High-resolution modelling of interactions between soil moisture and convective development in a mountain enclosed Tibetan Basin

The Tibetan Plateau plays a significant role in atmospheric circulation and the Asian monsoon system. Turbulent surface fluxes and the evolution of boundary-layer clouds to deep and moist convection provide a feedback system that modifies the plateau’s surface energy balance on scales that are currently unresolved in mesoscale models. This work analyses the land surface’s role and specifically the influence of soil moisture on the triggering of convection at a cross section of the Nam Co Lake basin, 150 km north of Lhasa using a cloud-resolving atmospheric model with a fully coupled surface. The modelled turbulent fluxes and development of convection compare reasonably well with the observed weather. The simulations span Bowen ratios of 0.5 to 2.5. It is found that convective development is the strongest at intermediate soil moisture. Dry cases with soils close to the permanent wilting point are moisture limited in convective development, while convection in wet soil moisture cases is limited by cloud cover reducing incoming solar radiation and sensible heat fluxes, which has a strong impact on the surface energy balance. This study also shows that local development of convection is an important mechanism for the upward transport of water vapour, which originates from the lake basin that can then be transported to dryer regions of the plateau. Both processes demonstrate the importance of soil moisture and surface–atmosphere interactions on the energy and hydrological cycles of the Tibetan Plateau.


Introduction
The Tibetan Plateau has an average elevation of more than 4500 m and is the world's largest mountain highland.Land use/land cover change, such as permafrost (Cheng and Wu, 2007) and grassland degradation or deforestation (Cui and Graf, 2009;Cui et al., 2006), but also effects on the carbon and hydrological cycles (Babel et al., 2014), are associated with overuse of resources and climate change.They also affect atmospheric circulation and hydrological resources of the Tibetan Plateau, especially the cryosphere (i.e.Yao et al., 2007;Ni, 2011;Kang et al., 2010;Yang et al., 2011Yang et al., , 2014)).Effects on the hydrological cycle are likely enhanced due to changes in phenology and length of the growing season (Shen et al., 2014;Che et al., 2014;Zhang et al., 2014).According to Immerzeel et al. (2010) more than 1.4 billion people in Southeast Asia live in the catchments of rivers originating on the Tibetan Plateau.It is therefore important to gain a better understanding of the hydrological processes in the region.
Surface-atmosphere interactions through the exchange of momentum, turbulent energy and water vapour play an important role in the development of convection and thus influence the energy balance of the surface and local precipitation.On the plateau scale, the surface energy balance impacts the plateau's role as an elevated heat source (i.e.Flohn, 1952; T. Gerken et al.: Soil moisture-convection interactions at Nam Co Lake Gao et al., 1981;Yanai et al., 1992).This heating is probably not the main driver of the Indian monsoon system (Molnar et al., 2010;Boos and Kuang, 2010), but acts to modify it and influences regional atmospheric circulation as well as eastern Asian precipitation (i.e.Liang and Wang, 1998;Zhou et al., 2004).While conventional atmospheric circulation models do not have a high enough resolution to resolve the relevant hydrological processes on the local scale, cloudresolving models with a fully coupled surface model, as introduced by Patton et al. (2005), present a valuable tool for the systematic investigation of surface-atmosphere interactions, convective development and locally generated precipitation.
The triggering of deep convection over topography is a major source of precipitation in mountainous environments (e.g.Banta and Barker Schaaf, 1987;Gochis et al., 2004).The summer monsoon season is associated with an increase of the total precipitable water in the atmosphere from approx.5 to > 15 mm and a destabilisation of the atmospheric profile (Taniguchi and Koike, 2008).A diurnal cycle in cloud development leading to the triggering of local convection before noon is commonly observed on the Tibetan Plateau and is organised by thermal valley circulations (i.e.Ueno, 1998;Ueno et al., 2009;Yatagai, 2001;Kuwagata et al., 2001;Kurosaki and Kimura, 2002).This organisation of local circulation leads to daytime precipitation over mountain ridges and to night-time precipitation from convergence in the basin centre (Ueno et al., 2009).Modelling studies (Kuwagata et al., 2001) and remote sensing (Yatagai, 2001) have determined valley scales of 160-240 km to be most effective for these processes.Yang et al. (2004) have also investigated thermal circulations on the Tibetan Plateau and found secondary triggering of convection within Tibetan valleys as a result of cold-pool fronts caused by convective events that were triggered over the mountains.Similar effects were simulated for Nam Co Lake as the result of a combined lake breeze-thermal mountain circulation (Gerken et al., 2014).Uyeda et al. (2001) measured maximum cloud top heights of 17 km a.s.l. in the Nagqu Basin occurring between 12:00 and 15:00 LT.Yamada and Uyeda (2006) found different structures of convection and precipitation on the Tibetan Plateau, which were associated with dry and moist surfaces.Dry surfaces during the pre-monsoon led to a high cloud base and increased re-evaporation of precipitation, while moist surfaces during the monsoon yielded lower cloud bases and higher rates of deposited precipitation.A modelling investigation (Yamada, 2008) confirmed this process and showed stronger convective motion for the dry case.Hence, complex interaction of local and larger scale circulation with local environmental conditions are essential for the evolution of clouds and the hydrological cycle on the Tibetan Plateau.
The influence of the monsoon declines to the North on the Tibetan Plateau (Tian et al., 2001(Tian et al., , 2003(Tian et al., , 2007)).At the same time local moisture recycling on the central plateau becomes essential keeping the regional water cycle active (Kurita and Yamada, 2008).Their research highlights the importance of the boundary layer as a major source of the precipitated water during night-time and of locally generated convection.Convection and precipitation at Nam Co Lake are both influenced by the lake's water body (Li et al., 2009) and the complex topography (Maussion et al., 2011(Maussion et al., , 2014)).
We previously investigated surface-atmosphere interactions, the development of mesoscale circulations and the evolution of convection in the Nam Co Lake basin.In Gerken et al. (2014), we applied a surface model (Gerken et al., 2012), coupled to the cloud-resolving ATHAM (Active Tracer High-resolution Atmospheric Model - Oberhuber et al., 1998;Herzog et al., 1998) and analysed convective triggering at Nam Co Lake.It was shown that the model can reproduce the most important atmospheric dynamics of the Nam Co Basin.Subsequently, the impact of different atmospheric profiles on convection and precipitation was investigated (Gerken et al., 2013a).These studies highlighted the importance of surface-atmosphere interactions in the generation of precipitation and convection.
Observed Bowen ratios at Nam Co station vary from 3 to 0.5, with maximum latent heat fluxes exceeding 400 W m −2 during wet conditions (Gerken et al., 2012).As a consequence Nam Co Lake can be classified as a semi-arid transitional region with strong soil moisture-climate coupling (Seneviratne et al., 2010).While the existence of feedbacks between land-surface moisture, rainfall and convective development is well known, there is disagreement on the dominating processes and the sign of the resulting feedback (Barthlott et al., 2011).Following the mechanism by Pal and Eltahir (2001), increased soil moisture increases the boundary layer's total convective available energy and lowers the lifting condensation level; both increase moist convection.At the same time, decreased solar radiation input due to larger cloud cover can reduce the available energy and thus convective development (Seneviratne et al., 2010;Pal and Eltahir, 2001;Pan et al., 1996).While soil moistureatmosphere interactions occur on many different spatial and temporal scales, interactions on the seasonal scale are well documented (e.g.Fischer et al., 2007).Taylor et al. (2012) showed that the likelihood of convective precipitation over wet soils was increased.Interactions on shorter timescales are less well understood.Clark et al. (2004) have demonstrated that moisture variability on the 10 km scale can induce regional circulations that enhance convection.Additional evidence for the influence of soil-moisture-induced mesoscale circulations on convective development was found in observational (Taylor et al., 2007) and modelling studies (e.g.Cheng and Cotton, 2004;Garcia-Carreras et al., 2010).Due to the localised convection triggering on the Tibetan Plateau, such processes might also play a role for precipitation at Nam Co Lake.
Compared to studies over flat terrain, much less is known about soil moisture-convection feedbacks above complex topography.Hohenegger et al. (2009) found a dependence on the model resolution in the Alpine region.Convection permitting resolutions led to predominantly negative feedbacks, caused by the presence of a stable layer above the boundary layer, while lower resolutions with parameterized convection showed positive feedbacks.Similarly, Barthlott et al. (2011) and Hauck et al. (2011) found considerable feedbacks between soil moisture and precipitation, but with a variable sign of the feedback.They identified convective inhibition as a major cause.
It is, therefore, key to understand soil moisture precipitation dynamics and its feedbacks at Nam Co Lake and similar basins to gain a better understanding of the hydrological cycle on the Tibetan Plateau scale.This work investigates the role of the interactions between the lake, the surrounding basin with complex topography and the atmosphere in the development of boundary-layer clouds and their evolution into deep moist convection.We analyse (1) the sensitivity of convective development to surface moisture, (2) the modification of the surface energy balance through feedbacks between surface fluxes, specifically evapotranspiration, and cloud development and (3) the implications for locally triggered, convective precipitation at Nam Co Lake.

Site description
Nam Co Lake, approx.elevation 4730 m a.s.l., is located circa 150 km north of Lhasa and just to the north of the Nyenchen Thangla mountain chain, which has maximum peak heights of more than 7000 m.The average elevation of the Nam Co Basin is 5230 m (Liu et al., 2010).Between 6 July and 6 August 2012 scientists from the University of Bayreuth conducted a field experiment at Nam Co Lake in cooperation with the Institute of Tibetan Plateau Research, Chinese Academy of Sciences.The Nam Co Lake Monitoring and Research Station for Multisphere Interactions (Ma et al., 2009) is located at 30 • 46.44 N, 90 • 57.72 E. It is situated at a distance of 300 m from a small lake, which is divided from Nam Co Lake to the north-west through a 500 m wide land bridge.The experiment is documented in Gerken et al. (2013b).Radiosondes were released on 8 days between 17 and 29 July for 00:00, 06:00 and 12:00 UTC.Eddycovariance flux measurements were performed with a CSAT-3 sonic anemometer and KH2O-krypton hygrometer (both Campbell Scientific).The KH2O was calibrated on site in accordance with Foken and Falke (2012).Standard atmospheric and soil temperature as well as moisture measurements were also carried out.
As described in Gerken et al. (2014), the TK2/3 software package (Mauder andFoken, 2004, 2011) was used to post-process the high-frequency measurements.All flux corrections and post-processing steps recommended in Foken et al. (2012) and Rebmann et al. (2012) were applied to  Foken, 2008;Foken et al., 2011).The presence of stationary, secondary circulations leads to a missing flux fraction of approx.30 % at Nam Co Lake.Therefore, before comparing measured and modelled fluxes, it is necessary to correct eddy-covariances fluxes according to the Bowen ratio (Twine et al., 2000).Additional information about flux dynamics at Nam Co Lake can be found in Biermann et al. (2014).
Vaisala RS92-SGP radiosondes on Totex-TA600 balloons were launched from a portable Vaisala sounding system with a mobile GPS antenna and signals were processed with the SPS-220 unit and DigiCora III MW21 software v3.2.1.This work uses mean local time throughout, which is close to UTC + 6 h or 2 h before Beijing Standard Time.On 17 July sunrise, solar noon and sunset were determined with the National Oceanic and Atmospheric Administration Sunrise/Sunset Calculator as 23:05, 06:02 and 13:00 UTC.
The soil was characterized in 2009 (Biermann et al., 2009) and the vegetation classification was reassessed in 2012 (Gerken et al., 2014, see Table 1).Except for some wetlands close to the lake, the basin and surroundings of the station are sparsely vegetated and classed as alpine meadow and steppe (Mügler et al., 2010).Figure 1 shows the Nam Co Lake basin and includes the flight path of the radiosonde launched for 17 July, 06:00 LT (00:00 UTC).The soil is estimated to have little water retention capacity and a low permanent wilting point (PWP) at 2 %.While the field capacity used for the modelling of 5 % is very small, it is not unreasonably low for a sandy soil and produced good agreement of measured and simulated fluxes in Gerken et al. (2012).

Model description
The non-hydrostatic, cloud-resolving ATHAM model (Oberhuber et al., 1998;Herzog et al., 1998Herzog et al., , 2003)), used in this study has originally been developed to simulate volcanic eruption plumes (Graf et al., 1999).Subsequently, it was applied to biomass-burning plumes (e.g.Trentmann et al., 2006) and atmospheric applications (Guo et al., 2004).The model uses a z vertical coordinate, and can be run both in 2-D and 3-D mode.Tracers, including all hydrometeors, are active.This means that the density of the gas-particle mixture and heat capacity of the fluid are influenced by the local concentration of all tracers (Oberhuber et al., 1998).
The simulations in this work include the following physical processes, and are similar to the work in Gerken et al. (2013a): shortwave radiation (Langmann et al., 1998), longwave radiation (Mlawer et al., 1997), and bulk microphysics (Herzog et al., 1998).Turbulence closure is of the order of 1.5 and predicts both turbulent length scale and turbulent kinetic energy (horizontal and vertical) (Herzog et al., 2003).The modified Hybrid (v6) land-surface model (Friend et al., 1997;Friend and Kiang, 2005;Gerken et al., 2012) and the Coupled Ocean-Atmosphere Response Experiment (COARE)algorithm v2 (Fairall et al., 1996a, b) water surface scheme compute turbulent energy fluxes above land and water.Nam Co Lake has a mean depth of > 50 m (Wang et al., 2009).The surface model and the coupled system were shown to perform well for this site (Gerken et al., 2012(Gerken et al., , 2013a(Gerken et al., , 2014) ) and the reader is referred to these publications for a more extensive model description.

Model set-up and cases
Due to the almost parallel orientation of Nam Co Lake and the Nyenchen Thangla mountains, it is possible to simulate the system's most important dynamical features such as the diurnal cycle of turbulent fluxes, the formation of a lake breeze and the interaction between locally generated clouds and the complex topography using a 2-D approach (Gerken et al., 2013a(Gerken et al., , 2014)).Our model domain cuts across the lake, the Nam Co station, as location of our measurements, and through the Nyenchen Thangla mountain chain, which is of relatively low height at this location compared to the glaciated peaks in the western range.The maximum elevation difference between the lake level and the mountain chain is approx.800 m.As the 2-D modelling set-up does not permit flow around the mountain chain, we expect to overestimate the influence of topography.
The 2-D domain used in this work has the same configuration as in Gerken et al. (2013a)  ing window was used to smooth the topography within the basin, convective development has a negligible sensitivity to the window size as shown in the supplement to Gerken et al. (2013a).Outside the Nam Co Basin the topography is set to the same elevation as the lake.Surface energy fluxes are gradually reduced to zero in the vicinity of the horizontal boundary, as we are interested in the behaviour of the mountain-lake system enclosed by mountains and hills.In contrast to cyclic lateral boundary conditions for momentum, hydrometeor tracers and water vapour concentrations exceeding the initialised profile are removed, without introducing a density perturbation.The 2-D simulations are integrated from 04:00 to 20:00 LT.The model time step is 2.5 s.
To assess the suitability of the 2-D simulations, we compare the convective development in 2-D with a 3-D simulation of the cross section that allows for the triggering of convection in three dimensions.The 3-D simulation spans the same domain in the principal horizontal and vertical dimension, and is integrated over the first 12 h.While the vertical resolution is identical to the 2-D simulations, the horizontal domain resolves the central 80 km and thus the lake basin and topographic elements at 200 m and then applies a grid stretching to span the total domain with a total of 490 grid points.The second horizontal dimensions span 10 km with 200 m resolution and uniform topography in the y direction.The lateral boundary in this dimension is fully cyclic.The 17 July 2012 was selected as the basis for the investigations presented in this work as this day was free of large-scale synoptic effects, which are not included in the model, and showed a characteristic evolution from clear sky with few boundary-layer clouds to locally generated "Cumulonimbus" activity.Gerken et al. (2013a) showed that the model realistically reproduced the observed weather and the evolution of the vertical profiles for temperature and moisture.Figure 2 displays the temperature and moisture profiles derived from the 00:00 UTC radiosounding of 17 July.The initial relative humidity (RH) of the atmospheric profile is reduced to a maximum of 90 %.This is reasonable as there was no closed cloud cover at the time of the sounding so that cloud passages of a radiosonde would lead to an overestimation of relative humidities at the respective level.As the focus of this work is the investigation of the impacts of varying soil moisture, the initial geostrophic wind is set to 3 m s −1 throughout the model domain, which exhibited the most realistic convection in Gerken et al. (2014).This also avoids wind shear to which 2-D simulations are very sensitive (i.e.Kirshbaum and Durran, 2004).The model was initialised with 6.8 and 2.6 • C as first and second layer mean temperatures T 1 and T 2 , which corresponds to an initial surface temperature (T 0 ) of 4.5 • C. For the complete description of the Hybrid surface model initialisation the reader is referred to Gerken et al. (2012).Soil moisture in the surface model is initialised as a fraction of field capacity (FC), which is the amount of water a soil can hold against gravity.In order to investigate the influence of soil moisture on convective development, we set the initial soil moisture of the simulations to 2.0, 1.5, 1.0, 0.75, 0.5 and 0.4 × FC.These cases are subsequently referred to by their initial soil moisture (e.g."1.0 × FC").
While the 2.0 × FC case corresponds to a realistic surface initialisation for 17 July, we leave soil temperatures and thus soil heat contents unmodified and reduce soil moisture gradually to the permanent wilting point (PWP = 0.4 × FC).The lake surface temperature is initialised with 10 • C, which is reported by Haginoya et al. (2009) as Nam Co Lake's mean surface temperature in July and August for 2006-2008.

Comparison of 2-D vs. 3-D
While 2-D simulations of deep convective development are frequently used to investigate the triggering of convection (i.e.Wu et al., 2009;Kirshbaum, 2011;Garcia-Carreras et al., 2011), they may overestimate convection due to the lack of entrainment in their convective plumes (e.g.Petch et al., 2008).As a consequence, we start the analysis by comparing the convective development for case 2.0 × FC between the 2-D and 3-D simulations.As in Gerken et al. (2013a), we consider grid elements with liquid water contents (both water and ice) of q t > 10 −3 g kg −1 as clouds.
Figure 3 shows the development of vertical velocities and cloud cover at several heights for 10:00 and 11:00 LT for the case 2.0 × FC in the 3-D set-up.During this time boundarylayer clouds that form near the mountains and hills develop into a convective plume that reaches 9 km a.g.l. at 10:30 LT (not shown) and then develops further.Overall, this results in maximum updraft strengths in excess of 10 m s −1 .Subsequently, the cloud development in terms of cloud base height, cloud top height and centre of the cloud mass (Z c ) as defined by Wu et al. (2009): where z is the level and dxdz the integration area, are displayed in Fig. 4. We compare the 2-D case with the corresponding value of the 3-D simulation averaged over the additional dimension.There is little difference in the cloud base height, while cloud top heights and Z c for the 3-D simulation are slightly higher than in the 2-D case, indicating stronger convection.This is also reflected in the deposited precipitation.At the same time, the timing of the precipitation with two distinct peaks is similar between the two cases.This is somewhat counter-intuitive, as one would expect increased entrainment for the 3-D case and less strong convection.However, due to the topographic forcing clouds are organised in bands so that entrainment is of less importance than anticipated.Our results are comparable to Garcia-Carreras et al. (2011), who also had stronger cloud development in the 3-D case.This might be due to the absence of lateral entrainment of dry air due to the limited extent and cyclic nature of the third dimension.Despite some differences in the structure of convective evolution, the resulting convective development is sufficiently similar to focus on a 2-D approach to further evaluate the behaviour of the system.

Turbulent fluxes
We investigate the locally driven evolution from boundarylayer clouds to precipitating convection as is frequently observed at Nam Co Lake.Variation of initial soil moisture contents in Hybrid leads to a shift in the Bowen ratio.Figure 5 displays the diurnal cycle of the simulated median turbulent latent (Q E ) and sensible heat fluxes (Q H ) as well as incoming solar radiation (SWD).The last panel shows the observed quantities at Nam Co station.As expected, the Bowen ratio changes from approx.2.5 for the dry surface at PWP to 0.5 for the moist surface at 2.0 × FC.While the incoming solar radiation initially follows the clear-sky radiation, there is a reduction of downward shortwave radiation seen in both the simulations and the measured fluxes after solar noon.This decrease in incoming solar radiation is the strongest for 1.0 × FC, while 1.5 × FC and 2.0 × FC show a slightly weaker decline during the afternoon.This highlights the impact of soil moisture on cloud development.Larger latent heat fluxes lead to increased cloud formation, which is further discussed in the next sections.Solar radiation and surface fluxes were measured at Nam Co station, while the surface model calculated surface fluxes are for the total model domain.Due to the stochastic nature of turbulence, it is not meaningful to directly compare a single model grid cell to the measured data.However, it is possible to say that the median, simulated solar radiation flux within the Nam Co Basin for case 0.75 × FC resembles most closely the measurements, while the moister cases have lower simulated solar irradiation.A comparison between the grid cell closest to the station and measurements (not shown) indicates that case 1.5 × FC matches most closely the observed fluxes in both terms of the Bowen ratio and the decrease of fluxes in the afternoon due to cloud cover, but the timing of the decrease was too early.At the same time there is large variation of fluxes as indicated by the interquartile difference, so that simulated solar radiation flux for all cases appears to be in reasonable agreement with the measured shortwave radiation.
The spatial variability of surface energy fluxes and solar radiation also demonstrates that surface atmosphere interactions change with increasing soil moisture.With the onset of cloud development, solar irradiance of shaded grid cells begins to drop, while the median irradiance and turbulent surface fluxes remain unaffected.With increasing cloud optical depth median shortwave radiation fluxes begin to decrease.This occurs first in the most shaded cells and then progresses to the mean fluxes.For the moister simulations (1-2 × FC) there is little difference between lower and upper quartile latent heat fluxes in the early afternoon, indicating widespread cloud cover as the main driver for the decrease.However, in the later afternoon, there is again increased variability in the simulated turbulent fluxes (both median and upper quartile) that is not accompanied by a corresponding increase in the median solar irradiation flux.This indicates that in addition to solar irradiance, local circulation systems likely play a role in the surface energy balance.The impact of local circulations on the surface energy balance, turbulent fluxes and convective triggering was investigated in Gerken et al. (2014).
Eddy-covariance measurements have a footprint area of similar magnitude to ATHAM's grid cells.Nevertheless, the same problem of comparing point measurements to a distributed simulation applies.Overall, there is a general agreement in the diurnal development of fluxes, with a decrease of turbulent fluxes in the afternoon.Case 2.0 × FC corresponds to a realistic surface initialisation.A comparison of simu- lated and measured fluxes in Fig. 5f and g shows that measured median simulated fluxes in the afternoon are lower than the measured ones, while the maximum fluxes around noon show similar magnitudes.This is likely due to an overestimation in the simulated convective development that is caused by the lack of the third dimension and entrainment of dry air into convective thermals.The inherently unclosed nature of turbulent fluxes measured by the eddy-covariance technique (Foken, 2008) introduces an additional degree of uncertainty into the comparison of modelled and measured fluxes.The observed mean of the ground heat flux at the Nam Co station is of the order of 30 % of net radiation (Biermann et al., 2014), leaving approx.30 % of the total surface energy balance unclosed.Using the energy balance correction of Twine et al. (2000) conserves the observed Bowen ratio, while the recently proposed buoyancy correction according to Charuchittipan et al. ( 2014) would attribute most of the unclosed flux to Q h and would thus change the Bowen ratio considerably.The Hybrid surface model was able to reproduce the turbulent flux measurement at the Nam Co station for a wide range of observed soil moisture (Gerken et al., 2012).Additionally, Kracher et al. (2009) showed that many landsurface models tend to reproduce measured Bowen ratios.
As a consequence, we assume in this work that the Bowen ratio method of energy balance closure (Twine et al., 2000) is applicable and that there is an overall reasonable agree-ment between the diurnal evolution of modelled fluxes for the moister surface initialisations and the measured fluxes.

Soil moisture-evapotranspiration response
Immediately connected to turbulent flux dynamics and of central importance for the hydrological response of the Nam Co Basin is the question of the relationship between soil moisture and evapotranspiration.Figure 6 displays the soil moisture variables for the simulated cases.The first panel shows the development of soil moisture expressed as a function of field capacity for the upper model layer.From Fig. 6 it becomes apparent that evapotranspiration during the daytime has a large effect on modelled soil moisture.The surface model Hybrid has a comparatively simple treatment of soil moisture, which includes drainage, molecular diffusion between soil layers for transport and evapotranspiration and precipitation as sources and sinks.While there is a general decreasing tendency of soil moisture during the course of the simulations, precipitation events can replenish the soil with water.In the case of 1.0 × FC, afternoon precipitation leads to soil moisture in the upper layer that are slightly larger than the initialised soil moisture and similar to soil moisture as in the moister 1.5 × FC case.Simulated evapotranspiration is expected to show a strong dependency on soil moisture for the relatively dry cases.Indeed, there is a large increase in simulated evapotranspiration between cases 0.5 × FC and 0.75 × FC, demonstrating a soil-moisture-limited regime as discussed in Seneviratne et al. (2010).A further doubling of soil moisture to 1.5 × FC leads to a similarly big increase in evapotranspiration.This increase, however, is no longer true for the case 2.0 × FC, which in fact has a lower actual evapotranspiration than the 1.5 × FC case.Between these two cases, the surface has switched from a moisture-limited regime to a radiation limited one due to an additional increase in cloud cover.At the same time, estimated potential evapotranspiration peaks for the 1.0 × FC case (Fig. 6c).Potential evapotranspiration, as calculated in Hybrid, is unlike the other quantities not a standard model output and had to be estimated with 5 min averaged model outputs, leading to some uncertainties in its magnitude.Nevertheless, it is striking that 1.0 × FC has the largest potential evapotranspiration, while 1.5 × FC and 2.0 × FC as the moistest cases in fact have the lowest.Additionally, for case 1.5 × FC actual evapotranspiration approaches potential evapotranspiration.In addition to the transition from a soil moisture-limited to a radiationlimited regime, there is also a feedback evident between soil moisture and convective development, which acts trough the generation of wind, precipitation and the reduction of in-coming shortwave radiation.This will be further discussed in Sect.3.4.
There is little variation of evaporation from the lake surface between cases and total daytime evaporation is less than 0.5 mm (Fig. 6d) and thus 50 % smaller than evapotranspiration from the land surface of even the driest case, highlighting the role of the land surface as a source of water.Additionally, the lake does not exhibit a strong diurnal pattern in simulated evaporation.This is in reasonable agreement to Biermann et al. (2014), who did not find a diurnal cycle in the measured lake evaporation.
Figure 7 shows the importance of soil moisture for the energy balance of the Nam Co Basin.Turbulent surface fluxes are integrated over space and time.The surface energy balance on the Tibetan Plateau is of key importance for understanding hydrological processes.Due to the clean atmosphere (i.e.Cong et al., 2009) at Nam Co Lake and its high elevation, the fraction of diffuse to direct radiation is small compared to more polluted environments so that clouds, through modification of the net radiation, have a stronger influence on the surface energy balance than in other regions.Soil moisture strongly modifies the partitioning of the available energy into sensible and latent heat, evident in the Bowen ratio, and thus the surface energy balance.We separated the integrated surface energy fluxes into the area of the lake, the non-elevated land around the lake, referred to as plain, the Nam Co Basin, which is defined from the lake to the top of the Nyenchen Thangla and an equal extent on the northern side of the lake.For completeness the whole domain values are also presented.Overall, sensible heat fluxes from the lake are negligible, while the importance of Nam Co Lake for latent energy fluxes compared with the plain area is very dependent on the initial soil moisture conditions: for the cases PWP and 0.5 × FC integrated latent heat fluxes from the lake have almost the same magnitude as the fluxes from the plain area, whereas for the moister cases they are comparatively small.As a consequence the importance of the lake as a moisture source for the surrounding area decreases with moister environmental conditions.
In general, there is a rather complicated behaviour of integrated energy fluxes for the plain and the basin area, which must be attributed to local dynamics such as the lake breeze, cloud cover and the evolution of convection.The 1.5 × FC has the largest transfer of latent energy from the land surface to the atmosphere, while the moister 2.0 × FC case has a smaller transfer of both latent and sensible energy.After 12:00 LT there is a visible change in the slope for 2.0 × FC, which is associated with increased cloud cover, while a similar kink for 1.5 × FC and 1.0 × FC is less strong.For 0.75 × FC there is a different behaviour.While there is also a reduction in fluxes after 12:00 LT, but unlike the moister cases, this results in only a temporary reduction in the surface to atmosphere transfer of latent heat which then increases again, while the transfer of sensible heat is as large as for case 0.5 × FC.

Convective development
We investigated the development of locally developed convection as it is frequently observed in the Nam Co Lake basin.As discussed in Sect.3.2 there is a direct interaction between soil moisture, Bowen ratio and turbulent surface fluxes, which to a large extent drives convective evolution.Therefore, differences in convective development are expected for the simulated cases.
During the course of the day, boundary-layer clouds forming at the beginning of the simulation grow into moist deep convection that produces precipitation in the model domain.Figure 8 displays the simulated averaged cloud liquid water concentration, as well as Z c , cloud base and cloud top height.While all simulations lead to the development of deep convection, there are some differences in the amount of generated convection, the convective timing and dynamics as well as deposited precipitation (Sect.3.5).The drier PWP and 0.5 × FC cases show only limited or isolated convection before solar noon and then some deeper convection around 14:00 LT, which does not lead to strong precipita-tion.Cases 0.75 × FC and 1.0 × FC on the other hand have a slightly stronger and earlier triggering of convection and then a stronger release of convective energy in the afternoon that is accompanied by a distinct precipitation event.The moistest simulations (cases 1.5 × FC and 2.0 × FC) have a stronger cloud development before noon and then lack a strong convective triggering in the afternoon.
These three regimes become especially visible, when analysing the convective mass flux M c defined as the product of moist density (ρ) and vertical velocity (w) integrated for a given height z: and integrated over cloudy grid cells with w > 0 (Fig. 9).Cases 0.75 × FC and 1.0 × FC are the only simulations that show a peak in convective mass flux in the afternoon at 5 km and 7 km a.g.l., while the remaining simulations show a different evolution.As the development of deep convection is sustained by the release of latent energy, it is expected to gain strength with height until latent energy release becomes Local Time 06:00 08:00 10:00 12:00 14:00 16:00 18:00 Local Time smaller than the reduction of buoyancy by the entrainment of dry air.Overall, there is little difference in M c at 3 km between the cases, but the dry cases lack the moisture for a subsequent release of latent energy and strong convection in the afternoon.It should be noted that the initial profile is quite moist for the lowermost 3 km, but then there is a dry zone between 3 and 6 km a.g.l.The equivalent potential temperature profile (θ e ) reveals a predisposition for convection.Hence, comparatively large Q E fluxes are necessary to provide the atmosphere with sufficient moisture for the development of convection and the release of latent heat.At the same time, the moister simulations also do not develop strong deep convection, as they have the smallest convective mass flux at 3 km around 14:00 LT.This means that the cloud cover displayed in Fig. 8e and f develops earlier and is not associated with strong convective motion.At the same time, the cloud base height decreases with increasing moisture as more water vapour accumulates in the boundary layer, which is consistent with the investigation by Golaz et al. (2001) and Yamada (2008), who also found a decreased buoyancy for their wet surface case.Furthermore, for the intermediate cases, local differences in precipitation affect surface moisture and potentially induce circulations that lead to additional convection triggering (Clark et al., 2004).
An analysis of convective available potential energy (CAPE) tells a similar story (Fig. 10).After sunrise, simulated CAPE values in the Nam Co start to diverge between cases due to differences in the surface energy balance.Cases PWP and 0.75 × FC develop the least amount of CAPE, which is fully released between 09:00 and 14:00 LT.The moistest cases in contrast develop the largest amounts of CAPE, but it is not fully released as high soil moisture and associated lack of sensible heat fluxes reduce convection development.Consequently, CAPE is exported from the basin.Cases 0.75 × FC and 1.0 × FC present an intermediate, for which there is ample development of CAPE, which is almost fully depleted by convection during the afternoon, so that they have the highest overall release of CAPE.
Overall, there are three different regimes of convective development in the simulations for Nam Co Lake that are associated with soil moisture.Convective development requires the availability of moisture and vertical instability through surface heating, and these conditions are best met for the intermediate moisture cases, while the drier cases do not provide a strong enough moistening and the moister cases inhibit the development of vertical motion due to increase cloud cover and lack of surface heating.

Precipitation and atmospheric moisture transport
The development of convection at Nam Co Lake is associated with daytime precipitation.betan valleys (e.g.Kurita and Yamada, 2008;Ueno et al., 2009).In addition to precipitation from mesoscale convective systems there are two types of locally developed precipitation: (1) daytime isolated convection triggered over the mountain chains or, in the case of Nam Co Lake, at the collision front between the lake breeze and the geostrophic wind (Gerken et al., 2013a) and (2) convergence in the centre of the valleys through thermal circulations leading to night-time precipitation.Due to the nature of our study, we only discuss the effects of locally generated, daytime precipitation.As was demonstrated in Sect.3.3 locally generated precipitation has a large influence on the availability of soil moisture and thus on surface fluxes.This is of special importance since potential evaporation rates of approx.4 mm day −1 are of similar magnitude to the water that is stored in the uppermost 10 cm of the soil at field capacity.Actual evaporation rates for a Kobresia meadow were measured at 2 to 6 mm day −1 (Babel et al., 2014).Therefore, the water balance of the Nam Co Basin and the regional transport of water vapour are of importance.Due to the simple Kessler-type microphysics used in this study, the actual precipitation rates only allow for a qualitative comparison between cases.
Table 2 displays the deposited precipitation for the convection simulations in terms of absolute mass and as fraction of evapotranspiration.For the entire domain between 6 and 10 % of the moisture evapotranspiration is deposited as precipitation.These values are smaller than commonly reported.This is at least partially due to the integration time of our simulations, which is shorter than the typical residence time of atmospheric water vapour.The lake surface has the Table 2. Total simulated deposited precipitation (Prec, [g m −2 ]) and normalised by evapotranspiration (Prec/ET [-]) for cases PWP, 0.5 × FC, 0.75 × FC, 1.0 × FC, 1.5 × FC and 2.0 × FC on 17 July 2012 for lake area, the plain adjacent to lake, the Nam Co Basin as defined in the text and the total domain.The area definitions are the same as those used in Fig. 7.  Case F q,ANw F q,ASe F q,ET F q,P F q,3 F q,10 F q,BNw F q,BSe q tot,A q tot,B largest spread with 1 to 44 % of recycled water.In general, the lake surface receives comparatively less precipitation per area than the rest of the basin, which is in line with expectations, as a cool water body suppresses convective activity.We observe increasing precipitation with increasing soil moisture.This trend is most clear for the driest cases, while it becomes less apparent for the moistest cases.This is in accordance with the observations of convective activity, which are strongest for intermediate initial soil moisture.Additionally, the basin and the plain are over-represented in terms of deposited precipitation in comparison to the total domain.A closer investigation into the moisture budget of the Nam Co Lake basin (Fig. 11) further points to the importance of vertical moisture transport within the region.We define two control domains for the Nam Co Basin: control domain A corresponds to the lowermost 3 km a.g.l., while control domain B is located directly above and extends to 10 km.The horizontal extent of the control domains is from the centre of the model domain to the top of the Nyenchen Thangla mountain chain and a similar extent to the north-west (see also Fig. 11g for illustration).It becomes apparent that the vertical transport of moisture dominates the overall mois-ture budget of both control domains for all displayed simulations, while evapotranspiration and precipitation are comparatively small terms.This leads to a net decrease of moisture in the lower control domain A during the daytime.During the night-time a reversal of the thermal circulation and subsidence should lead to a balancing of the moisture budget, which is not included in our simulations.Table 3 presents the full moisture balance for all cases.The convective transport from domain A to domain B is the strongest for 0.75 × FC and 1.0 × FC, while the remaining cases are of similar magnitude, underpinning again the preferential convective development for intermediate soil moisture.Lateral moisture transport, or advection is of only small importance for the lower control domain as the net flux is small.Vertical moisture transport dominates for domain A and it is the primary source of moisture for domain B. In the afternoon, when convective activity ceases, the moisture budget of the upper domain becomes increasingly more dominated by lateral transport of moisture with the background wind.This signifies that clouds and water vapour are removed from the basin, indicating locally occurring convection as a source of moisture for the larger region.However, as previously mentioned, the upper level wind speeds used in this work are unrealistically low, so that under realistic conditions one would expect an even faster transport of moisture from the basin to the surrounding area.Recent work by Froidevaux et al. (2013) investigated the importance of background wind speeds and horizontal transport of convective cells during their development.While wind speeds used in this work were lower than observed, the lake-land circulation (i.e.Gerken et al., 2014) provides the necessary horizontal propagation mechanism that allows for the development of convective cells.

Conclusions
In this work we have used the high-resolution atmospheric model ATHAM, coupled with the Hybrid surface model, to investigate the influence of soil moisture on convective development in the Nam Co Lake area.Surface moisture and, thus, turbulent fluxes are highly variable on the Tibetan Plateau and, therefore, not only are important for the Bowen ratio and the surface energy balance, but also have an effect on atmospheric dynamics.
It was established in this work as well as in previous studies that the modelling system used in this investigation is a sensible and cost-effective approach for the most important processes with regard to surface-atmosphere interactions, the evolution from boundary-layer clouds to deep and moist convection, and its interaction with background winds and complex topography.
The ATHAM model, with a fully coupled surface, was used to investigate surface-atmosphere interactions in a cross section through the Nam Co Lake basin for simulations with different initialised soil moisture.The model produces realistic turbulent energy fluxes of latent and sensible heat.The reduction in surface moisture leads to a shift in the Bowen ratio from 0.5 to 2.5.The fluxes of the most realistic surface configuration are in reasonable agreement to measured fluxes at Nam Co station.There is a feedback between surface fluxes, cloud cover and convection.Larger latent heat fluxes lead to increased cloud cover and consequently to smaller net surface radiation.This is seen in the simulated evapotranspiration.The surface changes from a water limited to a radiationlimited evapotranspiration regime, so that increased moisture actually means smaller net surface radiation.The role of the lake as a source of moisture depends on the initial surface moisture.Latent heat fluxes from the lake have a substantial contribution to the overall water budget of the basin only for very dry land surfaces.Our research clearly demonstrates the importance of soil moisture for the overall surface energy balance of the lake system and thus for potentially wider implications for the Tibetan Plateau's energy balance as a whole.
We subsequently investigated convective development and convective mass fluxes and found a "convective optimum" for the Nam Co Lake basin at intermediate soil moisture.At small soil moisture there is insufficient moisture available for latent heat release, which drives convection in the conditionally unstable Tibetan atmosphere.At very high soil moisture there is increased cloud cover, small sensible heat fluxes and decreased vertical motion.So while precipitation does occur, there is little convective development.
This has implications for the moisture budget of the Nam Co Basin.We found that the basin itself and especially the plain and the mountains are a preferential area for precipitation so that weather stations close to the lake are not representative for the entire basin.At the same time vertical transport of moisture through convection and the subsequent lateral transport with the background wind, may be an important water source for the larger region.Overall, precipitation effects can be directly seen in the soil moisture response of the coupled model and thus have an influence on turbulent fluxes, convective development and the surface energy balance.As precipitation events are both small in scale and short in duration, it is extremely important to get a good representation of regional precipitation in time as well as location.This requires mesoscale models at the appropriate resolutions of the order of a few hundred metres.Additionally, both Hohenegger et al. (2009) and our previous work (Gerken et al., 2013a) highlight the importance of atmospheric profiles, through stability and convective inhibition, in the triggering of convection.This demonstrates a further need for accurate high-resolution data sets for soil and atmospheric data.
As current environmental impacts, such as climate change and land use change, not only are expected to impact surface cover and soil moisture, but also it is important to assess how these changes will affect convective development and thus the hydrological cycle on the Tibetan Plateau.

Figure 1 .
Figure 1.Land use map of Nam Co Lake created from Landsat data.The yellow line shows the path of radiosonde ascent on 17 July 2012, 00:00 UTC.The yellow circle indicates the location at which the radiosonde reaches 10 km a.g.l. and the cold-point height.The cyan cross indicates the location of Nam Co station.The inset, reproduced from Gerken et al. (2012), shows the area near the small lake and the location of Nam Co Station (NamITP).© 2012, The Microsoft Corporation, © Harris Corp, Earthstar Geographics LLC.

Figure 3 .
Figure 3. Development of convection in the 3-D simulation of case 2.0 × FC at 10:00 LT for heights 1, 3, 5, 7 and 9 km a.g.l.(a-e); at 11:00 LT (g-k).The filled contours indicate vertical velocities (w, [m s −1 ]) and the red contour lines show the location of clouds; (f, l) display the topography of the model domain.The location of the lake is indicated by the blue line.

Figure 4 .
Figure 4. Comparison between the 2-D and 3-D simulation of case 2.0 × FC for (a) cloud base height, (b) cloud top height, (c) the centre of cloud mass (Z c as defined in text) and (d) mean deposited precipitation.

Figure 5 .
Figure 5. Development of turbulent surface fluxes in the Nam Co Lake basin for cases PWP, 0.5 × FC, 0.75 × FC, 1.0 × FC, 1.5 × FC and 2.0 × FC on 17 July 2012 (a-e).Thick black lines correspond to the median flux over land.For SWD and Q E also the upper and lower quartiles are given as thin lines, presenting a measure of spatial flux variation.The dashed line is clear-sky SWD; (g) measured turbulent fluxes near Nam Co research station.Thin lines correspond to directly measured eddy-covariance fluxes.Thick lines are energy-balance corrected fluxes according to Twine et al. (2000).

Figure 7 .
Figure 7. Spatially and temporally integrated turbulent energy fluxes (E) at Nam Co Lake for cases PWP, 0.5 × FC, 0.75 × FC, 1.0 × FC, 1.5 × FC and 2.0 × FC on 17 July 2012: (a, b) sensible and latent energy over the lake,(c, d) over the plain adjacent to the lake, (e, f) the Nam Co Basin as defined in the text and (g, h) in total domain.

Figure 8 .
Figure 8. Modelled development of convection at Nam Co Lake for cases PWP, 0.5 × FC, 0.75 × FC, 1.0 × FC, 1.5 × FC and 2.0 × FC on 17 July 2012: contours correspond to mean cloud water equivalent concentrations in the Nam Co Lake basin [g m −3 ].The dashed line indicates the height of the centre of cloud mass (Z c ) and black lines indicate cloud top and cloud bottom heights.The upper panel of each subplot indicates mean deposited precipitation within each 5 min interval [g m −2 ].

Figure 11 .
Figure 11.Instantaneous moisture flux f q [kg m −1 s −1 ] for Nam Co Lake basin.(a, c, e) Moisture transport into control volume A, as indicated in (g) of this figure, representing the Nam Co Basin up to 3 km a.g.l.: evapotranspiration (blue); precipitation (cyan); net horizontal flux (black); net vertical flux (red); and resulting total flux (grey, dashed).The precipitation flux is not displayed as it is of negligible magnitude.(b, d, f) Transport into the mid-tropospheric control volume B with the same horizontal extent and from 3 to 10 km vertical extent; (g) is a schematic drawing of the control volumes.

Table 3 .
Total integrated moisture flux (F q ) into Nam Co Lake basin between 06:00 and 16:00 BST.The control domains denoted with subscripts A and B are indicated in Fig.11g.The subscripts Nw, Se stand for the north-west and south-east boundaries.Positive fluxes across the lateral boundaries indicate flux into the domain.ET and P are the evapotranspiration and precipitation flux; 3 and 10 indicate the vertical boundaries of the control volumes, with a negative sign for upward flux.q tot is the total net moisture flux for a volume.