Volume 118, Issue 22 p. 12,317-12,331
Regular Article
Free Access

Uncertainty in atmospheric profiles and its impact on modeled convection development at Nam Co Lake, Tibetan Plateau

Tobias Gerken

Corresponding Author

Tobias Gerken

Department of Micrometeorology, University of Bayreuth, Bayreuth, Germany.

Department of Geography, Centre for Atmospheric Science, University of Cambridge, Cambridge, UK.

Corresponding author: T. Gerken, Department of Micrometeorology, University of Bayreuth, 95440 Bayreuth, Germany. ([email protected])Search for more papers by this author
Wolfgang Babel

Wolfgang Babel

Department of Micrometeorology, University of Bayreuth, Bayreuth, Germany.

Search for more papers by this author
Fanglin Sun

Fanglin Sun

Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences, Lanzhou, China.

Search for more papers by this author
Michael Herzog

Michael Herzog

Department of Geography, Centre for Atmospheric Science, University of Cambridge, Cambridge, UK.

Search for more papers by this author
Yaoming Ma

Yaoming Ma

Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China.

Search for more papers by this author
Thomas Foken

Thomas Foken

Department of Micrometeorology, University of Bayreuth, Bayreuth, Germany.

Member of Bayreuth Center of Ecology and Environment Research, Bayreuth, Germany.

Search for more papers by this author
Hans-F. Graf

Hans-F. Graf

Department of Geography, Centre for Atmospheric Science, University of Cambridge, Cambridge, UK.

Search for more papers by this author
First published: 29 October 2013
Citations: 12


[1] This work investigates the influence of atmospheric temperature and relative humidity profiles obtained from radio soundings, NCEP-I and ERA-Int reanalysis and GFS-FNL analysis data on the simulated evolution of clouds and convection at Nam Co Lake on the Tibetan Plateau. In addition to differences in moisture, the initial atmospheric profiles exhibit considerable differences in near-surface temperatures that affect vertical stability. Our analysis is carried out during 2 days in summer 2012 using a 2-D high-resolution modeling approach with a fully interactive surface model so that surface fluxes react to changes in cloud cover. Modeled convection for the radio-sounding profile compares reasonably well with weather observations for the first day, but less well for the second day, when large-scale synoptic effects, not included in the model, become more important. The choice of vertical profile information leads to strongly differing convection development, translating into modifications of the surface energy balance and of the energy and water cycle for the basin. There are strong differences spanning one order of magnitude in the generated precipitation between the model simulations driven by different vertical profiles. This highlights the importance of correct and high-resolution vertical profiles for model initialization.

Key Points

  • Convection development is highly dependent on source of atmospheric profile
  • Modeled convection for the radiosounding profile agrees well with observations
  • Changes in convection lead to large differences in surface energy balance

1 Introduction

[2] The Tibetan Plateau is the largest mountain highland in the world and has an average elevation of more than 4500m above sea level (asl). As a summer heat source, it acts to modify the monsoon circulation and influences precipitation patterns downstream in Eastern Asia [i.e., Gao et al., 2006; Xu et al., 2008; Chen et al., 2012]. A changing climate and socioeconomic factors contribute to land-use change and pasture degradation [Cui and Graf, 2009], which have an impact on regional circulation, precipitation, cloud cover, and hydrological resources [i.e., Cui et al., 2007a, 2007b, 2006; Immerzeel et al., 2010; Yang et al., 2011], and may adversely affect livelihoods.

[3] On the Tibetan Plateau, observations are sparse and there are no permanent weather stations above 4800m [Maussion et al., 2011], leading to a bias toward lower altitudes. At the same time, reanalysis data sets and gridded precipitation products such as TRMM (Tropical Rainfall Measuring Mission) have large errors due to terrain effects [Frauenfeld et al., 2005; Yin et al., 2008; Ma et al., 2008]. Differences between terrain height and surface levels in global models range from several hundred to almost 2000m [Wang and Zeng, 2012]. While most comparison studies between observations and reanalysis data focus on ground-based data [Frauenfeld et al., 2005; Ma et al., 2008, 2009; Wang and Zeng, 2012], where systematic biases in elevation can, with some effort, be accounted for, there is to our knowledge only one study, addressing errors in vertical atmospheric profiles of temperature, humidity, and wind speed [Bao and Zhang, 2013]. They compare temperature, moisture, and wind at standard pressure levels, obtained from the GAME-Tibet (Global Energy and Water Exchanges [GEWEX] Asia Monsoon Experiment, 1998) radio soundings with commonly used reanalysis data and find relatively small biases for wind speed and temperature in the averaged data, but considerable biases for relative humidity (RH): Below 200hPa, ERA-Int (European Centre for Medium-Range Weather Forecasts Interim reanalysis, [Dee et al., 2011]) has a positive bias of approximately 10% RH, while the RH-bias in NCEP/NCAR RA-I (National Centers for Environmental Prediction/National Center for Atmospheric Research Reanalysis-I, [Kalnay et al., 1996]) decreases from +10% close to the surface to −5%at 300hPa. There was a generally negative temperature bias for both products below 200hPa (−0.5°C increasing to −2°C with height for ERA-Int and −1.5°C close to the surface decreasing to −0.5°C). There was also a temporal pattern in the discovered biases with the smallest values during the daytime. It should be noted, however, that such a comparison of the mean over a large number of profiles does not address the quality of individual profiles nor the nature of observed profiles between standard pressure levels, which is important for the evolution of convection.

[4] With increasing horizontal and vertical model resolutions, the quality of atmospheric profiles becomes more important as regional weather models start to resolve convection directly. This requires the atmospheric profiles to reflect local conditions and to be of sufficient vertical resolution to adequately model convection.

[5] The total amount of precipitable water in the atmospheric column increases from approximately 5mm to >15mm between premonsoon and monsoon seasons [Taniguchi and Koike, 2008]. Stable isotope investigations indicate that the importance of the Bay of Bengal and the Arabian Sea as a water source for summer precipitation decreases northward on the Tibetan Plateau [Tian et al., 2001, 2003, 2007] as the influence of the monsoon becomes weaker. Isotope measurements at high temporal resolutions show a complex pattern of synoptic, monsoonal precipitation, and subsequent water recycling through locally generated precipitation [Kurita and Yamada, 2008].

[6] Atmospheric and ecosystem modeling provide important tools to investigate these developments if they (1) include the relevant processes in their model physics and parameterizations, (2) can be applied on relevant spatial and temporal scales, and (3) can be supplied with initial conditions that reflect the state of the system to be modeled. In Gerken et al. [2012, 2013a] we have demonstrated the first two assertions and will discuss here the third point with respect to atmospheric profiles of temperature and moisture. Five different atmospheric profiles from 17 and 18 July 2012 are combined with field measurements for high-resolution studies of convection development at Nam Co Lake.

[7] During the monsoon season, there is frequent development of deep convection at Nam Co Lake, starting with shallow boundary layer clouds in the morning, which then develop into moist convection. This is partially caused by the interactions between the lake and the complex topography. Scenes taken from MODIS-terra/aqua, which show the development of locally triggered convection at Nam Co Lake, illustrate this development (Figure 1).

Details are in the caption following the image
Modis visible composite pictures for 17 and 18 July 2012. Nam Co station (N30°46.44 E90°57.72, 4730m asl) is indicated by a blue circle.

[8] The simulations conducted in this study are designed to investigate the impact of different vertical profiles on the simulated convection development at Nam Co Lake. We (1) show that the model is able to reproduce realistic convection development for the local radiosonde profile, (2) investigate the impact of other profiles on simulated convective activity, and (3) discuss its impact by investigations of the energy and water cycle.

[9] These simulations provide valuable information about the impact of uncertainties in atmospheric profiles for the estimation of surface-atmosphere interactions and the surface energy balance in general and specifically for remote and data-sparse regions.

2 Materials and Methods

[10] In this work we conduct 2-D simulations with the ATHAM (Active Tracer High-Resolution Atmospheric Model) model for a cross section through the Nam Co basin (Figure 2). We use realistic topography and a fully interactive surface model in order to gain a better understanding of physical processes involved in the development of convection at the lake. This work uses a similar setup as tested and applied in the 2009 summer monsoon season [Gerken et al., 2012, 2013a], but uses directly measured profiles and reanalysis data in order to investigate the influence of atmospheric profiles on convection evolution.

Details are in the caption following the image
Land-use map of Nam Co Lake created from Landsat data. The yellow and cyan lines indicate the path of radiosonde ascents on 17 and 18 July 2012. The markers (yellow and cyan circles) indicate 10km above ground level (agl) and the cold point heights. The black cross indicates the location of Nam Co station ©2012 The Microsoft Corporation ©Harris Corp, Earthstar Geographics LLC.

2.1 Site Description

[11] From 6 July to 6 August 2012 a field experiment was conducted by the University of Bayreuth in cooperation with the Institute of Tibetan Plateau Research, Chinese Academy of Sciences at the Nam Co Lake Monitoring and Research Station for Multisphere Interactions (N30°46.44; E90°57.72, 4730m asl) located approximately 300m from a small lake, which is directly adjacent to the southeast shore of Nam Co Lake (Experiment documentation in Gerken et al. [2013b]). During this period, radiosondes were launched on 8 days for 00, 06, and 12UTC between 17 and 29 July, while eddy covariance, soil, and standard atmospheric measurements were carried out continuously. The eddy-covariance measurements were post-processed using the TK2/3 software package [Mauder and Foken, 2004, 2011], applying all flux corrections and post-processing steps for turbulence measurements recommended in Foken et al. [2012] and Rebmann et al. [2012]. A portable Vaisala radio-sounding system was deployed, using RS92-SGP sondes, Totex-TA600 balloons, mobile GPS antenna, and processed with SPS-220 and DigiCora III MW21 (v3.2.1). The error of the RH measurements is given as <5% by the manufacturer.

[12] Local solar time (LST) is assumed to be UTC+6, which is in almost perfect agreement with the times of sunrise, noon, and sunset at 23:05, 06:02, and 13:00UTC as given by the National Oceanic and Atmospheric Administration Sunrise/Sunset Calculator for 17 July. Hence, the soundings correspond to 1 h after sunrise, solar noon, and 1 h before sunset.

[13] Soil parameters at the station were measured in 2009 [Biermann et al., 2009], and vegetation parameters were reassessed in 2012 (Table 1). The area around the station is sandy, rather sparsely vegetated, and the soil has little water retention capacity. Gerken et al. [2012] showed that the Bowen ratio at Nam Co station varied considerably from 3 to 0.5 depending on the soil moisture.

Table 1. Description of the Soil and Surface Parameters Determined for Nam Co Station (N30°46.44; E90°57.72) in Summer 2009 as Well as Leaf Area Index and Vegetation Height Estimated for July 2012
Texture sandy
Porosity 0.39
Field capacity 0.05
Wilting point 0.02
Heat capacity (cp) (Jm−3 K−1) 2.2×106
Thermal conductivity (Wm−2 K−1) 0.20
Surface albedo (α) 0.2
Surface emissivity (ε) 0.97
Vegetated fraction 0.6
Leaf area index (m2 m−2) 0.6
Vegetation height (m) 0.07

2.2 Model Description

[14] We use the nonhydrostatic, cloud resolving ATHAM-model [Oberhuber et al., 1998; Herzog et al., 2003], which was designed for the study of volcanic plumes [Graf et al., 1999] and then further developed for biomass-burning plumes [e.g., Trentmann et al., 2006] and clouds [Guo et al., 2004]. ATHAM's dynamic core is capable of solving the Navier-Stokes equations in two or three dimensions, and transported tracers such as all hydrometeors are active in the sense that they influence heat capacity and density of the mixture at each grid point [Oberhuber et al., 1998].

[15] The physical processes included in the model for this work are as follows: 1.5-order turbulence closure predicting horizontal and vertical turbulent kinetic energy as well as turbulent length scale [Herzog et al., 2003], short- and long-wave radiation [Langmann et al., 1998; Mlawer et al., 1997], bulk-microphysics [Herzog et al., 1998], 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, 1996b] water surface scheme for turbulent energy fluxes above land and Nam Co Lake, which has a mean depth of >50 m [Wang et al., 2009]. The surface models were demonstrated to perform well at this site in Gerken et al. [2012].

2.3 Model Setup and Cases

[16] Both the lake shore and the Nyenchen Thanglha mountain chain are oriented almost parallel to each other so that a 2-D cross section is capable of simulating the most important dynamical features of the lake-mountain system. The mountain chain is located approximately 10km south of Nam Co Lake. In the absence of large-scale forcings, the circulation in the basin is primarily driven by a lake breeze and a thermal mountain circulation, which was successfully simulated in 2-D [Gerken et al., 2013a]. The horizontal domain cuts across Nam Co Lake research station, where our measurements were situated, and through a relatively low section of the mountain chain. As 2-D simulations have a tendency to overestimate the influence of topography, we deem this acceptable. The domain size is 153.6km using 200m horizontal resolution, which gives a total number of horizontal grid points of nx=770. The lateral boundary conditions are cyclic for momentum, but hydrometeors and water vapor in excess of the initial profile are removed without perturbing density. There are 175 layers, starting at 50m vertical resolution for the first 50 layers, then stretching to a constant vertical resolution of 200m for the last 50 layers below the model top, set to 17.5km above ground level (agl). The model topography uses the ASTER-DEM (Advanced Spaceborne Thermal Emission and Reflection Radiometer-Digital Elevation Model) with 90m resolution smoothed with a 2km moving window, removing vertical cliffs and single grid point depressions. A sensitivity study into the effects of topographic smoothing on our results showed, that smoothing windows ≤ 2km only have a small influence on the shape of topography and the maximum elevation of the relief. The variation in generated precipitation was considerably smaller than the variation between the cases chosen in this work (see supporting information for details). As we are interested in the Nam Co basin, the topography outside is set to the lake level and turbulent surface fluxes are gradually reduced to zero near the lateral boundary. The simulations are integrated from the initial profile for 12h with a time step of 2.5s from 04:00LST to 16:00LST. No synoptic effects or external changes to the profile are included in the model, so that large-scale weather developments cannot be reproduced.

[17] We acknowledge the limitations of our approach, which arise from the lack of a third dimension and thus result in a simplified flow field and reduced entrainment of dry air into convective updrafts, potentially leading to an overestimation of convective activity. But as the focus of this work is on the comparison of the convection development between profiles rather than attempting to simulate specific days, we believe that the chosen approach yields valuable information about the interaction of processes during convection evolution and about the impact of uncertainty in atmospheric profiles on convection. Additionally, we lack distributed observations that would be required to validate the results of 3-D simulations.

[18] This work investigates the development of moist convection on 17 and 18 July 2012 using five different vertical atmospheric profiles of temperature (T) and relative humidity (RH). The cases denoted RS use the original measured radiosonde profile, StdLev uses standard pressure levels (500, 400, 300, 250, 200, 150, 100, 70, and 50hPa) extracted from RS, NCEP uses the NCEP/NCAR Reanalysis-I [Kalnay et al., 1996], ERA uses the ERA-Interim reanalysis [Dee et al., 2011] and GFS uses the Global Forecasting System's final analysis product (GFS-FNL) [Kanamitsu et al., 1991; Caplan et al., 1997]. For model initialization, the initial RH is reduced to a maximum of 90% in order to prevent cloud formation at the start of the integration time.

[19] In this work the focus was placed on the impacts of T and RH, so that the initial geostrophic wind speed was uniformly set to 3ms−1, which produced most realistic convection in Gerken et al. [2013a]. This also addresses that the 2-D simulations are highly sensitive to wind shear [i.e., Kirshbaum and Durran, 2004].

[20] The surface configuration for both days is presented in Table 2. The lake surface temperature is initialized with 10°C, which corresponds to the mean temperature of Nam Co Lake in July and August for 2006–2008 [Haginoya et al., 2009].

Table 2. Soil Model Initialization for Temperature and Upper-Layer Soil Moisture: T0, urn:x-wiley:jgrd:media:jgrd50941:jgrd50941-math-0001, and urn:x-wiley:jgrd:media:jgrd50941:jgrd50941-math-0002 Correspond to “Skin” Temperature and Mean Layer Temperatures for the First and Second Model Layers (0.1 and 4m Layer Depth)a
T0 urn:x-wiley:jgrd:media:jgrd50941:jgrd50941-math-0003 urn:x-wiley:jgrd:media:jgrd50941:jgrd50941-math-0004 SM0
Date (°C) (°C) (°C) (-)
17 July 4.5 6.8 2.6 2.0
18 July 7.1 9.0 3.3 2.0
  • a SM0 is upper-layer soil moisture expressed in terms of field capacity. The model initialization procedure follows Gerken et al. [2012] using soil data measured at Nam Co station.

2.4 Atmospheric Profiles

[21] Both days selected for our analysis show a development from fair weather cumulus in early morning to observed Cumulonimbus activity from the midmorning onward (10:00 and 9:00LST for 17 and 18 July), as is frequently observed in the Nam Co Lake basin. Figure 3 displays the vertical profiles for temperature, relative humidity, and wind speeds for these 2 days at 00UTC (6:00LST). The radiosondes launched at Nam Co show that both measured profiles have inversion layers of more than 0.5K temperature increase in the midtroposphere. These are most likely the result of large-scale subsidence. While there is little synoptic activity on 17 July, the NCEP-I reanalysis and the decrease in relative humidity indicate that the Nam Co region becomes influenced by large-scale subsidence on the following day. The cold point in the profiles is approximately at 85hPa or 13km agl (not shown). For T, there is little difference between the sounding and the gridded products above 300hPa, whereas differences of several K are found below. In both days, ERA underestimates temperatures close to the surface, while GFS overestimates them considerably. In terms of RH, there are much greater differences between the profiles. The gridded data sets reflect less the observations from radiosondes. For 17 July, there is a good agreement between RH at Nam Co and Nagqu, a permanent sounding station located 2° to the East and 0.75° to the North, while the Lhasa sounding station (1° to the South) shows a different profile. This indicates that moisture fields are not purely determined by local conditions but are also dependent on air masses and their advection. There is little agreement between gridded products. GFS and NCEP agree reasonably well with the sounding in the lower troposphere, but there is no RH data available above 300hPa for NCEP. ERA-Int and GFS are excessively moist aloft, with ERA-Int being generally too moist, despite being the highest spatial resolution data set. The sounding of 18 July is overall moister than the day before. Interestingly, NCEP closely reflects the Lhasa sounding for both days, which is assimilated into the reanalysis. The gridded products show little variability between each other and between days, while the soundings are noisier and deviate at some altitudes where high wind speeds are found.

Details are in the caption following the image
Vertical profiles for 17 and 18 July 2012 00 UTC (6:00h Local time) at Nam Co station (N30°46.44; E90°57.72) as determined by radiosonde ascent (black), NCEP reanalysis (red), ERA-Interim (orange), and GFS-FNL analysis (yellow). The gray circles and triangles indicate the radiosonde measurement at Nagqu (N31°28.8; E92°03.6, circle) and Lhasa (N29°39.6; E91°07.8, triangle): (a, b) T (°C); (c, d) RH (%); (e, f) U (ms−1); and (g, h) V (ms−1). The black crosses indicate the height of inversion layers with more than 0.5K increase in temperature. Gridded products are bilinearly interpolated to the station location.

3 Results

[22] Our simulations show the development of clouds, which are defined for the purpose of this work as grid cells with a total content of condensed ice and water of qt>10−3 gkg−1, and convection within the Nam Co basin. Shortly after the start of the simulations, shallow boundary layer clouds form, which subsequently grow deeper with moist and deep convection developing later in the day. As in Gerken et al. [2013a], we use the center of the cloud mass (Zc) as a diagnostic for the activation of clouds [Wu et al., 2009], which will develop into moist convection:
with z as the vertical coordinate and dxdz as the area element to be integrated over.

[23] For both days it is apparent that simulations initialized with atmospheric profiles from different sources lead to big differences in convective development. Convection occurs in several phases with all simulations reaching similar cloud morphology, but showing differences in timing that are related to feedbacks between vertical profiles, surface fluxes, and cloud microphysical structure (Figures 4 and 5). On 17 July for RS, there are shallow boundary layer clouds until 7:00LST and then there is a transitional period with activated clouds that form and dissolve, while at the same time the cloud top height and the center of the cloud mass increase in height. At approximately 10:30LST, simulated convection suddenly becomes deep and substantial rain is generated. The weather observations for Nam Co station for 17 July show the first Cu congestus clouds at 8:00LST and a heavy thunderstorm was recorded over the plain south of Nam Co station at 11:30LST, which dissolved after approximately 30 min. More Cumulonimbus clouds were observed during the afternoon. This illustrates the rapid convection development in the Nam Co basin. Compared to RS, StdLev shows a more continuous convection development with time, which only accelerates in the afternoon. The convection development in RS on the other hand is characterized by sudden increases in the center of the cloud mass. This is likely to be associated with layers of high stability at 1.5, 3, and 5km agl found in the sounding (Figure 6). Convection development in NCEP appears to be an intermediate between RS and StdLev both in terms of Zc and mean cloud concentrations, which is not surprising given the close similarity of NCEP to the sounding at the standard pressure levels. ERA, the moistest case, has the fastest development of convection and produces the largest amount of clouds and precipitation. GFS in contrast shows the longest delay between the activation of clouds, which is followed by a relatively fast growth in convective height after 10:00LST. It also produces by far the least amount of rain and has lower cloud concentrations, than RS.

Details are in the caption following the image
Modeled development of convection at Nam Co Lake for cases (a) RS, (b) StdLev, (c) NCEP, (d) ERA, and (e) GFS on 17 July 2012: Contours correspond to mean cloud particle concentrations in the Nam Co Lake basin. Each contour level corresponds to 0.1gm−3. The dashed line indicates the height of the center of cloud mass (Zc), and black lines indicate cloud top and cloud bottom heights. The blue shaded area shows the accumulated precipitation.
Details are in the caption following the image
As in Figure 4, but for 18 July 2012.
Details are in the caption following the image
Atmospheric lapse rate (dθdz−1, Kkm−1) for simulations of (a, b) 17 July 2012 at 6:00 and 12:00LST and (c, d) 18 July 2012 for profiles from cases RS (black), StdLev (blue), NCEP (red), ERA (orange), and GFS (yellow).

[24] On 18 July there are similar differences in precipitation and convection development between the cases: RS shows a step-pattern in the development of convection that is associated with a strong inversion layer at around 4km agl. After the first activation of boundary layer clouds, clouds are confined below the inversion until deep convection develops at approximately 10:30LST. StdLev, in contrast, develops both faster and stronger resulting in maximum precipitation rates comparable to the much moister ERA case. Convection in ERA develops later with all clouds confined to the first 2 km above ground until 10:00LST. The NCEP run appears to be comparable to StdLev for cloud concentration, convection development, and precipitation timing although the amount of precipitation is smaller than StdLev. As on the previous day, the GFS case develops late and produces little precipitation. GFS's high cloud top height in the morning is not caused by convection, but results from mountain waves leading to condensation and a layer that is considerably moister than the day before.

[25] The following sections discuss the generation of precipitation (section 3.1), the influence of vertical stability (section 3.2), and the impact of convection on the surface energy budget (section 3.3). Sections 3.4 and 3.5 discuss the profile evolution and feedbacks between clouds and the surface.

3.1 Precipitation

[26] A key variable governing environmental processes on the Tibetan Plateau is precipitation. There is a large difference in the dynamics and total amount of simulated precipitation between the cases. Atmospheric stability and surface flux dynamics impact the development of convection and thus of precipitation. The microphysics used in this work includes graupel and rain. Snow is not represented separately. We are more interested in the comparison between the different cases than in the absolute values. The 18 July produces a larger amount of simulated precipitation than 17 July for all cases except GFS, which is related to the moister initial profiles. In reality though, this day was influenced by upper level drying, a synoptic effect not included in our simulations, suppressing convection. There is approximately one order of magnitude difference in the deposited precipitation between the driest case (GFS) and StdLev or ERA, which are the wettest (Table 3). When distinguishing between precipitation that falls within the basin (PrecB) and the total precipitation (Prectot), we find that a substantial fraction (25–60%) of the simulated precipitation occurs within the Nam Co basin, which is defined here as the area between the lake and the top of the Nyenchen Thanglha mountain chain in the south and an equal area north of Nam Co Lake. Nam Co Lake is situated at the northern edge of the area influenced by the monsoonal circulation. The lake itself may be a water source for the region through the export of moisture from the basin, especially during breaks in the monsoon. A competing hypothesis to this is that Nam Co Lake mainly constitutes a closed water cycle, where the evaporated water is deposited as precipitation on the surrounding mountain slopes and thus remains within the catchment.

Table 3. Deposited Total Precipitation at Nam Co Lake for the Model Runs on 17 and 18 July 2012 Within the Nam Co Basin as Defined in the Text PrecB and Entire Domain Prectot
17 July 2012 18 July 2012
PrecB Prectot PrecB Prectot
Run (m3) (m3) (m3) (m3)
RS 9.8 17.3 11.6 23.2
StdLev 11.1 28.1 43.4 73.7
NCEP 5.7 18.5 34.2 52.9
ERA 19.4 58.5 19.9 71.3
GFS 3.9 9.5 2.2 6.1

[27] The precipitation simulated in this work is consistent with locally generated cumulus convection frequently observed on the Tibetan Plateau [i.e., Uyeda et al., 2001], which is controlled by the surface and triggered over mountain ranges [Gerken et al., 2013a]. According to Fujinami et al. [2005] and Sato et al. [2007], there is a second major source of precipitation within Tibetan mountain valleys, occurring in the evenings, when the reversal of the daytime thermal circulation leads to convergence and moist convection in the valley centers. This effect, not included in our studies, would increase the percentage of locally deposited precipitation. Overall, recycling of locally generated precipitation is likely to be an important part of the water cycle at Nam Co Lake. Our findings also highlight the importance of initialized profiles for the generation of precipitation and show that the considerable uncertainty in RH and surface effects, impacting stability, lead to large uncertainties in precipitation.

3.2 Influence of Atmospheric Stability

[28] We expect atmospheric stability to play an important part in controlling convective evolution at Nam Co Lake. Additionally, vertical profile resolution and inversion layers matter. For 17 July (Figures 6a and 6b) there is direct evidence from the step-wise increase in cloud heights as displayed in Figure 4 that the stable inversion layers in the radiosonde below 4km agl profile have a delaying effect on convection development and precipitation generation. Until 7:00LST, clouds remain below the inversion layer at 1km agl and later the center of the cloud mass is confined to altitudes below the higher stable layers until after 10:00LST, when the inversion layers of the initial profile have been eroded. While the potential temperature gradient (dθ/dz) of StdLev is close to the mean gradient of RS, there is a more gradual development of convection in the absence of distinct inversion layers. In general, the stability of all profiles in the middle troposphere is comparable, but there are differences below 3km agl that influence cloud development: GFS is the most stable profile near the surface and the least stable for higher altitudes. Hence, the late activation of clouds is followed by rapid growth in cloud height around 10:00LST. There is some convergence between the profiles of different sources for 12:00LST that highlight the influence of surface heating. However, especially for GFS and RS, some features of the initial profile remain distinguishable. The development of the stable layer in NCEP is associated with the absence of moisture in the initial profile above 300hPa, which causes instability in the atmospheric profile initialized by ATHAM.

[29] There is a similar situation for 18 July (Figures 6c and 6d), where the layer of high stability below 4km agl in the RS case coincides with a plateau in the convection development between 9:00 and 10:30LST. However, StdLev has a similar plateau but no inversion layers or particularly high stability below 6km agl. We suspect two reasons for the effect. First, the stability of StdLev is highest of all profiles at this altitude and, second, the feedback between cloud cover and surface fluxes leads to a reduction of surface fluxes and thus of convective activity. This is further discussed in the next section. ERA is the least stable profile in proximity to the surface and then most stable below 2km agl, explaining the late activation of clouds and the fact that moist convection does not occur before 10:00LST.

3.3 Energy Transfer to the Atmosphere

[30] The surface energy balance of the Tibetan Plateau is a key factor to understand regional climate. This section discusses the impact of different profiles for the simulated surface energy balance. The ATHAM modeling system is set up with an interactive surface. Clouds modify the available radiation and thus have direct impact on the surface energy balance and consequently on turbulent energy fluxes. Due to the remoteness and thus low aerosol loading on the Tibetan Plateau [Cong et al., 2009] and its high elevation, there is little indirect shortwave radiation, so that cloud shading effects are stronger compared to other regions. The surface and its reaction to shading impact the development of convection by controlling sensible heat and water vapor fluxes. As a consequence, and unlike simulations with prescribed idealized fluxes as are commonly used in high-resolution simulations, we expect feedbacks between convection development and the total energy exchanged through turbulent surface fluxes. One measure of these interactions are the spatially and temporally integrated turbulent surface fluxes. Despite the removal of condensate at the lateral boundary, which reduces cloud cover and increases insolation near the boundary, simulations are comparable to each other and we believe that the comparison of the runs in terms of relative values yields more information than the total amount of energy that is supplied to the atmosphere.

[31] Figures 7 and 8 show the integrated turbulent surface fluxes separated for the lake surface, the land area adjacent to the lake (called plain), and the total domain. The plain is defined as the area between the lake and the Nyenchen Thanglha mountain chain plus an equal area to the north of the lake. Sensible heat fluxes from the lake, both simulated and measured [Biermann et al., 2013], are small compared to the land surface, so that the variations' influence on the total energy balance are negligible. For latent heat, where the lake does have a significant contribution to the total energy and water vapor that is supplied within the basin, there are differences of up to 50% between runs.

Details are in the caption following the image
Spatially and temporally integrated turbulent energy fluxes (E) at Nam Co Lake for cases RS, StdLev, NCEP, ERA, and GFS on 17 July 2012: (a) Sensible and (d) latent energy over lake, (b, e) over plain adjacent to lake, and (c, f) in total domain.
Details are in the caption following the image
As in Figure 7, but for 18 July 2012.

[32] On 17 July (Figure 7), there is little variation in the sensible and latent energy contributions between the simulated cases. For the basin fluxes (Esens, plain and Elat, plain) StdLev is approximately 30% larger than the rest of the fluxes. The simulations using GFS, which has the latest triggering convection, do not have a higher energy contribution than the rest of the simulations due to strong modeled convection after 11:00LST, leading to reduced energy input during the time of strongest solar radiation. For the whole domain there are two groups with NCEP and ERA having smaller Esens, plain and Elat, plain than the other simulations. This is despite the substantially differing initialization profiles and differences in simulated convection. However, as discussed in the next section, in both cases, the model has developed similar atmospheric profiles of T and RH in the middle troposphere at 12:00LST.

[33] For 18 July (Figure 8), there is a larger scatter in surface heat fluxes than during the previous day, and there are different evolutions for the latent and sensible heat fluxes. For both Esens, plain and Elat, plain, StdLev and NCEP have the lowest energy transfer, which is explained by their similar convection dynamics and early triggering, while ERA has the largest energy flux despite being the moistest close to the surface. GFS has a large contribution to the latent heat flux over land, but not for sensible heat. Interestingly, for GFS, lake fluxes behave differently to land fluxes, which is due to higher water vapor mixing ratios over water for GFS.

[34] In general, there is considerable variation in the total amount of energy that is transferred from the surface to the atmosphere ranging up to 50% between the cases. The absolute variation is larger for latent than for sensible heat but is comparable when normalized with the magnitude of the fluxes.

3.4 Profile Development

[35] The above analysis is dependent on reasonable model performance. We, therefore, compare the development of the vertical profiles in the simulated cases with respect to the radio sounding and reanalysis data. As previously mentioned, we chose 4:00LST as start time of the simulations, which corresponds to 1h before sunrise, while the atmospheric data is from 6:00LST or 1h after sunrise. We assume that there is little change in the profile over this 2 h period, which is confirmed for all cases, because modeled and observed surface fluxes are small, so that little heat is transferred from the surface to the atmosphere.

[36] Figures 9 and 10 show the modeled atmospheric profiles compared to the radio soundings and reference profiles at 6:00LST and 12:00LST (00UTC; 06UTC). On the first day (Figure 9), the simulated temperature profile in the RS case (Figure 9a) closely resembles the sounding profile for 12:00LST, except that the inversion layer at 3.7km agl has eroded. StdLev is very similar to RS. For RH, the situation is different: Up to 3km agl, the mean RS humidity profile matches the sounding, but between 3 and 5km agl and above the height of the initial inversion layer, there is substantially more moisture in the case RS than in the observed profile. This is due to simulated convection. While there is some increase of moisture in the 06UTC sounding above 3km and weather observations report the occurrence of Cb incus clouds from 10:00LST as well as a thunderstorm at 11:30LST, the convective transport of moisture is overestimated in the RS simulation. This is likely to be attributed to the 2-D approach, which underestimates entrainment of dry air into thermals. Above 5km, the 06UTC sounding profile has dried considerably compared to the 00UTC sounding, which is not reproduced in the simulations due to the exclusion of synoptic effects.

Details are in the caption following the image
Comparison of the mean profiles of the ATHAM simulations initialized with (a) full radiosonde profile (RS, black lines), (b) radiosonde standard level information only (StdLev), (c) NCEP-I (red), (d) ERA-Int (orange) reanalysis, and (e) GFS-FNL (yellow) products compared to the respective profile data for 17 July 2012 00 UTC (dashed, 6:00LST) and 06 UTC (solid, 12:00LST). The gray lines are the simulated profiles for each case.
Details are in the caption following the image
As in Figure 9, but for 18 July 2012.

[37] For the cases initialized with gridded products, it clearly shows that the simulations and their RH profiles are dominated by the initialized water vapor contents. Cases ERA and GFS are much too moist above 3km compared to the observed 06UTC data. In general, the differences in temperature and RH between the different simulations for the lowermost 3km have decreased, which indicates the influence of the surface on both the boundary layer (approximately 2km high) and the lower troposphere in general.

[38] All simulations are moister than the measured profile in close proximity to the surface, which might be due to the fact that the mean profile also contains lake cells, whereas the radiosonde was launched in several hundred meters distance to water bodies.

[39] For the second day (Figure 10), the situation is different. With respect to temperature, there is still a reasonable agreement between the simulated and measured temperature profiles, but for RH they show little agreement. While one could argue that boundary layer moisture profiles are reasonably close to the observed sounding data and the gridded products, the 06UTC radio sounding is very dry above 4km agl and is substantially drier throughout the entire column. As RH is influenced by both absolute water contents and temperature, the drier boundary layer can be explained by the higher temperatures, but above that there is no indication of a substantially warmer profile. However, the NCEP reanalysis shows the region to be in subsidence, which would lead to a reduction in RH and, in addition, advection of dry air under the prevailing west-wind conditions might be responsible. Nevertheless, there were Cb incus clouds observed in the Nam Co basin during the afternoon of 18 July, showing that convection was indeed taking place.

[40] Overall the development of profiles shows that the 2D model somewhat overestimates convective evolution, as expected, but also that the profile evolution is in good agreement with the initial conditions.

3.5 Cloud—Surface Flux Interactions

[41] Feedbacks between the highly variable cloud cover and surface fluxes are of importance for the surface energy balance. This section investigates the interactions between cloud development and surface fluxes. Due to space constraints we focus on 17 July, but 18 July exhibits similar behavior. Figure 11 shows the temporal development of the sensible and latent heat fluxes (QH and QE) for the five simulations and fluxes measured at Nam Co station. For the simulated fluxes, the median of the fluxes is given together with the upper and lower quartiles of shortwave radiation and latent heat flux as a measure of variability. The energy balance of the measured eddy-covariance fluxes was closed according to the Bowen ratio [Twine et al., 2000], assuming a ground heat flux corresponding to 10% of the net radiation. Even though it is difficult to compare the spatially averaged simulated fluxes with measurements that have the footprint of approximately 1 grid cell, it is apparent that both have similar Bowen ratios (Bo≈0.5) and flux evolution. For all cases except StdLev, both QH and QE increase gradually during the morning hours and then decrease sharply in the afternoon due to cloud cover, reducing incoming shortwave radiation (SWD). StdLev and measured fluxes show a more gradual decrease and higher SWD. As previously discussed, 2-D simulations tend to overestimate convection so that simulated fluxes may be excessively low due to cloud shading. On the other hand, the flux measurements were carried out in close proximity to the lake, which induces subsidence and thus are not entirely representative for the basin. Hence, we cannot determine which of these processes is responsible for the difference.

Details are in the caption following the image
Development of turbulent surface fluxes in the Nam Co Lake basin for cases (a) RS, (b) StdLev, (c) NCEP, (d) ERA, and (e) GFS on 17 July 2012, respectively (QH — ; QE —-; down-welling shortwave radiation (SWD) gray). Thick black lines correspond to the median flux over land. SWDQ and QE,Q are upper and lower quartiles of SWD and QE, giving a measure of spatial flux variation. SWDcl is clear sky SWD; (f) 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].

[42] We generally find strong evidence for a positive correlation between surface fluxes and SWD (p<0.05, not shown). There is, however, considerable scatter and generally only 15 to 50% of the variance is explained by changes in shortwave solar radiation as determined by orthogonal regression. This highlights the importance of additional factors affecting surface-exchange processes. One possible approach to analyze the relative importance of SWD-flux interactions is to estimate the ratio of the standard deviation of the flux and its spatially averaged value (σFlux/<Flux>, Figure 12). For this we only consider grid cells at the altitude of the lake in order to eliminate the influence of topography on fluxes. In the diurnal cycle of this quantity, there is a minimum in the morning hours, when shallow cumulus dominates. Earlier in the morning, the variation of fluxes is higher reaching a maximum around 7:00LST, when the simulated surface fluxes start to increase. This occurs approximately 1 h later than observed, indicating some inertia in the surface model. With the development of moist convection later in the day, the spatial variation of fluxes increases and becomes as big as the mean value of the fluxes. This is both due to the decreasing magnitude of fluxes and increased differences in solar shading. When comparing the variances of fluxes between shaded and unshaded areas, it becomes apparent that the shaded areas show the larger variance, sometimes exceeding the mean flux, while σFlux/<Flux> in unshaded areas remains below a maximum of 50% and below 30% except during times with large liquid water paths (lwp). This variation must be attributed to other processes than cloud-surface feedbacks. At higher lwp, shaded and unshaded areas show different behavior. The variation of the shaded areas increases with lwp (mature convection), while unshaded areas have a maximum flux variation with intermediate lwp (developing convection). A possible explanation for this is the development of cold pools and fronts and their influence on surface wind speeds and turbulent exchange. In general, the variation of sensible heat fluxes is larger than the variation of evapotranspiration, which is expected, because QH depends directly on skin temperature, while QE is dependent on more variables in both the surface model [Gerken et al., 2012] and reality.

Details are in the caption following the image
Normalized standard deviation of turbulent surface fluxes at lake level on 17 July (σFlux/<Flux>) for cases (a) RS, (b) StdLev, (c) NCEP, (e) ERA, and (e) GFS: diurnal development; (f–j) with respect to domain-averaged liquid water path (lwp,(mm)). Unshaded areas: QH (red), QE (blue); shaded areas: QH (orange), and QE (black).

4 Discussion and Conclusions

[43] In this work we use a coupled high-resolution atmospheric and surface model to perform 2-D simulations of daily cycles, especially of the evolution from shallow cumulus to deep, precipitating convection at the Nam Co Lake basin on the Tibetan Plateau, which is frequently observed during the summer monsoon. We investigate the influence of different atmospheric profiles. The aim of this work is less to quantitatively simulate the convection development of specific days, but rather to qualitatively investigate the atmospheric profiles' influence on simulated convection. This is then used to evaluate the impact of profile choice on surface-atmosphere-interactions and energy exchange. Due to the sparseness of observations, there are considerable uncertainties in the initial atmospheric profiles. While gridded atmospheric products, such as reanalysis data, are readily available, the available vertical resolution is coarse and grid cells may not accurately reflect local conditions.

[44] As reported by Bao and Zhang [2013], we found gridded atmospheric products at Nam Co Lake to have a reasonable quality for temperature, but show larger biases with respect to moisture. Furthermore, their lack of vertical resolution leads to the omission of inversion layers and abrupt changes of water vapor, which also have an impact on convection development. For both these limitations there is presently no feasible approach for correction, highlighting the importance of direct measurements through atmospheric soundings.

[45] Radio soundings on the other hand, are cost-intensive and difficult to conduct in remote areas. Additionally, they give very local and temporal “snapshots”, which may not be representative of the entire area. This work shows that the uncertainties in vertical profiles are important for modeling studies both for the development of convection and impact both the surface energy balance and the generation of precipitation on the basin scale. Hence, careful choice and evaluation of profiles is needed.

[46] ATHAM is generally able to adequately simulate the convection development and its dynamics at Nam Co Lake. There are limitations to the 2-D approach such as a likely overestimation of convective activity and cloud cover. We find that differences in stability and relative humidity in the initial profiles are consistent with differences in convective activity and generated precipitation. Our simulations match weather observations at Nam Co Lake quite well for 17 July, a day with negligible contribution from large-scale advection. As our model does not include synoptic-scale forcings, the 18 July matches less well observations and later soundings, which is due to the selected case and not to problems with the applied model.

[47] Local water recycling of locally generated convection is an important part of the regional water cycle. With the choice of profile having a large impact on deposited precipitation, accurate profiles of temperature and moisture are crucial for modeling of the water cycle.

[48] From the integrated energy fluxes, we see that, depending on the profile used, there are considerable differences in the amount of water vapor and sensible heat that are transferred from the surface to the atmosphere. One of the main causes for this are changes in cloud cover and thus net radiation. This highlights the importance of the vertical profiles for the surface energy balance. Potentially, also studies of climate change with common coarse-grid numerical models will be affected by these processes. Overall, due to the uncertainties in the gridded profile information, users of such data should be aware of the potential impacts of their choice with respect to the surface energy balance and convection development. Ideally, the robustness of modeling results should be tested by using input and validation data from several independent sources.

[49] Feedbacks between cloud shading and surface fluxes are likely to play a significant role and need to be investigated. We investigated some of the connections between radiation input and surface fluxes and find that modeled and observed fluxes are comparable. Simulations utilizing StdLev profiles have the best agreement, while the other profiles produce too strong reduction of fluxes in the afternoon. Differences in radiation input are the main contributing factor to spatially and temporally varying fluxes, but additional effects, such as the generation of fronts or cold pools associated with deep convection, are also important as shown by the dependency of the flux variance to liquid water path. The soil model used in this work is able to capture the fundamental dynamics of fluxes. For the future, we believe that specifically adapted surface models that react on the timescales of large eddies as presented in Liu and Shao [2013] would further increase the value of our approach.


[50] This research was funded by the German Research Foundation (DFG) Priority Programme 1372 “Tibetan Plateau: Formation, Climate, Ecosystems” as part of the Atmosphere - Ecology - Glaciology - Cluster (TiP-AEG): FO 226/18-1,2. The work described in this publication has been supported by the European Commission (Call FP7-ENV-2007-1 grant 212921) as part of the CEOP-AEGIS project (http://www.ceop-aegis.org/) coordinated by the University of Strasbourg." We thank Kathrin Fuchs for her help during the radiosonde measurements. The map of Nam Co was produced by Sophie Biskop and Jan Kropacek within DFG-TiP and Phil Stickler of the Cambridge Geography Department. MODIS images were provided through AERONET, and we thank the MODIS team for their work. GFS-FNL data were produced by the National Center for Environmental Prediction (NCEP). NCEP Reanalysis data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.esrl.noaa.gov/psd/. ERA-Int reanalysis was downloaded from http://data-portal.ecmwf.in. Sounding data for Nagqu and Lhasa were obtained through the University of Wyoming's Atmospheric Sounding Portal (http://weather.uwyo.edu).