shortwave radiative heating processes.The domain-averaged winds are relaxed to the initial wind throughout the integration.As the boundary layer turbulence is explicitly simulated by the vorticity equation, the convection is directly triggered by the horizontal inhomogeneity of buoyancy, which can be generated by processes such as differential heating over terrain, sea breeze fronts, or cold pools.

Future ensemble (PGW)
In order to cleanly attribute the change in precipitation and local circulation over the complex topography to the first-order structure in thermodynamic forcing and the modulation in cloud dynamics, we design the 'Future' ensemble experiment using the PGW approach.The warming scenario is based on the historical simulation and the future climate projections of Taiwan Earth System Model version 1 (TaiESM1; Lee et al 2020).The temperature and moisture profiles averaged around Taiwan are examined, and the averaged vertical temperature profile is projected to be 3 K warmer than the historical simulation sometime between the near future and the far future, with the timing depending on the anthropogenic emission scenarios of the various Shared Socioeconomic Pathways.The warming is quite vertically uniform throughout the troposphere below 250 hPa, and the magnitude and structure of the warming are similar regardless of seasonal means or only on days with synoptic patterns favoring local circulation development.For moisture profiles, the TaiESM simulations are wetter in mid-and high levels than the Banqiao sounding observations, indicating a systematic wet bias in the model.However, the mean RH profiles in the near-future and far-future projections stay similar to the historical simulations.Therefore, for simplicity, in our pseudo-global warming experiments, we add a vertically uniform 3 K increment to the temperature profiles of the initial conditions of the Current ensemble and adjust the moisture profile by conserving the RH for each member, respectively.The initial wind profile remains the same as that in the corresponding Current ensemble member.With such an experiment design, the land-sea temperature contrast is maintained in the Future ensemble simulations, leading to similar sea breeze intensity and development in the simulated diurnal cycle.
We can first examine the changes in the thermodynamic environment after applying the temperature and water vapor increment to the 30 initial conditions.The changes in ensemble mean profiles are shown in figure 3(a) (dashed lines), and the statistical distribution of thermodynamic parameters is shown in figure 3(b).Owing to the marine layer environment, the low-level CWV is greatly enhanced after warming with the constant RHs, leading to a moderate decrease in CIN and the level of free convection (LFC), while there is a significant increase in CAPE due to warming effects.The overall convective environment in the 3 K warming scenario suggests enhanced convective intensity as higher energy can be acquired for air parcels that break through the CIN layer and a higher probability of deep convection initiation as energy limits and height constraints on the spontaneous rising of air parcels will become lower.Therefore, more frequent and stronger development of deep convection is expected in the future ensemble, while the high-resolution simulations will quantify the detailed changes in convection intensity and location with explicitly resolved cloud dynamic response to the thermodynamic forcing.

Ensemble mean precipitation
The Central Weather Administration (CWA) in Taiwan operates a spatially dense and temporally continuous surface rain gauge network covering the entire island.Figure 4 Figure 4(c) shows the ensemble mean precipitation in the future scenario simulations.The overall ensemble areal mean precipitation increases from 4.8 mm d −1 to 7.1 mm d −1 (∼16% K −1 ).The precipitation enhancement is far higher than the average increase in the topical mountainous region, indicating strong feedback involving convective-scale dynamics (Fowler et al 2021) under the selected local circulation-dominant weather regime.As for the spatial distribution, the precipitation hotspots remain in similar locations with enhanced intensity under the future scenario, and the rainfall region expands toward the western plain.For the diurnal evolution, the Future ensemble's mean peak time shifts earlier by 30 min while the amplitude is enhanced by 35% relative to the Current ensemble.There are also more cases exhibiting rainfall in the late evening, leading to longer durations of the diurnal cycle.The differences between the Future and Current ensemble mean, as shown in figure 5(a), reveal a more distinct structure of the precipitation response to the warming.The areas that exhibit significant precipitation enhancement (>6 mm d −1 ) are located over the western slopes and foothills adjacent to the current hotspots towards the plains, with a maximum increase of 18.7 mm (∼128%) in mean precipitation.There are scattered areas with rainfall reduction (−2 to −14 mm d −1 ), mainly situated around the elevation of around 1000 m, the altitudes between the current hotspots and the higher ridges.
The pattern of precipitation change signifies clear modulation in cloud dynamics in the future ensemble.The shift of the mountainous precipitation hotspots towards the foothill and plain can be related to more pronounced propagation of the orographically locked convective systems or more frequent convection triggering and development over the western foothill and plain.This can be further illustrated when compared to the estimated thermodynamic precipitation response using the C.-C. scaling, as shown in figure 5(b).The estimate assumes that the RH and precipitation efficiency remain the same as in the Current simulations.Thus, for a given temperature change, the thermodynamic-only precipitation response would scale with the change rate in saturation vapor pressure described by the C.-C. equation (Trenberth et al 2003, Lenderink andvan Meijgaard 2008).The thermodynamic precipitation response is exclusively positive but much weaker in magnitude (mostly <6 mm d −1 and maximum increase ∼7.8 mm), and its spatial distribution follows the ensemble mean distribution that the orographically locked hotspots exhibit a more pronounced precipitation increase.The difference between the overall precipitation response and the thermodynamic precipitation response highlighted that cloud dynamics play a dominant role in determining the structure and intensity of future changes in the orographically locked afternoon thunderstorms in Taiwan.

Response of ECSs
In order to assess the response in cloud dynamics, the convective system tracking algorithm (Chang et al 2021) was applied to the LES output.The ECSs are identified as the convective systems exhibiting a maximum rain rate greater than 100 mm h −1 within the tracked lifecycles.The occurrence counts of the ECSs in figure 6 highlight the orographically locked nature of these extreme systems.Most of the ECSs occur over the precipitation hotspots in the southwestern mountains and foothills in both the Current and Future ensembles (figures 6(a) and (b)), while the counts are clearly enhanced in the Future ensemble.The maximum count increases from 52 in the Current ensemble to 86 in the Future ensemble.Using the contour  (2) the overlapping rain cells were temporally linked with the iterative consideration of merge and splitting; (3) the convective systems were identified by collocating the three-dimensional cloud objects (grids with hydrometeor mixing ratio > 10 −4 ) with cloud base lower than 0.5 km, cloud depth thicker than 1.0 km, and the center of cloud mass higher than 0.5 km on top of the tracked rain cells; (4) the convective systems that have maximum rain rate > 100 mm h −1 within the tracked lifecycles were selected as the ECSs.lines of ECSs occurrence over 40, figure 6(c) reveals the detailed spatial difference over the southwestern hotspots (red box in figure 5(a)).In the Current ensemble, more ECSs occur on the slopes between altitudes of 500 m and 1500 m.In the Future ensemble, the region experiencing more frequent ECSs expands toward lower elevations, but the boundaries at the higher-elevation side remain unchanged.The spatial change in the occurrence of ECSs is consistent with the change in ensemble mean precipitation (figure 5(a)), suggesting that ECSs predominantly contribute to the mean precipitation response.
Figure 7 quantifies the change in cloud dynamic features over the southwestern mountains and plains (red box in figure 6(a)) using the system-based statistics of the tracked ECSs.Overall, the system lifetime distribution (figure 7(a)) becomes wider in the Future scenario, with the interquartile range (IQR) expanding from 4.1-7.4h to 2.9-7.8h, featuring more frequent occurrence of ECSs with a shorter lifetime.The maximum rain rate (figure 7(b)) is moderately enhanced, accompanied by larger variability (23.3-29.5 mm h −1 in IQR).This suggests an increasing occurrence of the short-duration, intense precipitation events.In terms of the morphology and internal structure of the ECSs (figures 7(c)-(f)), there are significant increases in the median and variability of both cloud size (3.9 × 10 4 km 3 to 7.6 × 10 4 km 3 in median number; 4.9 × 10 4 km 3 to 8.6 × 10 4 km 3 in IQR) and in-cloud updraft velocity (31.6-41.6 m s −1 in median number; 7.2-13.0m s −1 in IQR).Together with higher median cloud top height (17.7 km to 18.5 km) and cloud water path (28.2-32.9kg m −2 ), the statistics show that both the organization, intensity, and variabilities of ECSs are enhanced, a clear response to the increase in CAPE and CWV in the initial conditions.
To relate the changes in system lifetime with convective structure, we can further divide the ECSs into short-lived and long-lived sub-groups based on the median number of lifetimes in figure 7(f) (6.0 h).The joint distribution of lifetime and cloud size for the two subgroups in figure 8(a) reveals that in the Current ensemble (blue boxes), the long-lived ECSs exhibit larger cloud size, suggesting more organized structures than the short-lived ECSs.In the Future ensemble, both sub-groups of ECSs grow in size (red boxes), but the growth of the short-lived ESCs is more significant, ranging from two to three times their original sizes.Moreover, many shorter-lived ECSs exhibit an even shorter duration (2-4 h) in the Future ensemble, while the lifetime distribution of long-lived ECSs remains.Figure 8(b) shows that the long-lived ECSs have stronger in-cloud updrafts, consistent with their higher degree of organization.The intensification of in-cloud updraft in the Future ensemble for short-lived ECSs is comparable to that for the long-lived ones.Overall, the statistics provided by tracking the extreme systems suggest that, under the given warming environment, more extreme systems can be effectively triggered over the southwestern region, and they develop faster and stronger and become more organized, whereas they also dissipate earlier, leading to a shorter lifetime.On the other hand, the more long-lived, organized type of ECSs is still present and intensified in the Future scenario.Together with the spatial distribution change of the ECSs (figure 6), we can anticipate that the long-lived ECSs remain to develop over the hotspots in the southwestern mountains in the Future ensemble.With their stronger intensity, these systems are likely to produce more pronounced cold pools, leading to longer propagation and/or more effective triggering of new systems towards the foothills and plains.The convectively more unstable environment in the Future ensemble also implies a higher probability of convective systems being triggered over the plain by sea breeze front or sea breeze confluence.The detailed changes in the simulated convection propagation and triggering in the PGW experiments will be investigated in subsequent studies.

Discussion
The present study focuses on the specific large-scale regime when the summertime subtropical high-pressure system dominates around Taiwan.We chose this environment to carry out the first set of the storyline approach simulations because it best demonstrates the advantage of the TaiwanVVM framework.The active development of local circulation over the complex terrain and the subsequent triggering and organization of ECSs are appropriately captured in the Current ensemble, owing to the sensitive response of the fine-scale circulation to the buoyancy gradient of the differential heating, as well as the appropriate representation the physical feedback modulating convective dynamics.However, the typical summertime synoptic conditions in Taiwan include not only the high-pressure dominant state investigated in this study but also the state with active moisture transport by the low-level southwesterly monsoon when the subtropical high weakens or shifts eastward.With the presence of moderate to strong low-level wind and vertical wind shear in highly Locations of precipitation hotspots and convection structures can become different, and hence their response to future warming.As our next step, the change of convection in Taiwan under various large-scale regimes will be systematically assessed using the TaiwanVVM framework.
Our PGW experiment imposed a first-order change in the thermodynamic environment based on the future projection of TaiESM.Under the spatially uniform 3 K warming with constant RH assumption, within the specified large-scale regime, the changes in thermodynamic profiles correspond to significant increases in CAPE and minor decreases in CIN due to a marked rise in low-level CWV.Our ongoing work will adopt the perturbation of various complexities in thermodynamic conditions in the PGW experiments (Brogli et al 2019(Brogli et al , 2023)).For example, differential surface warming between the island and the surrounding ocean can be applied to generate changes in the sea breeze intensity.Moreover, perturbation in temperature and moisture profiles that consider different temperature lapse rates and vertically varying RH conditions can represent future changes in the vertical structure of the subtropical high.The systematic exploration will reveal the sensitivity of the local circulation-dominated diurnal convection over complex topography to the variety of thermodynamic scenarios within the projected range of future regional climate change.
The value of the sub-kilometer LESs in this study is to provide critical data towards climate service regarding a specific regime of local weather under climate change.This is nicely visualized in figure 9 by overlaying the geographical information of frequent ECS occurrence under the future scenario with key water resource facilities and disaster-prone locations in the southwestern foothills.In the future, when the weather in Taiwan is under the influence of the sub-tropical high, the intensified and expanded coverage of afternoon extreme thunderstorms will provide more ample rainfall to supply several major reservoirs in southern Taiwan.This includes the Tsengwen Reservoir (also shown in the cross-section in figure 1), which holds the largest effective storage capacity on the island and plays an important role in water resource management.On the other hand, the area covered by enhanced ECSs also exhibits a high risk of natural hazards, as potential debris flow torrents and landslide locations are densely distributed along the strenuous terrain.The increasing occurrence of strong precipitation with short duration in the future can, therefore, trigger more numerous and severe geological disasters.Our approach provides the necessary and comprehensive guidance for impact assessment, land use development plans, and risk and resource management for a specific weather type in the future.

Summary and conclusion
The orographic-locked diurnal convection over complex topography offers both challenges and opportunities to investigate the key processes of convection and their interaction with the environment.The present study designs a framework that utilizes the TaiwanVVM LES ensembles and a carefully selected large-scale regime to assess how these localized, extreme-prone convective weather events in Taiwan Island will manifest in a changing climate.The results show that the precipitation hotspots and diurnal cycles identified by long-term observations are well represented by the present-day ensemble simulations.In the PGW simulations, Taiwan is projected to experience marked increases in CAPE alongside small decreases in CIN, provided the RH remains unchanged in the adjacent marine boundary layer.This leads to significant rainfall enhancement in the warming environment over the original precipitation hotspots, particularly along the western foothill, as well as earlier diurnal peak time and stronger amplitudes.There is a rise in the occurrence of ECSs in the warming scenario, especially in short-duration intense events.Tracking statistics reveal more numerous short-lived ECSs (2-4 h), and, overall, the ECSs exhibit more intense maximum updrafts (+10 m s −1 ), reach greater heights (+1 km), and have broader spatial extents (+1.5 times).The changes in spatial distribution and the different relationships between structure and lifetime suggest a modulation in the propagation and triggering location of the ECSs in the warming climate, which requires further examination.
In conclusion, this study has provided valuable insights into the impact of warming on orographically locked diurnal convection under the very specific synoptic regime that favors the development of local circulation in Taiwan.The design of the PGW experiments reveals the strong modulation in cloud dynamics that can enhance extreme rainfall.The sub-kilometer simulations using a model that well represents the key physics, given a highly constrained synoptic regime, offer critical information for local impact assessment and water resource management associated with future changes in extreme rainfall.Systematic simulations similar to the ones presented here but for different large-scale weather types and under various scenarios of future thermodynamic perturbation will be carried out in the subsequent work to provide a more complete investigation of the responses in the warm season convection in a changing climate.information of the landslide areas from the Geological Survey and Mining Management Agency, Ministry of Economic in Taiwan: https://landslide.geologycloud.tw/swagger/api-docs/api.
All data that support the findings of this study are included within the article (and any supplementary information files).

Figure 1 .
Figure 1.Vertical cross-section of a deep convective system (shadings) and local circulation (vectors) over the southwestern slope (along the AB red line in the upper-left inset) simulated in one of the TaiwanVVM ensemble members.The gray shadings show the cloud water content (ice plus liquid), and the color shadings represent the rainwater content.

Figure 2 .
Figure 2. Composite mean column-integrated water vapor (color shading), 850 hPa winds (vectors), and the contour of mean 500 hPa geopotential height equals 5880 gpm (orange line) from ERA5 reanalysis (Hersbach et al 2020) for the 198 local circulation-dominant days between the years 2005 and 2015 from May to September.

Figure 3 .
Figure 3. (a) The skew-T log-P diagram of the 30 initial conditions (thin lines) used in the Current simulations and the ensemble average (thick solid lines), and their wind directions (in degree-angle) and wind speeds (in radial distance; m s −1 ) at 500 hPa level and the low level (averaged between surface and 850 hPa).The brown lines represent the temperature, and the purple lines represent the dew point temperature, respectively.The thick dashed lines are the ensemble average of initial conditions in the Future simulations (see text for details).(b) The box-whisker plots of the CAPE, CIN, LCL, LFC, boundary layer CWV, and mid-level CWV of the initial conditions in the Current (blue) and Future (red) simulation ensembles.The dots represent the mean values.The bottom and top whiskers represent the range between the 10th and the 90th percentiles, and the boxes signify the 25th, 50th, and 75th percentiles.
(a)  shows the observed climatological rainfall distribution composited from the 198 d with environmental conditions favoring local circulation development from May to September 2005-2015.Major precipitation hotspots cluster in the central to southern part on the western slope of the central mountain range, while there are also some isolated hotspots in the northern mountains.The composite mean temporal evolution of precipitation shows a distinct diurnal cycle with peak time at ∼15LST and an amplitude of ∼1.1 mm h −1 .These features are accurately captured by the ensemble mean precipitation of the Current scenario simulations, as shown in figure4(b).The transition of precipitation intensity from the slope towards the western plain is also well represented in the simulations.The simulated composites reveal additional hotspots in the northwestern mountains where rainfall stations were not established owing to the strenuous terrain.The successful representation of the climatological precipitation pattern by the TaiwanVVM ensemble highlights the capability of the model to resolve the fine-scale structure of the local circulation and its effects on convection development over Taiwan's complex topography.The Current ensemble mean diurnal evolution of precipitation intensity closely matches the observation in both peak time and amplitude.The ensemble spread of the daily peak values captures the range of ±1 standard deviation of the observations.

Figure 4 .
Figure 4. (a) Composite mean precipitation intensity observed by 270 Central Weather Administration surface rain gauge stations during the 198 local circulation dominant days between the years 2005 and 2015 from May to September.The upper panel shows the spatial distribution of daily mean precipitation.The gray shading shows 500 m and 1000 m topographic contours, and the black lines represent the county and city boundaries.The lower panel shows the diurnal evolution of precipitation intensity averaged over all stations.The thick line shows the composite mean of all 198 d, and the thin lines are individual days.(b) The ensemble mean precipitation in the Current simulations.The upper panel shows the spatial distribution of daily mean precipitation.The lower panel shows the diurnal evolution of precipitation intensity averaged over the land.The thick line shows the ensemble mean, and the thin lines show individual members.(c) Similar to (b) but for the Future simulations.

Figure 5 .
Figure 5. (a) Difference in ensemble mean precipitation between the Future and Current simulations.The black contours mark the 1000 m elevation.(b) The estimated thermodynamic precipitation responses by the C.-C. scaling.The precipitation increments at each grid and at each time step in the Current simulations are scaled using the changes in initial near-surface temperature between each pair of the Current and the Future ensemble member.This ensures that the elevation effects on temperature and the initial temperature variability are taken into account (Berg et al 2013).

Figure 6 .
Figure 6.The occurrence counts of ECSs in (a) the Current simulations and (b) the Future simulations.(c) The regions where ECSs occur more than 40 times in the Current simulations (blue) and the Future simulations (red) over the hotspots in southwestern Taiwan (box in figure 6(a)).The gray shading shows the orographic heights.The ECSs are derived by the convective system tracking algorithm (Chang et al 2021): (1) rain cells were identified by the spatially contiguous rainy (>5 mm h −1 ) grids;(2) the overlapping rain cells were temporally linked with the iterative consideration of merge and splitting; (3) the convective systems were identified by collocating the three-dimensional cloud objects (grids with hydrometeor mixing ratio > 10 −4 ) with cloud base lower than 0.5 km, cloud depth thicker than 1.0 km, and the center of cloud mass higher than 0.5 km on top of the tracked rain cells; (4) the convective systems that have maximum rain rate > 100 mm h −1 within the tracked lifecycles were selected as the ECSs.

Figure 7 .
Figure 7. Box-whisker plots of the tracked extreme precipitation systems over the southwestern hotspots (red box in figure 6(a)) in the Current (blue boxes; 47 ECSs) and the Future (red boxes; 67 ESCs) simulations.The distribution of (a) the overall lifetime and the maximum values during the tracked lifetime for (b) rain rates, (c) cloud object volume, (d) in-cloud vertical velocity, (e) the cloud top height, and (f) the vertically-integrated cloud water path (liquid plus ice) the ECSs are displayed.The dots represent the mean values of all ESCs.The bottom and top whiskers represent the range between the 10th and the 90th percentiles, and the boxes signify the 25th, 50th, and 75th percentiles.

Figure 8 .
Figure 8.(a) Joint box-whisker plots of system lifetime and maximum cloud object volume and (b) system lifetime and maximum in-cloud updraft velocity.The ECSs are divided into short-lived and long-lived groups based on the median of the tracked lifetime in figure 7(f) (6 h for the Current ensemble and 5.3 h for the Future ensemble).The whiskers show the range between the 10th and 90th percentiles, crossing at the medians.

Figure 9 .
Figure 9.The locations of ECS occurrence exceeds 40 times in the TaiwanVVM Future simulations (red contour, same as figure 6(c)) and the geographic information of the key reservoirs (dark blue) and their catchments (light blue patches), the potential debris flow torrents (orange lines), and landslide areas (green patches) over the southwestern hotspots provided by the open data from Taiwan government agencies (see sources in the data availability statement).The gray shading shows the orographic heights.