Theoretical and Applied Climatology

, Volume 109, Issue 3, pp 577–590

Sensitivity of MM5-simulated planetary boundary layer height to soil dataset: comparison of soil and atmospheric effects

Authors

    • Department of MeteorologyEötvös Loránd University
  • Ferenc Ács
    • Department of MeteorologyEötvös Loránd University
  • Borbála Laza
    • Department of MeteorologyEötvös Loránd University
  • Ákos Horváth
    • Hungarian Meteorological Service
  • István Matyasovszky
    • Department of MeteorologyEötvös Loránd University
  • Kálmán Rajkai
    • Research Institute for Soil Science and Agricultural Chemistry of the Hungarian Academy of Sciences
Original Paper

DOI: 10.1007/s00704-012-0597-y

Cite this article as:
Breuer, H., Ács, F., Laza, B. et al. Theor Appl Climatol (2012) 109: 577. doi:10.1007/s00704-012-0597-y

Abstract

The effects of two soil datasets on planetary boundary layer (PBL) height are analyzed, using model simulations. Simulations are performed with the MM5 weather prediction system over the Carpathian Basin, with 6 km horizontal resolution, investigating three summer days, two autumn, and one winter day of similar synoptic conditions. Two soil datasets include that of the United States Department of Agriculture, which is globally used, and a regional Hungarian called Hungarian unsaturated soil database. It is shown that some hydraulic parameter values between the two datasets can differ up to 5–50%. These differences resulted in 10% deviations in the space–time-averaged PBL height (averaged over Hungary and over 12 h in the daytime period). Over smaller areas, these relative deviations could reach 25%. Daytime course changes in the PBL height for reference run conditions were significant (p < 0.01) in ≈70% of the grid points covering Hungary. Ensemble runs using different atmospheric parameterizations and soil moisture initialization setups are also performed to analyze the sensitivity under changed conditions. In these cases, the sensitivity test showed that irrespective of the radiation and PBL scheme, the effect of different soil datasets on PBL height is roughly the same. PBL height is also sensitive to field capacity (Θf) and wilting point (Θw) changes. Θf changes seem to be more important for loamy sand, while Θw changes for the clay soil textural class.

1 Introduction

Land-surface modeling schemes in numerical weather prediction systems have been required to address problems associated with vegetation and soil characteristics, especially with the appearance of the mesoscale models. In soil hydrology, soil texture has a crucial role. Even today, it is still important to include data for representing horizontal and vertical distribution of soil texture, not only in the mesoscale but also in global medium-range weather forecast systems. For example, the spatial variability of soil hydraulic properties was taken into account by the European Centre for Medium-Range Weather Forecasts (ECMWF) at the end of 2007 with the introduction of the revised Hydrology of Tiled ECMWF Scheme for Surface Exchanges over Land module (Balsamo et al. 2010). The modelers recognized that turbulent heat fluxes cannot be simulated with acceptable accuracy without having more than two soil layers (Smirnova et al. 1997; Abramopoulos et al. 1988). Many of them (e.g. Pan and Mahrt 1987; Alapaty et al. 1997; Mölders 2005) underlined the importance of the knowledge of actual soil moisture content (SMC). Later, it was recognized that the available soil moisture content (ASMC) is at least as or even more important than the actual soil moisture content (Reen et al. 2006; Teuling et al. 2009). Since ASMC is so important, it is obvious that knowledge of the accurate values of soil hydraulic parameters (field capacity and wilting point) becomes as important as the knowledge of soil hydraulic functions (Braun and Schädler 2005; Ács 2005).

The ratio between the actual plant available soil moisture content and ASMC (RASMC) is determined by many factors. The most important, which have been extensively analyzed, is the soil texture (e.g. Alapaty et al. 1997; Niyogi et al. 1999; Raman et al. 2005), soil hydraulic functions (Cuenca et al. 1996; Shao and Irannejad 1999; Braun and Schädler 2005), variation of soil hydraulic parameters (Ek and Cuenca 1994; Mölders 2005; Teuling et al. 2009), actual soil moisture content (McCumber and Pielke 1981; Kim and Entekhabi 1998; Quintanar et al. 2008), and the spatial variability of soil properties (Rajkai 1991; Bell et al. 1980). To our best knowledge, among these studies, none of them analyzed the sensitivity of land-surface–atmosphere interaction processes to the used soil database. Such first attempts were made by Horváth et al. (2006, 2007), who focused on deep convection processes analyzing 24-h sums of convective precipitation. The sensitivity was analyzed in terms of skill score (Horváth et al. 2009) and Spearman correlation (Ács et al. 2010) coefficients. In the latter study, it was shown that the Spearman correlation differences for precipitation spatial distribution were significant on the p < 0.05 level.

The aim of this study is to show that soil parameters derived from different soil databases could have notable effects not only on convective precipitation but also on PBL height simulations. We used the MM5 mesoscale modeling system and two soil datasets: the United States Department of Agriculture (USDA) (Cosby et al. 1984) and the Hungarian unsaturated soil database (HUNSODA) (Nemes 2002). These are introduced in Sections 2 and 3. On the chosen simulation days, the simulation area was completely or almost completely cloudless. In the latter case, only a small number of cirrus clouds were observed. This extreme weather is caused by a high-pressure system center over or near to the Carpathian Basin; its appearance frequency is quite high. More information about the climatic aspects of the investigated weather type is given in Section 3.4. Numerical experiments are described in Section 4. The results are presented in Section 5. In the analysis, we will focus on planetary boundary layer (PBL) height evolution and its spatial distribution. In some cases, PBL height changes will be separately considered for each soil textural class because such an aspect is also important. In Section 5.3, the obtained sensitivity to soil dataset will also be analyzed in terms of different atmospheric effects. Finally, the main conclusions are drawn.

2 Theory

2.1 MM5 model

The sensitivity studies were performed with the MM5 Version 3 numerical weather prediction system (Grell et al. 1994). In this model, a terrain-following sigma coordinate system is used, the horizontal discretization applies the Arakawa B grid (Arakawa and Lamb 1977). The vertical velocity, temperature, specific humidity, and the different type of hydrometeors (cloud water, rain, snow, graupel, and cloud ice particles) are predicted at the center of the grid; while the horizontal momentum components are predicted on grid corners. Model runs were performed using a horizontal resolution of 6 km. For the prediction, 20 vertical levels were used with nine further diagnostic levels defined in the lowest 1.5 km of the atmosphere. Model runs can be performed using different physical parameterization schemes. We used the following basic settings for the reference run: Grell’s scheme (Grell et al. 1991) for convection, Reisner’s scheme (Reisner et al. 1998) for microphysics, the rapid radiative transfer method (RRTM) scheme (Mlawer et al. 1997) for radiation transport, the Mellor–Yamada–Janjic (MYJ) Eta scheme (Janjic 1990, 1994) for planetary boundary layer processes, and the Noah land surface model (LSM) (Chen and Dudhia 2001a) for simulating lower land-surface transport processes. In the following sections, the most important components for this study will be briefly described.

2.2 Planetary boundary layer and radiation schemes

Janjic’s Eta PBL scheme was applied for describing turbulent mixing in the PBL and to estimate PBL height. This is a local vertical mixing (exchange only exists between neighboring atmospheric levels) scheme based on the prediction of turbulent kinetic energy (TKE). Analyzing the TKE from the surface upwards, PBL height is defined at a height where TKE falls below 0.4 m2s − 2 in stable and 0.6 m2s − 2 in unstable stratification after reaching a maximum in the lower atmospheric layers. This height is between the height of the first model layer and 5,000 m. If the geopotential height is lower than sea level height, then the PBL height is 0 m. The MRF scheme (Hong and Pan 1996) completely differs from the Eta scheme. It uses a nonlocal closure scheme based on the Troen and Mahrt (1986) representation of counter gradient heat and moisture transfer and K-theory. The PBL height is determined from the critical bulk Richardson number. As opposed to the Eta scheme, vertical mixing in clouds is also included in the MRF.

In the RRTM scheme, both the short- and long-wave radiation transfer schemes are implemented. The absorption spectrum includes the effect of water vapor, carbon dioxide, and ozone. Radiation will be altered by the clouds and precipitation if they are present according to the cloud scheme used. The CCM2 scheme (Hack et al. 1993) is similar to the RRTM scheme; however, it also includes the effect of unresolved clouds if the relative humidity exceeds a threshold and also provides radiative fluxes at the surface.

2.3 Land surface model

As mentioned above, the Noah LSM (Chen and Dudhia 2001a, b) was used for simulating lower boundary conditions. Its core is a multilayer soil model together with a single layer snow and vegetation model. Evapotranspiration is calculated as a sum of soil evaporation (Ee), transpiration (Et), and evaporation of intercepted water (Ew) on the vegetation. The three components depend on the potential evaporation (Ep), which is Penman-based, using stability dependent aerodynamic resistance (Mahrt and Ek 1984). Soil evaporation depends on RASMC, defined directly from wilting point (Θw) and field capacity (Θf):
$$ E_{\rm e}=\left( 1-\sigma_{\rm f} \right) \cdot E_{\rm p} \cdot {\rm RASMC} $$
(1)
where RASMC = (Θ − Θw)/(Θf − Θw). Transpiration depends on Θw and Θf via stomatal resistance and also on saturated soil moisture content (Θs) if there is an intercepted canopy water content (I) of the vegetation:
$$ E_t=\sigma_{\rm f} \cdot E_{\rm p} \cdot B_{\rm c} \cdot \left[ 1- \left( \frac{I}{\Theta_{\rm s}} \right)^n \right] $$
(2)
The parameter Bc expresses the effect of canopy resistance based on Ek and Mahrt (1991). Stomatal resistance is calculated as a function of air temperature, air humidity, solar radiation, and soil moisture. Soil moisture affects stomatal resistance similarly to Eq. 1 (Jacquemin and Noilhan 1990). Evaporation of intercepted water depends solely on Θs.
$$ E_{\rm w}=\sigma_{\rm f} \cdot E_{\rm p} \cdot \left( \frac{I}{\Theta_{\rm s}} \right)^n $$
(3)

The ratio between Ee and Et is determined from the green vegetation fraction (σf). The model was set up to make calculations in four soil layers having their lower boundary at depths of 10, 30, 60, and 100 cm. Soil hydraulic functions are parameterized after Campbell (1974).

2.4 Significance test

The following statistical approach is used to analyze the relationship between PBL height and soil parameter database. Irrespective of soil type, the incoming solar radiation is the main driver of the diurnal course of PBL height. Therefore, the diurnal course must be considered in order to estimate the effect of different soil parameters on PBL height. For i and j (i = 1,45; j = 1,111) grid points, the expected value mi,j(t) of the simulated PBLHU at time t is approximated by sine and cosine waves of 1 day and one half-day period lengths. The latter cycle is introduced to describe the asymmetry of the diurnal course. Coefficients in this approximation are estimated by least squares. The expected value μi,j(t) of PBLUS at time t can be obtained using the same method. The variance \(d_{ij}^2(t)\) of PBLHU at time t and the variance \(\delta_{ij}^2(t)\) of PBLUS at time t can be obtained by the above procedure except when squared centralized data are used. (A centralized dataset consists of original data minus its actual estimated expected value.) Finally,
$$ X_{ij}(t)=\frac{{\rm PBL}_{ij}^{\rm {HU}}(t)-m_{ij}(t)}{d_{ij}(t)}-\frac{{\rm PBL}_{ij}^{{\rm US}}(t)-\mu_{ij}(t)}{\delta_{ij}(t)} $$
(4)
is formed for further purposes. The null hypothesis is that no systematic difference exists between PBLHU and PBLUS, and so the expected value of \(Y_{ij}={\rm PBL}_{ij}^{\rm {HU}}(t)-{\rm PBL}_{ij}^{\rm {US}}(t)\) is zero. If the null hypothesis holds the test statistic,
$$ Pt_{ij}=\frac{\overline{Y_{ij}(t)/s_{ij}(t)}}{\sqrt{\frac{1+\alpha_{ij}}{1-\alpha_{ij}} \cdot \frac{1}{T}}} $$
(5)
can be well approximated by a standard normal random variable, where T represents the number of data and \(s_{ij}^2(t)\) is the variance to Yij(t) estimated to fit the data \(\big[\big({\rm PBL}_{ij}^{{\rm HU}}(t) {\kern-1pt}-{\kern-1pt} m_{ij}(t) \big) {\kern-1pt}-{\kern-1pt} \big( {\rm PHR}_{ij}^{{\rm US}}(t) {\kern-1pt}-{\kern-1pt} \mu_{ij}(t)\big)\big]^2\) to sine and cosine waves of lengths of 1 day and one half-day periods. Note that Eq. 5 is a modification of Student t test due the autocorrelation between consecutive values of Yij(t). The denominator is obtained by modeling the dataset Yij(t) as a first order autoregressive process. Specifically, the autoregressive coefficient αij is identical with the one-lag autocorrelation calculated from dataset Xij(t). For any fixed significance level p, the critical value for rejecting the null hypothesis is obtained utilizing the Gaussian approximation to Eq. 5. In cases where the PBL height does not show a diurnal course (e.g., constant) or the diurnal course is present for shorter than 90 min (usually in winter), the method cannot be applied. These are treated as errors.

3 Data

3.1 Spatial distribution of surface parameters

Two different land-use datasets can be applied by default to MM5: USGS-25 and SSiB-17. The latter set’s spatial resolution over Hungary is merely 1° × 1°, which is insufficient for mesoscale prediction, while the former is available at a 0.1° horizontal resolution for Europe. The categories from the fine mesh are then interpolated to the 6 × 6 km grid, which were applied in the model of Dudhia et al. (2005). Over Hungary, the most common land use category is dryland, cropland, and pasture (Fig. 1a), which gradually changes to grassland and deciduous broadleaf forest in the hilly areas of the northeastern and southwestern parts of Hungary. Vegetation parameters for the three most common land use types are presented in Table 1. Leaf area index is set to four for all vegetation types.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig1_HTML.gif
Fig. 1

a Land use distribution according to USGS. b Soil texture distribution according to FAO in the model domain

Table 1

Vegetation parameters of the USGS land use categories

Category

Albedo

Number of root layers

Minimal stomatal resistance (s/m)

Visible solar flux (W/m2)

Roughness length (m)

Grassland

0.19

3

40

100

0.08

Dryland cropland v

0.19

3

40

100

0.07

Deciduous broadleaf forest

0.19

4

100

30

0.8

Spatial distribution of the soil textural classes is presented in Fig. 1b. The FAO (Food and Agriculture Organization)-STATSGO distribution is used, which has a 5’ horizontal resolution. From the available 12 USDA classes, only 6 textural classes appeared in the model area. They are as follows: loamy sand (lS), sandy loam (sL), loam (L), sandy clay loam (scL), clay loam (cL), and clay (C). The most frequent in Hungary is cL (41.2%), followed by C (28.6%), and L (13.1%). The relative coverage of the remaining three textures is below 7%; for lS, sL, and scL, it is 4.2%, 6.9%, and 5.2%, respectively.

3.2 Soil datasets

Two soil datasets are used in this study: the USDA—Cosby et al. (1984) and the HUNSODA (Nemes 2002). The USDA contains ≈1,400 soil samples, while the HUNSODA, ≈470. In the USDA, an average of 78% of soil samples are sand (d > 0.2 mm) and about 10% are silt (0.2 mm > d > 0.02 mm). In the HUNSODA, an average of 38% of soil samples are sand and 42% are silt (Nemes et al. 2005). This obvious difference, together with the differences in the definition of the soil textural classes (USDA—Cosby et al. 1984, HUNSODA—Filep and Ferencz 1999), is the reason why there are pronounced differences in the soil parameter values. As already mentioned, Campbell’s 1974 hydraulic function parameterization was used:
$$ \Psi(\Theta)=\Psi_{\rm s} \cdot \left( \frac{\Theta}{\Theta_{\rm s}} \right)^{-b} $$
(6)
$$ K(\Theta)=K_{\mathrm s} \cdot \left( \frac{\Theta}{\Theta_{\rm s}} \right)^{2b+3} $$
(7)
Except for saturated soil hydraulic conductivity, Ks, all other parameters (Θs, Ψs and b) were determined using Eq. 6, applying simple linear regression. Ks is calculated after Cosby et al. (1984) in the USDA, while after Fodor and Rajkai (2005) in the HUNSODA. Θf is determined by different methods in the two databases. For US soils, field capacity is defined as the Θ value for which K is equal to 5.787·10 − 9 m/s (Hillel 1980), that is
$$ K(\Theta_{\rm f})=5.787 \cdot 10^{-9}~\rm{m/s} = 0.5~\rm{mm/day} $$
(8)
For HU soils, Θf is defined by a critical value of soil water potential (Várallyay 1973):
$$ \Psi=10^{2.3}{\rm cm\ H_2O} $$
(9)
The calculated parameter values for both databases are presented in Table 2. The parameter values are given only for those soil textural classes, which are represented in the model domain.
Table 2

Soil parameter values referring to USDA and HUNSODA database

Soil texture

Θs (m3/m3)

Θf (m3/m3)

Θw (m3/m3)

Ψs (m)

b

q (kg/kg)

Ks (m/s)

USDA

   Loamy sand

0.421

0.383

0.028

0.036

4.26

0.82

1.41·10 − 5

   Sandy loam

0.434

0.383

0.047

0.141

4.74

0.6

5.23·10 − 6

   Loam

0.439

0.329

0.066

0.355

5.25

0.4

3.38·10 − 6

   Sandy clay loam

0.404

0.314

0.067

0.135

6.66

0.6

4.45·10 − 6

   Clay loam

0.465

0.382

0.103

0.263

8.17

0.35

2.45·10 − 6

   Clay

0.468

0.412

0.138

0.468

11.55

0.25

9.74·10 − 7

HUNSODA

   Loamy sand

0.598

0.479

0.08

0.126

3.9

0.82

2.52·10 − 5

   Sandy loam

0.476

0.379

0.064

0.143

3.99

0.6

1.14·10 − 5

   Loam

0.468

0.406

0.088

0.207

4.2

0.4

4.58·10 − 6

   Sandy clay loam

0.439

0.354

0.061

0.206

4.21

0.6

7.98·10 − 6

   Clay loam

0.58

0.479

0.139

0.234

4.74

0.35

3.05·10 − 6

   Clay

0.541

0.489

0.147

0.224

6.21

0.25

8.00·10 − 7

Θs saturated soil moisture content, Θf field capacity, Θw wilting point, Ψs saturated soil water potential, b Campbell’s porosity index, q quartz content, Ks saturated soil moisture conductivity

The foremost difference between the databases appears in the porosity index b: 11.6 for USDA and 6.2 for HUNSODA. The difference of Θf and Θw between the two datasets is 14% and 23% on average, respectively. Note that for the loamy sand textural class, it is around 20% and 60%, respectively. As these parameters define the ASMC (Θf − Θw), it can be concluded that ASMC for HUNSODA is, on average, larger (with 0.05 m3/m3, ≈15%) than the ASMC for USDA. Similarly to Θf, Θs also shows higher soil water-holding capacity for HUNSODA, amounting to a 5–30% difference. It is obvious that the Θw, Θf, and Θs parameters, which are relevant for the calculation of latent heat flux, vary substantially between the two databases.

3.3 Initial and boundary conditions

The initial and boundary conditions were taken from the ECMWF’s meteorological archive and retrieval system (MARS) database. The pressure levels of the initial conditions were the same as in our model. Apart from the atmosphere’s state variables, land-surface and soil variables such as skin temperature, soil moisture content, and soil temperature in the four soil layers were derived from the ECMWF as well. During the time span from which the simulated days were chosen, the ECMWF’s surface model was unchanged. At that time, a uniform soil texture was used over Europe (ECMWF 2007). The simulation domain is defined by the northwestern 45.6°N/15.6°E and southeastern 49.3°N/22.8°E coordinates (690 km × 294 km), which includes Hungary and some parts of the Alps and the Carpathian mountains as well. For a 6-km horizontal resolution, the overall number of points was 115 × 49 = 5,635; from these, 2,307 were in Hungary. Every model run was started at 00 UTC producing a 24-h simulation. The boundary conditions were updated every hour.

3.4 Climatic aspects

To analyze the relationship between soil characteristics and PBL height, special weather conditions are required which can produce completely or almost completely cloud-free sky. Such events require high-pressure systems creating divergence in the lower atmosphere. The frequency occurrence of these synoptic situations can be determined using macrosynoptic classification systems such as the Péczely (1961) and the Hess and Bresowsky (1952) systems. We preferred Péczely’s system because this classification is based on locating high- and low- level pressure systems with respect to the Carpathian Basin. According to the classification, there are 7 weather situations determined by high-pressure systems out of 13 categories. Among them, the synoptic conditions encoded as As (high-pressure system south from the Carpathian Basin), A (high-pressure system above Carpathian Basin), AF (high-pressure system above Scandinavia), and Aw (high-pressure system west from the Carpathian Basin) are the most relevant (Gyurácz et al. 2003). In a 120-year data series (Pongrácz 2003), the appearance of these four types is 5%, 13%, 5%, and 14%, respectively. Though the frequency of all types of high-pressure system is twice larger than the frequency of low pressure systems, the probability of finding a summer day without cloud formation in the whole model region, or finding an autumn or winter day without the appearance of daylong fog is quite low. According to our experience, there are only few (about 5–10) such days in the 1–2-year period. So, in an 18-month period between June 2006 and November 2007, we found only six such days. It has to be emphasized that such ideal conditions were chosen on the behalf of the sensitivity nature of this study. The number of days where cloud-free sky is prevailing over at least half of Hungary is much higher, more than 70 days in a year. If low- or mid-level clouds are present until noon, the reduced surface fluxes and turbulence result no visible effects of the soil parameters on simulated PBL height.

3.5 Synoptic conditions of the simulated day(s)

Numerical experiments are performed on 6 days. Table 3 shows some basic weather elements for these days. In these cases, a high-pressure system determined the weather, and the advection was low or moderate. From the 6 days, 3 were in summer, 2 in autumn, and 1 on a considerably warm winter day. High-level clouds were observed on 2 days, and in two cases, fog occurred in the early hours of the morning. One day, 18 July 2007, was chosen for a more detailed sensitivity test. On this day, the pressure system over Europe showed low spatial pressure gradient with sea level pressure between 1,015 and 1,020 hPa from western Russia to the coast of the Atlantic Ocean, and from the Baltic to south Greece. The fresh morning quickly became a hot day with over 36°C in the whole country because of clear sky; the average maximum wind gust was around 8 m/s.
Table 3

Short description of the simulated days

Days

Péczely type

Tmax (°C)

T min (°C)

Cloudiness

Sunshine duration

19 Jul. 2006

A

34

11

Ci

13–15

12 Sep. 2006

An

27

5

10–12

15 Jan. 2007

A

13

–9

Ci, fog

7–8

18 Jul. 2007

A

41

14

13–15

26 Jul. 2007

A

31

7

10–15

10 Oct. 2006

A

22

0

fog

8–10

A high pressure system above Carpathian Basin, An high pressure system north from Carpathian Basin

4 Numerical experiments

In total, 100 model runs were performed; 20 runs on 18 July 2007 (see Table 4) and 80 runs on the other days (see Table 5). By the runs specified in Table 4, we can get an insight into the soil-PBL height relationship; while by the runs specified in Table 5, we can compare the sensitivity of PBL height to soil and atmospheric effects. In Table 4, run RefH is the reference run. In the reference run, Grell’s convection scheme, the RRTM radiation scheme, Janjic’s PBL-Eta scheme, and HUNSODA soil database were used. Run RefU differs from run RefH only in the soil database; in this case, the USDA soil database is used. Runs PTF and PTW differ from run RefH only in Θf values. In run PTF, Θf is increased; while in run MTF, it is decreased by 30% of its reference value. Similarly, runs PTW and MTW differ from run RefH only in Θw values. Θw changes in the same way as Θf. Comparing runs RefH and RefU, the sensitivity to soil dataset is to be investigated. The sensitivity to Θf changes is to be quantified by comparing PTF and MTF runs on one hand and the RefH on the other. Analogously, the sensitivity to Θw changes is to be estimated by comparing PTW, MTW, and RefH.
Table 4

Conditions used in the runs for 18 June 2007

 

Run name

RefH

RefU

PTF

MTF

PTW

MTW

Soil database

HUNSODA

USDA

HUNSODA

Change in Θf

0%

+30%

−30%

0%

Change in Θw

0%

0%

+30%

−30%

Table 5

Conditions used for sensitivity runs in this study

Run name

PBL

Radiation

Θini (%)

JR0H/JR0U

Janjic

RRTM

 

MR0H/MR0U

MRF

100

JC0H/JC0U

Janjic

CCM2

 

MC0H/MC0U

MRF

 

JR8H/JR8U

Janjic

RRTM

80

MR8H/MR8U

MRF

 

JR2H/JR2U

Janjic

120

MR2H/MR2U

MRF

 

The run conditions used on other days are shown in Table 5. The combination of physical parameterizations, initial conditions, and soil database is denoted by the combination of letters and numbers compared to the reference run. The first letter always indicates the PBL scheme used, J for the Janjic, and M for the MRF. The second refers to the radiation scheme (R for RRTM and C for CCM2), and the fourth to the soil database, H for HUNSODA, and U for USDA. The number in the third position shows if the initial soil moisture content had changed. For 20% decreased, unchanged, and 20% increased SMCini conditions, the number is 8, 0, and 2 respectively.

5 Results

For the present purpose, clear-sky days are needed when clouds do not disturb the incoming solar radiation at all. In this case, the PBL evolution is undisturbed, and the soil effects can be better observed. If only a small amount of low-level clouds forms in the studied area, the simulated PBL height instantly falls below 1 km even if it reached, e.g., 2.5 km before. This aspect has to be highlighted if we want to investigate the effect of soil parameters.

5.1 Reference run

Among soil factors, soil texture is one of the most important. Through RASMC, this factor also determines PBL height as discussed below. For the reference run conditions, the relative frequency distribution of soil texture classes for different PBL height intervals in Hungary on 18 July 2007 is presented in Fig. 2.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig2_HTML.gif
Fig. 2

Relative frequency distribution of soil texture classes for different PBL height intervals as obtained by the reference run in Hungary on 18 July 2007

Above sandy clay loam (scL, horizontal line filled), PBL heights appeared only in a narrow range between about 300 and 680 m. PBL heights ranging between 460 and 530 m were the most frequent with a relative frequency of about 55%. Above clay soils, the PBL heights appeared in a larger interval between about 530 and 1220 m. Interestingly, PBL heights beyond 900 m appear only above clayey soils, the clay loam (cL), and the clay (C). For loamy soils (loamy sand, sandy loam, and loam), PBL heights range between about 300 and 840 m.

It is obvious that the range of PBL heights becomes broader going from sandy to clayey soils. This fact is also determined by soil effects. The soil was pretty dry on the chosen day as shown by the spatial distribution of daytime mean (for time period 5–17 UTC) relative available soil moisture content (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig3_HTML.gif
Fig. 3

Spatial distribution of the daytime mean (5–17 UTC) relative available soil moisture content (m3/m3) on 18 July 2007

Inspecting Figs. 1b and 3, it is obvious that RASMC is much lower for C than for the scL textural class. Accordingly, sensible heat flux and PBL height are higher for C than for scL (Fig. 2). Furthermore, Fig. 1b shows that C covers a considerably larger area than scL. These two effects together resulted in the range of the PBL heights being fairly larger for C than for scL textural class.

5.2 PBL height sensitivity to soil dataset

The diurnal course of the spatially averaged PBL height and its standard deviation obtained by using US and HU soil datasets on 18 July 2007 in Hungary is presented in Fig. 4. Both courses are governed by the incoming solar radiation. From about 09:30 UTC, PBLHU (PBL height obtained using the HU soil dataset) is systematically larger than the PBLUS (PBL height obtained using the US soil dataset). PBLHU reaches its maximum at 1,380 m, while PBLUS at 1,130 m. Afterwards, the PBL height decrease occurs rapidly, and it collapses at 17:00 UTC. The PBL courses differ not only in the height but also in the dynamics, as the PBLUS collapses about 30 min before the PBLHU. These systematic differences are caused by the differences in the maximum available soil moisture content (Θf − Θw). Note that (Θf − Θw)HU is, on average, about 50 mm larger than the (Θf − Θw)US. Both \(\Theta_{\rm f}^{\rm HU}\) and \(\Theta_{\rm f}^{\rm HU}\) are higher than the US database values (except for sL and scL, with differences 0.004 m3/m3 and 0.008 m3/m3, respectively). Therefore, RASMCHU is less than RASMCUS; accordingly, the sensible heat flux and the PBL height will be larger for the HU than for the US soil dataset.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig4_HTML.gif
Fig. 4

Daytime course (5–17 UTC) of the spatially averaged PBL height and its standard deviation obtained by using US and HU soil datasets in Hungary on 18 July 2007

The standard deviation of the PBL height is caused by the heterogeneities in its spatial distribution. However, it also follows the diurnal course of the incoming solar radiation. Therefore, its course is similar to the course of PBL height. Since PBLUS is, on average, smaller than PBLHU, correspondingly σUS is also smaller than σHU. From midday on, σ is about one-third of the averaged PBL heights. This fact shows that the spatial variability of PBL heights is large. σ reaches its maximum around 16:00 UTC when PBLHU–PBLUS differences are the largest. This time is also close to the time of the beginning of the PBL collapse.

Spatial distribution of the daytime mean values of the PBL height differences obtained by using HU and US soil datasets on 18 July 2007 is presented in Fig. 5. The largest differences amount to about 300–400 m. They are located in the southeastern and central northern parts of Hungary mainly over cL soil texture. The smallest differences occur over regions where the RASMC (Fig. 3) is over 0.3 or the difference between the HU and US soil parameters are negligible. RASMC over 0.5 is found at lakes and snow pack. The correspondence between RASMC and PBL height difference only deviates in a small number of cases. These areas are usually found in the mountains, where the advection transports the surplus heat in RefH compared to RefU, causing higher PBL heights.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig5_HTML.gif
Fig. 5

Spatial distribution of the daytime mean PBL height differences (m) obtained by using HU and US soil datasets (PBLHU–PBLUS) on 18 July 2007

Spatial distribution of significant differences (p < 0.01) in the diurnal course of the PBL height obtained by using HU and US soil datasets on 18 July 2007 is presented in Fig. 6. The differences are significant over 71.7% of the territory of Hungary. For most of these grid points, RASMC is low (mostly below 0.15). The least differences occur in areas with soil textures sL and scL. For lS and cL soil textures, where the HU/US soil parameter differences are the largest, almost all grid points show significant differences.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig6_HTML.gif
Fig. 6

Spatial distribution of the significantly different (p < 0.01) diurnal courses of PBL heights obtained by using HU and US soil datasets (PBLHU–PBLUS) on 18 July 2007

5.3 PBL height sensitivity to Θw and Θf

In Noah LSM, Θf and Θw are the basic soil parameters. At the same time, Θf and Θw differ between the two datasets by an average of 14% and 23%, respectively. This was the reason why we tested the sensitivity of PBL height to changes of Θf and Θw. Just as in Section 5.1, we analyzed the relative frequencies. These relative frequencies of PBL height for PTF, PTW, MTF, and MTW runs, together with the reference run, are presented in Fig. 7. Twelve-hour mean values of PBL heights are considered for each run. The χ homogeneity test was applied for the presented distributions; in all cases, the histograms are significantly (p < 0.0001) different from the reference run.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig7_HTML.gif
Fig. 7

Relative frequency distribution of PBL height for reference, MTF, MTW, PTF, and PTW runs in Hungary on 18 July 2007

On the simulated day, soil moisture content was close to wilting point for clay textures. For low SMC, the difference between soil hydraulic functions is greater (Ács 2003) than for moist soils, so any modification in either field capacity or wilting point has a notable effect on calculated latent and sensible heat fluxes. In MTF and MTW runs, the soil is moister in respect to the reference run. Accordingly, PBL height obtained for MTF (PBLMTF) and MTW (PBLMTW) is generally lower than the PBL height obtained for the reference run (PBLref). PBLref–PBLMTF is about 135 m, while PBLref–PBLMTW is somewhat less, amounting to 95 m. In PTF and PTW runs, the soil is drier with respect to the reference run. So, the PBL height obtained for PTF (PBLPTF) and PTW (PBLPTW) is generally higher than PBLref. In this case, PBLPTF–PBLref is about 115 m, while PBLPTW–PBLref is about 200 m. Note that PBLref is around 730 m. PBLref values that are lower than 610 m are on about 30% of Hungary’s territory. PBLMTF and PBLPTF values which are lower than 610 m occupy 59% and 15% of Hungary’s territory, respectively. Note that PBLMTF values are not larger than 990 m. At the same time, 27% of the PBLPTF values are larger than 990 m. Also, it can be said that the number of affected grid points is proportional to the change (30%) of the soil parameter values.

Now, we examine the relationship between the PBL height differences obtained for different runs and soil texture classes. The relative frequency of PBL height differences which are larger than 75 m (≈10% of average PBLref height) obtained by different runs and for different soil textural classes are presented in Fig. 8.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig8_HTML.gif
Fig. 8

Relative frequency of PBLMTW–PBLref (briefly denoted as MTW), PBLMTF–PBLref (MTF), PBLPTF–PBLref (PTF), and PBLPTW–PBLref (PTW) of which differences in the absolute values are larger than 75 m for different soil textural classes

Figure 8 shows also the mean value of relative frequencies for all soil textural classes. For the MTW test for instance, this is 22.4%. The relative frequency of PBL height differences larger than 75 m averaged for all tests, and soil textural classes amounts to about 30%. This is about 15% of the total amount of PBL height changes. The sensitivity to different test types is observable. The relative frequency of the PBL height changes for PTF and MTF runs is higher than that obtained for MTW and PTW runs. Note that the dependence of the relative frequency of PBL height changes on soil textural classes is similar, for MTW and PTW on the one hand and for MTF and PTF on the other. For MTW and PTW runs, the relative frequency of PBL height changes are larger for clayey (cL and C) than sandy (lS) soils. That is, PBL height changes via RASMC changes are more sensitive to the changes of Θw for clay than for loamy sand when the soil is drier. Opposite to this, the relative frequency of PBL height changes is larger for lS than for the C textural class for MTF and PTF runs. That is, Θf changes cause a behavior precisely opposite to the previous case. Relative frequency of the PBL height changes decreases, going from sandy to clayey soils for both PTF and MTF runs. In this tendency, sandy loam is the exception. This unique performance of the sL textural class can be attributed to the fact that about half of it appears along rivers, such as the Danube (Fig. 1); therefore, it cannot affect the surrounding area significantly.

The spatial distribution of the significant (p < 0.01) PBL height changes for MTF and PTW runs is presented in Fig. 9a, b, respectively. For the MTF run, 72.6% of the points located in Hungary were significant. Similarly, 63.7% of the grid points showed significant differences for the PTW run. For the PTW run, the points of nonsignificant differences appeared at the location where the maximum available soil moisture content was low. In these points, we obtained nonsignificant differences even if the actual soil moisture content was low. Based on Fig. 8, similar results can also be expected for PTF and MTW runs; that is, both Θf and Θw changes are important irrespective of whether they increase or decrease.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig9_HTML.gif
Fig. 9

Spatial distribution of the significantly different (p < 0.01) diurnal courses of PBL heights obtained by using a the MTF run and the reference run (PBLMTF–PBLref) and b the PTW run and the reference run (PBLPTW–PBLref) on 18 July 2007

5.4 PBL height sensitivity to physical parameterizations and soil moisture initialization

PBL height sensitivity to soil database as well as to Θf and Θw parameter changes was shown only for 1 day in the previous sections. In these simulations, reference run conditions are used. So, the uncertainties related to the behavior of the atmosphere as well as to the initialization of soil moisture content were not considered. To this end, simulations were performed on different days using different PBL and radiation transfer modules and initial soil moisture content values. These run types are specified in Table 5. A total of 6 days are investigated. Note that albeit the weather type on the chosen days is the same, the temperature, wind, and cloudiness conditions were quite variable. The basic question is whether the sensitivity of PBL height to soil dataset is valid also for ensemble runs which include the above-mentioned uncertainties. This question is analyzed and answered based on Fig. 10.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig10_HTML.gif
Fig. 10

Daytime course (5–17 UTC) of the spatially and 6-day-averaged relative PBL height differences obtained by using HU and US soil datasets in Hungary. The run conditions and the referring symbols are specified in Table 5

In early morning hours, there are practically no differences between the different runs. Around noon, the relative differences spread between 8% and 15%. Note that relative differences obtained by the MRF PBL scheme are always less than those obtained by the MYJ scheme. It is also obvious that the deviation of the curves obtained by using different radiation modules is much smaller than those obtained by different PBL schemes. The sensitivity to soil moisture content initialization is somewhat larger than the sensitivity to radiation module use, but somewhat smaller than the sensitivity to PBL scheme use. The spread of the curves is the highest sometime in mid-afternoon just before the beginning of the PBL collapse. Then, the highest relative differences amount to about 25%. This increment is caused by the time shift in simulating PBL collapse which exists between different runs. This ensemble run test shows that the rate of the curve scatter (about 6% around noon) is almost half of the typical value of the relative differences (about 10% around noon) in the greater part of the daytime period. This is especially true in the morning periods, when PBL height increases, but it is not valid in the time period around the beginning of the PBL collapse. As a conclusion, the sensitivity to soil parameter differences is greater than the sensitivity to model radiation and PBL parameterizations. Ensemble runs are also analyzed from the point of view of significant differences; the spatial distribution of the significantly different PBL heights was investigated. In doing so, the atmospheric and soil moisture content initialization effects are separately considered. Spatial distribution of the significantly different (p < 0.01) diurnal courses of PBL heights obtained by using HU and US soil datasets (PBLHU–PBLUS) and different radiation and PBL module configurations in 6 days is presented in Fig. 11. Since 6 days are simulated using two radiation modules and two PBL schemes, there is a total of 24 days of simulations.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig11_HTML.gif
Fig. 11

Spatial distribution of the significantly different (p < 0.01) diurnal courses of PBL heights obtained by using HU and US soil datasets (PBLHU–PBLUS) in 6 days, performing runs by combining RRTM and CCM2 radiation modules with Janjic and MRF schemes (a total of 48 (2 × 6 × 2 × 2) runs for 24 days (6 × 2 × 2))

Figure 11 indicates the relationship between the areal distribution of significant differences and soil textural classes. Note that the HU/US parameter differences increase going from sandy to clayey soils. So, the differences were significant in 18 and more cases (indicated with light shading) for almost all clayey soils. For loamy sand textural class, these significant differences appeared only on 12–14 days. Of course, the number of significantly different days is less than 10 in areas where there are cities and lakes. The spatial distribution of the significantly different (p < 0.01) diurnal courses of PBL heights obtained by using HU and US soil datasets (PBLHU–PBLUS) in 6 days for different soil moisture content initializations is presented in Fig. 12.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-012-0597-y/MediaObjects/704_2012_597_Fig12_HTML.gif
Fig. 12

Spatial distribution of the significantly different (p < 0.01) diurnal/daytime courses of PBL heights obtained by using HU and US soil datasets (PBLHU–PBLUS) in 6 days, performing runs by Janjic and MRF schemes increasing and decreasing the initial soil moisture content by 20% with respect to its reference value (a total of 48 (2 × 6 × 2 × 2) runs for 24 days (6 × 2 × 2))

As in the previous case, there are simulations for 24 days. The pattern obtained is very similar to the previous one. The relationship between the areal distribution of significant differences and soil textural classes can be unequivocally demonstrated. Above loamy sand areas, the differences were significant in 12–14 days; in some cases, only on 8–10 days. In south Hungary, between the Danube and Tisza rivers, the number of significant cases is also 12–14. There, the prevailing texture is sandy loam and sandy clay loam. All these results (Figs. 11 and 12) suggest that the investigated soil signal is strong enough with respect to other atmospheric and soil moisture initialization effects; that is, it can be observed irrespective of which model configuration is used.

6 Conclusion

The effect of soil hydraulic properties on PBL height changes is a rather old theme in numerical weather prediction (McCumber and Pielke 1981). Nowadays, forecasters endeavor to include the effect of soil texture, land use type, and parameterizations of soil properties on atmospheric events as accurately as possible. There are still open questions associated with the representation of soil hydraulic properties (Reen et al. 2006) which, regulating the surface energy budget, also greatly affect the PBL evolution.

Despite these efforts, there are still many blind spots regarding the effect of surface characteristics. One such area not investigated until now is the sensitivity of PBL height changes to soil database use. To analyze this effect, we simulated diurnal courses of PBL heights using the USDA-Cosby (Cosby et al. 1984) database on the one hand and the HUNSODA database (Nemes 2002) on the other, applying the MM5-Noah modeling system. Diurnal courses of PBL height differences obtained by using HU and US soil datasets (PBLHU–PBLUS) were significant not only for reference but also for ensemble runs when different radiation and PBL parameterization schemes and soil moisture initialization were used. The differences were significant to a greater extent above clay and clay loam, than above loamy sand and sandy loam textural classes. This is unequivocally caused by HU/US parameter differences which increase going from sandy to clayey soils (see Table 2. The results suggest that PBL height depends more on RASMC than on SMC, when the model is coupled to the Noah land-surface scheme. The RASMC is dependent on SMC and the Θf and the Θw soil parameters equally, or sometimes less on the SMC. The sensitivity of PBL height to variations of Θf and Θw is different for different soil textural classes. Θf changes seem to be more important for loamy sand, while Θw changes for clay (see Fig. 8). It is to be underlined that the analyzed sensitivity can only be observed when the shadowing effect of clouds is negligible. The appearance probability of this type of weather in a common year is about 20%, but the probability of the appearance of days without clouds over the domain is considered in the study much lower. In the cloudy-sky case, this sensitivity cannot be observed since PBL height changes caused by clouds’ shading effects are much larger than those changes which are caused by soil database use.

Acknowledgements

The research is supported by the Hungarian Scientific Research Fund (OTKA K-81432) and by the European Union and is cofinanced by the European Social Fund (grant agreement no. TÁMOP 4.2.1./B-09/1/KMR-2010-0003).

Copyright information

© Springer-Verlag 2012