Departmento de Impacto Ambiental de la Energía, Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas, Madrid, Spain

Transcription

1 880 JOURNAL OF APPLIED METEOROLOGY Simulations of Mesoscale Circulations in the Center of the Iberian Peninsula for Thermal Low Pressure Conditions. Part I: Evaluation of the Topography Vorticity-Mode Mesoscale Model FERNANDO MARTÍN, SYLVIA N. CRESPÍ, AND MAGDALENA PALACIOS Departmento de Impacto Ambiental de la Energía, Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas, Madrid, Spain (Manuscript received 27 September 1999, in final form 10 August 2000) ABSTRACT The Topography Vorticity-Mode Mesoscale (TVM) model has been evaluated for four different cases of thermal low pressure systems over the Iberian Peninsula. These conditions are considered to be representative of the range of summer thermal low pressure conditions in this region. Simulation results have been compared with observations obtained in two intensive experimental campaigns carried out in the Greater Madrid Area in the summer of The wind fields are qualitatively well simulated by the model. Detailed comparisons of the time series of simulations and observations have been carried out at several meteorological stations. For wind speed and direction, TVM results are reasonably good, although an underprediction of the daily thermal oscillation has been detected. The model reproduces the observed decoupled flow in the nighttime and early morning along with the evolution of mixing layer flow during the day. In addition, the model has simulated specific features of the observed circulations such as low-level jets and drainage, downslope, upslope, and upvalley flows. The model also simulates the formation of hydrostatic mountain waves in the nighttime in some cases. 1. Introduction The Greater Madrid Area is located in a 700-m high plateau at the center of the Iberian Peninsula. It is bordered to the north-northwest by a high mountain range (Sierra de Guadarrama), 40 km from the city, and to the northeast and east by lower mountainous terrain. The former and closest is about 200 km long and is aligned along the southwest northeast axis, with a mean altitude of 2000 m. The highest summit reaches up to 2400 m and is located 50 km northwest of the city. The climate in Madrid is somewhat extreme, typical of a continental area, with hot dry summers and cold winters, with most days being under clear-sky conditions. These topographic and climatological features along with a heat island effect contribute to complex mesoscale circulations and mixing conditions, which have an important influence on atmospheric pollution episodes. The geographical features of the Iberian Peninsula and its particular location in the Mediterranean area create specific meteorological conditions in which the ther- Corresponding author address: Fernando Martín, Grupo de Modelizacion de la Contaminacion Atmosferica, Dpto. Impacto Ambiental de la Energía, CIEMAT, Avda. Complutense 22, Madrid, Spain. fernando.martin@ciemat.es mal low dominates summer atmospheric conditions. In spite of the high frequency of this mesoscale pressure system in the south of Europe and in other parts of the world (Junning et al. 1984; Barry and Chorley 1987), it has hitherto received little attention from the scientific community, and a significant lack of thermal low and associated air circulation studies exist. Although the thermal low in Spain is more frequent in summer, it has also been detected at the beginning of autumn and even during the last days of winter near to the spring season (Font 1983). These infrequent cases are associated with weak synoptic conditions, long dry periods, and strong surface heat fluxes. Surface heating and atmospheric convective motions are the common characteristics in thermal low development. However, several peculiarities lead to a significantly different thermal low in the Iberian Peninsula, as compared with other countries. The particular horizontal and vertical sizes (less than 1000 km and 3000 m, respectively), as well as the high intensity and persistence of this mesoscale system, are also related to the geographical location of the Iberian Peninsula, which is almost completely surrounded by sea. Temperature differences between the air over the heated ground and air over the sea, along with mountain range orientation, produce strong air convergence, which is channeled by the main mountain valleys (Millán et al. 1991). A strong 2001 American Meteorological Society

2 MAY 2001 MARTÍN ET AL. 881 reduction of the inland surface pressure reaches a maximum in the early afternoon, when heating of the ground is most marked and when convective air cells are completely developed. When the solar energy begins to decrease, a slow dissipation of the thermal low takes place until it disappears during night hours. Therefore, the thermal low is a 24-h meteorological system clearly associated with intense solar radiation over the arid regions. Portela and Castro (1991) presented a climatic description of thermal lows in the Iberian Peninsula. They found the formation of thermal low pressure systems over the Iberian peninsula is very related to the deficit of evaporation in semiarid soils. It could explain why the thermal lows also can be observed in spring or early autumn but being less frequent than in summertime. They also made a high-resolution analysis of the pressure fields that allowed a classification of the thermal low systems taking into account the location of the area of maximum pressure gradient and the thermal low intensity. Extensive experimental documentation about the behavior of air convergence under summer thermal low conditions exists (Millán et al. 1991; Martín and Palomino 1995). These studies prove that the thermal low and sea breezes force an inward flow of coastal pollutant emissions toward the center of the Iberian Peninsula. The strong links between the local air circulation of the sea breeze and this mesoscale system are related to the particular orientation of mountain valleys near the Spanish coast that favors inland air motions. Moreover, the strong heating of the Iberian Peninsula soils can intensify the inland penetration of air masses. Experimental results have shown that sea-breeze penetration is significantly greater when air convergence associated with thermal low conditions interacts with the sea-breeze circulation (Martín and Palomino 1995). Furthermore, the thermal low can inject pollutant air masses to upper atmospheric levels, where they can then be transported long distances. Under thermal low conditions, the local air circulation over the central plateaus of the Iberian Peninsula, where the convergence zone is usually located, can be very different. Extensive instrumentation deployment over the Madrid area (Plaza et al. 1997) has allowed detection of significant differences of air circulations when thermal low conditions affect the Iberian Peninsula. These differences could explain the levels of pollutants detected in the center of Spain. Several works have been devoted to modeling the complete thermal low pressure system structure. In the case of large tropical thermal lows, Leslie (1980) incorporated a simple surface heat balance scheme into a large-scale numerical forecast model for Australia. The Gaertner et al. (1993) study of the Iberian thermal low consisted of a two-dimensional simulation with a hydrostatic, high-resolution, primitive-equation model. Three-dimensional simulations of the structure and flows of the Iberian thermal low system were done by Portela (1994) and Portela and Castro (1996) using the Pronóstico a Mesoscala (PROMES) model (Gaertner 1994) and by Ibarra et al. (1994) using the Regional Atmospheric Modeling System (RAMS) model. In both cases, working with almost the same spatial domain (the entire Iberian Peninsula), the main features of the thermal low were well simulated, but differences were found in predicting smaller-scale flows, such as circulations in the plateaus and in areas near large mountain ranges. In the PROMES simulation, the horizontal grid spacing was 20 km 20 km, and hence smaller-scale flows were better resolved than with the RAMS simulations, for which the horizontal resolution was 32 km 32 km. In contrast with these regional-scale modeling studies simulating the flows over the entire Iberian Peninsula, this paper is focused on the meso- -scale modeling in a smaller area of it. The current work is presented in two parts. Part I is presented in this first paper. The objective is to evaluate the performance of the Topography Vorticity-Mode Mesoscale (TVM) model in simulating the evolution of the meso- -scale atmospheric conditions in the center of the Iberian Peninsula under the forcing of a summer thermal low pressure system. To do this, the TVM predictions in the Greater Madrid Area are compared with wind and temperature observations from the surface and upper-air meteorological stations for four different cases of thermal low pressure situations. The paper includes a description of the TVM model, the cases selected for modeling, the data sources, and the model configuration, along with a discussion of the model results compared with observations. In Part II (Martin et al. 2001), the variability of polluted air parcel trajectories computed with the TVM model in the Greater Madrid Area under thermal low pressure conditions is discussed. The practical use of a mesoscale meteorological model requires understanding its characteristics and range of application. For the quantification of the accuracy of model results, it is also necessary to estimate input data accuracy and how it affects the results, to evaluate the uncertainties in model assumptions and parameterizations, and to judge how the model represents reality. The evaluation procedure will ensure that users can assess the degree of reliability and accuracy inherent in the model (Moussiopoulos 1996). The mesoscale prognostic TVM model has been evaluated previously to simulate the atmospheric flows for winter anticyclonic conditions in the center of the Iberian Peninsula. Model results agreed in significant aspects with observed wind flows over the Greater Madrid Area under anticyclonic conditions in wintertime, for example, the daily cycle of thermally driven flows, the displacement of the surface wind convergence line toward the south as a result of the influence of the synoptic flow on the mesoscale flow, and the model s prediction of two layers with very different flows (Martín et al. 1996). First simulations of the flow in the center of the Iberian Peninsula were carried out by Gaertner (1994)

3 882 JOURNAL OF APPLIED METEOROLOGY for several meteorological conditions with the PROMES model including two cases of the summer thermal low and a winter anticyclonic system (Martín et al. 1996). 2. TVM model The TVM model is a prognostic mesoscale meteorological model. It is a three-dimensional mesoscale vorticity-mode numerical model for atmospheric flows in complex terrain based on the Urban Meteorology (URB- MET) model (Bornstein et al. 1987). The URBMET model is a three-dimensional, hydrostatic, shallow-convection, Boussinesq (incompressible) model to simulate urban influences and sea-breeze fronts by use of the vorticity equations with both hydrostatic horizontal vorticity components and two streamfunctions. Several versions have been developed during the 1990s. Initially, the TVM model was an URBMET version for nonflat topography. Schayes and Thunis (1990) made the first reformulation by applying sigma-height coordinates. TVM uses Cartesian coordinates horizontally and a terrain-following coordinate in the vertical. A complete explanation of the formulation and descriptions of some applications of this version of TVM can be found in Schayes et al. (1996) and Bornstein et al. (1996). The version of the TVM model used in the current study (TVMNH20a) was developed by Thunis (1995): it is a nonhydrostatic, incompressible, and Boussinesq mesoscale model. However, TVMNH20a can be run in a hydrostatic mode, which keeps the two horizontal hydrostatic vorticity equations but does not include the additional terms for the nonhydrostatic formulation. The TVMNH20a model assumes two main layers: an atmospheric layer and a subsurface layer. The atmospheric layer is further separated into two sublayers: a constant flux surface that corresponds to the surface boundary layer (SBL) and a transition layer in which TVMNH20a uses a level-1.5 closure scheme in which turbulent kinetic energy (TKE) is computed from a prognostic equation involving advection, TKE shear production, buoyancy destruction/production, vertical diffusion, molecular dissipation, and horizontal diffusivity. Horizontal wind speed, potential temperature, and specific humidity in the SBL are assumed to obey Monin Obukov similarity scaling via the Businger forced and mixed convective functions. The infrared radiative fluxes have been computed from the scheme of Sasamori (1968), with carbon dioxide concentrations fixed at 320 ppm and with absolute humidity computed as a function of temperature, pressure, and specific humidity. A soil model based on the force restore method (Deardorff 1978) is included in TVMNH20a to compute surface temperature from the surface soil heat flux and the temperature of the lower soil layer, assumed constant within each soil class. The surface soil heat flux is obtained using a residual method from the surface energy balance equation (Schayes 1982; Schayes et al. 1996). Soil moisture is computed by a prognostic equation, which depends on the SBL latent heat flux and surface latent heat flux, computed by use of the Penman Monteith formulation (Monteith 1981). This formulation accounts for dynamic and vegetation effects by use of aerodynamic and surface resistances that are constant for each land use type. Remaining boundary conditions are as follows. 1) At the upper boundary, the wind is geostrophic, vorticity is zero, and temperature and humidity values match synoptic-scale values. 2) A filter is used in the uppermost computational levels to smooth all prognostic variables, except TKE, at each time step to avoid reflections of vertically propagating gravity waves (Schayes et al 1996). 3) At lateral boundaries, zero-gradient (open) boundary conditions are assumed. 4) At the surface, stream functions are zero and TKE 2 is fixed at 4 u * (Therry and Lacarrère 1983), where u * is friction velocity. How TVMNH20a solves the PBL hydrodynamic and thermodynamic transport equation at each time step is described in Schayes et al. (1996). The elliptic equations of streamfunctions and vorticity are solved via a modified biconjugate gradient method, and advection is solved by the third-order parabolic piecewise method. The model is only applicable (i) to the meso- -scale, because horizontal synoptic-scale variations cannot be considered; (ii) to steady synoptic conditions; and (iii) to clear-sky, light-wind, and weak pressure-gradient meteorological cases, (iv) with no initial direction shear. Other versions of the TVM model have been evaluated for a number of case studies and geographic locations (e.g., Athenian Photochemical Smog Intercomparison of Simulations air quality study of Athens, Greece; Fos, France, experiment; and New York sea breezes, Boulder windstorm, and Madrid, Spain, experiments), and compared with other mesoscale meteorological models such as RAMS, MEMO (Mesoscale Model), PROMES, and MAR (Model Atmospherique Regional) (e.g., Thunis et al. 1993; Bornstein et al. 1993, 1996; Martín et al. 1996, 1997). 3. Cases studied In the framework of the Regional Cycles of Atmospheric Pollution in the Mediterranean Area (RECAP- MA) project, the atmospheric pollution team of the Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas (CIEMAT) carried out two experimental campaigns during the summer of 1992 to study the dynamics and chemistry of air pollution in the Greater Madrid Area. RECAPMA field campaigns lasted three days during July and September of Meteorological conditions during these periods consisted of thermal low pressure systems over the Iberian Peninsula on five days (14 16 July, and 15 and 17 September). During

4 MAY 2001 MARTÍN ET AL. 883 the afternoon of 15 July, thunderstorms occurred, and hence this day was rejected for simulations because the TVMNH20a model assumes clear-sky conditions and lacks any treatment of clouds or moist processes. Hence, the selected days for simulations were 14 and 16 July and 15 and 17 September. Surface weather charts at 1200 UTC (Fig. 1) show the thermal low pressure system in the final phase of development. The center was located over the Iberian Peninsula because of the strong heating of the semiarid land of the peninsula in contrast to the surrounding sea and ocean. The low pressure system was clearly formed at 1200 UTC on 16 July and on 15 and 17 September. On this last day, there was also an interaction with a frontal system that was going over the northern coast of the peninsula. Synoptic charts at the 850- and 500- hpa levels only showed low pressure systems over or close to the Iberian Peninsula on 16 July at the 500- hpa level (when a weak cold low pressure system from 15 July appeared) and 17 September (when a weak trough affected northern regions of the peninsula). The anticyclone was well developed at 850 and 500 hpa on 15 September. Although the thermal low pressure systems over the Iberian Peninsula appeared to be similar, there are significant differences in the upper atmosphere among the four selected days. In this study, the selected days thus cover a wide range of atmospheric conditions under the label of a summer thermal low pressure system. 4. Input data and model configuration The geographical domain corresponds to an area of 340 km 310 km in the center of the Iberian Peninsula that covers most of the Tajo Valley, the mountain range called Sistema Central (which includes Sierra de Gredos in its western area, Sierra de Guadarrama in its center, and Sierra de Ayllón in its northeastern area), Montes de Toledo, and part of the Northern and Southern Plateaus (Fig. 2). The Greater Madrid Area is close to the center of the domain, between the two ranges. It is a large conurbation extending about 50 km 50 km and including the city of Madrid and several satellite towns. The main input data of TVMNH20a consist of: 1) Topography: Data for a mesh of cells with a spacing of 10 km 10 km were obtained from a digital terrain model (CNIG 1995) for Spain with a spacing of 1 km 1 km (Fig. 2). These data were smoothed by applying a Shapiro filter to reduce slopes to a maximum of 6% and to filter out 2 x waves, where x is the length of a grid cell. 2) Land use data: These data have been obtained by processing the CORINE land use data for Spain (I. Rábago and M. Sousa 1995, personal communication) with a 5 km 5 km spacing (Fig. 3). CORINE is the acronym of coordination of the Collection of Information on the State of the Environment. Land use data were translated to the same mesh as the topography data. The eight different land use types and their associated physical parameters are shown in Table 1. Surface and (10 cm) deep soil temperatures are listed in Table 2. Because observations were unavailable, deep soil data were obtained by trial and error, using values within climatological ranges. 3) Meteorology: Synoptic wind speed and direction must be imposed as top boundary conditions. In addition, initial potential temperature and wind speed profiles, along with surface specific humidity data, are needed. All of this information has been obtained from wind profiles measured from free soundings carried out by CIEMAT and those of the National Institute of Meteorology (Instituto Nacional de Meteorología: INM, hereinafter). Specified synoptic winds need to account for the meteorological situation modeled. Deep anticyclonic or cyclonic situations usually affect the entire troposphere, but the Iberian thermal low is usually embedded in a subtropical anticyclone. Air masses affected by the strong buoyancy in the center of the Iberian Peninsula can be frequently injected up to 3000 m above ground level (AGL) under thermal low pressure conditions. Therefore, while the lowest 3000 m deep layer of atmosphere is affected by the thermal low system, the anticyclonic conditions remain in the middle and upper troposphere. In this case, it is clear that the synoptic wind data (which are used as upper boundary condition) have to be obtained from altitudes no higher than 3000 m AGL using soundings launched at noon or the early afternoon (see Table 3). The initial wind and temperature profiles were obtained by analyzing smoothed radiosonde soundings from two locations close to Madrid to obtain an initial sounding for each simulated case. These profiles were extended to the entire domain to represent the initial state of the atmosphere for each case. In Fig. 4, the initial meteorological profiles are depicted up to m in potential temperature and 4500 m in wind to show clearly the initial conditions in the lower layers where the vertical changes are sharper. 4) Model configuration: Although the model can use nonregular or stretched grids with variable resolution, for the current simulations a regular grid was selected. Details about the miscellaneous aspects of input data are given in Dutrieux (1997). The number of grid points currently selected was in the horizontal, with 24 vertical levels. The top of the domain was set at m above mean sea level (MSL). The vertical resolution was variable, with the minimum separation at the bottom levels (20 m) increasing with altitude. The filter to smooth all the prognostic variables, except TKE, is applied to the five uppermost computational levels (i.e., above m MSL). The simulations cover 37 h starting

5 884 JOURNAL OF APPLIED METEOROLOGY FIG. 1. Meteorological charts from the Spanish National Institute of Meteorology for four selected cases.

6 MAY 2001 MARTÍN ET AL. 885 FIG. 2. Modeled area showing main geographical features. FIG. 3. Land use types.

7 886 JOURNAL OF APPLIED METEOROLOGY TABLE 1. Physical parameters associated with land use types. Soil heat capacity per area unit was estimated assuming the depth of upper soil layer is equal to 10 cm. Here, z 0 is roughness length. Land use type Albedo Emissivity Surface resistance to evaporation (s m 1 ) Soil heat capacity per area unit (J K 1 m 2 ) z 0 (m) 1 Water reservoir 2 Urban area 3 Herbaceous and shrubs 4 Forest 5 Olive trees and groves 6 Orchards and vineyards 7 Cropland 8 Pasture at 1200 UTC [in Madrid, the difference between universal coordinated time (UTC) and local mean solar time is about 12 min]. The TVMNH20a model computes the suitable time step for each resolution, but an upper limit is imposed; for these cases, 30 s was considered. 5. Data used in model evaluation Six meteorological 10-m towers were deployed in the region of Madrid (Fig. 5). Wind speed and direction at 10 m AGL and temperature at 1.5 m AGL were automatically recorded every 10 min at all stations except the one installed at the CIEMAT headquarters (CI). This one was equipped with temperature sensors at 10 and 80 m AGL and with wind speed and direction sensors at 80 m AGL. Data (averages, standard deviations, vertical gradients, atmospheric stability, etc.) from this station were recorded every hour. Observed wind data from the 14 INM synoptic stations were recorded every 3 h, but, unfortunately, gaps during the night are frequent. Several free and tethered soundings were launched daily by the CIEMAT team at the Villanueva site. Details about the experimental deployment are in Crespí et al. (1995). These data were used to evaluate the performance of the model in its simulation of the vertical structure of the lower troposphere, especially the boundary layer. TABLE 2. Data of soil temperature (K) for every scenario. Here, T g is the soil surface temperature, T g1 is the surface temperature of water areas, and T 2 is the temperature at the bottom boundary of the soil model. T g (K) T g1 (K) T 2 (K) 14 Jul 16 Jul 15 Sep 17 Sep Simulations The nonhydrostatic TVM model (version TVMNH20a) has been run to simulate the four cases described in section 3. Comparison between model results and observations are presented from the analysis of surface (10 m) wind fields, wind and temperature time series from surface stations, and vertical profiles of wind and temperature at the Villanueva site. a. Surface wind patterns 1) 14 JULY The wind field is influenced by a 150 synoptic wind with a speed of 4 m s 1. By 0300 UTC (hereinafter, all time references will be UTC), 15 h after the start of the model simulation (Fig. 6), drainage flows are notable in many areas of the domain (e.g., southeasterly flows over the Northern Plateau and northeasterly flows over in the Tajo Valley and Greater Madrid Area). Another important feature of the wind field is the intense downslope flow observed in the northern slope of the mountain range, which is related to the flow acceleration observed when air is flowing over the ridge in stable conditions, giving rise to the formation of mountain waves, as can be seen in the vertical cross-sectional results (see section 7). In addition, because the synoptic flow has a northward component, southward-directed katabatic flows on southern slopes are weak, except those on the steepest southern slopes of Sierra de Gredos. Most observations are in the Greater Madrid Area and its surroundings. In this area, the simulated wind fits the observations well. The rest of the domain has sparse coverage at this hour. By 0900, the weak land warming is enough to weaken the strong southeasterly downslope flows on the northern slope, but weak southeasterly flows remain in the Northern Plateau. In the Greater Madrid Area, winds are generally weak and start to rotate clockwise to easterly or southeasterly flows because of the heating of the southern slopes of the Sistema Central. These features are confirmed by the observations. By 1500, upslope winds are dominant on the southern slopes, but no clockwise rotation to westerly or southwesterly flows (in contrast with the other cases simulated) is observed in the Greater Madrid Area and in Tajo Valley. This situation probably is due to the forcing of the southeasterly synoptic flow produced by the lo-

8 MAY 2001 MARTÍN ET AL. 887 TABLE 3. Meteorological inputs. Surface potential temperature (K) Surface specific humidity (10 3 kg kg 1 ) Synoptic wind direction ( ) Synoptic wind speed (m s 1 ) 14 Jul 16 Jul 15 Sep 17 Sep cation of the thermal low center over the southwestern area of the Iberian Peninsula (Fig. 1). On the northern slopes of the Sistema Central, the upslope wind is inhibited, in agreement with the observations. It probably is due to a downward momentum transfer from aloft in a well-mixed layer (Atkinson 1989). It can also be the reason that southeasterly flows are also dominant in the Southern and Northern Plateaus. By 2100, TVM shows that upslope flows have disappeared, which is confirmed by the observation at the station on the southern slope of the Sistema Central. TVM results also simulate weak winds in the west and north of the Greater Madrid Area, but not in the east and southeast. This result agrees with the observations. On the northern slopes of the Sistema Central and Northern Plateau, southeasterly flow is again dominant and shows a clear tendency toward the nighttime flow shown at ) 16 JULY Although some similarities exist between the simulated wind fields for 16 July and those for 14 July, some differences exist (Fig. 7). On 16 July, the synoptic wind blew from 190 with a speed of 3 m s 1. The first panel of Fig. 7, at 0300, shows a considerable difference in the flow over the southern slope of the Sistema Central as compared with the former case. Drainage flow on the very steep southern slope of Sierra de Gredos is almost inhibited, because the synoptic wind is nearly opposite to the katabatic flow. By 0900, almost no differences exist between the estimated wind fields for both days. As for the 14 July case, a downward momentum transfer in the mixing layer can be the reason that upslope flows in the lee side of the Sistema Central do not appear and southerly and southwesterly flows are dominant in most part of the domain at By 2100, a weak flow with some drainage in the Greater Madrid Area and in Tajo Valley exists. The strongest winds are blowing from a southerly direction on the northern slopes (downslope winds) and on the Northern Plateau. Modeled drainage flows in the southern slopes, just west and north of the Greater Madrid Area, are stronger than those simulated for 14 July as confirmed by the observations. 3) 15 SEPTEMBER Under the thermal low conditions of 15 September, the synoptic wind is from the northwest with a speed of4ms 1, in contrast to the two former cases. The wind field pattern is correspondingly different, especially over the Sistema Central (Fig. 8), where northwesterly flows are found during most of the entire day, although oscillations are detected. During the night, downslope winds on the southern slope are strong (more FIG. 4. Initial meteorological profiles used in simulations.

9 888 JOURNAL OF APPLIED METEOROLOGY FIG. 5. Simulation domain showing locations of meteorological stations and topographic height contours. than in any other case) and can also be observed at the Hoyo station. In this case, the effect of leeside accelerations of downslope flows was more significant than in other cases. These are observed in the wind fields at 0300 and As demonstrated in section 7, mountain waves also are formed. Bear in mind that sunset in September is about 1.5 h earlier than in July. Thus, wind field features at 2100 are closer to nighttime flow patterns than are those in July. In addition to the wind over the Sistema Central, the flow over the Greater Madrid area and the Tajo Valley consists of weak northerly and northeasterly drainage flows at night. A convergence zone is detected in the narrower western area of the Tajo Valley, where interaction of the downslope flows is strong. Unfortunately, no observations exist in this area to confirm this feature. By 0900, northwesterly downslope flows on the southern slopes of the Sistema Central become weak, drainage flows almost disappear, and upslope winds start in both slopes of the Sistema Central with a convergence line sited over the southern slopes but close to the crests. After midday, (by 1500), the strong daytime flow pattern is completely developed, and westerly and southwesterly flows are dominant in the Tajo Valley and the Greater Madrid Area. In contrast, the flow over the mountains and plateaus is northwesterly. The convergence line parallel to the Sistema Central was moved southward about 20 km with respect to its position at 0900 by the forcing of the northwesterly synoptic flow. This southward movement of the convergence line continues progressively during the daytime. This convergence line usually appears just over the peaks of the range in simulations when a zero synoptic wind speed is assumed (not shown here). Some similar features were observed in simulations of flows for winter conditions (Martín et al. 1996). The speed of this movement may depend on the component of the synoptic wind vector perpendicular to the Sistema Central and on the intensity of solar heating, as seen in other simulations (not shown here). 4) 17 SEPTEMBER The 260 synoptic wind blew with a speed of 4 m s 1 on 17 September. As in the former case, the northwesterly downslope flow is important on the southern slopes of the Sistema Central (Fig. 9), but the wind speed is lower and wind direction has an important west component at The downslope wind on the northern slopes is also significant, but less so than on the southern slopes. The effect of the direction of the synoptic wind, almost parallel to the Sistema Central, is clear in the simulated flow. There are also important downslope winds on the northern slopes of the Montes de Toledo, and significant southwesterly flows on the Northern Plateau. In contrast, the wind is very weak in the Greater Madrid Area and in Tajo Valley, also confirmed by observations. Later in the morning, (by 0900), downslope flows have weakened significantly, upslope winds are starting, and the flow in the Tajo Valley is clearly from the southwest. By 1500, the southwesterly flow is intense in the entire domain, as evidenced by observations and in the

10 MAY 2001 MARTÍN ET AL. 889 FIG. 6. Simulated wind fields (gray arrows) and observed winds (black arrows) for 14 Jul at 10 m. simulated wind field. This general southwesterly flow over the entire domain did not occur in the other cases. It is likely that the 260 synoptic wind has favored this flow, but perturbation of the mesoscale flows is also important. On the one hand, the southwesterly upvalley flow in the Tajo Valley and in the Greater Madrid Area is stronger than in any other case. On the other hand, the wind direction is disturbed by the Sistema Central, where a convergence area appears. However, the model does not fit the observation in the northern slope of the Sistema Central where weak upslope flows are observed in contrast with the modeled southwesterly flows. In this case, the model probably overestimates the downward transfer momentum in the mixing layer, which implies a destruction of the upslope flows in the lee sides. At night, by 2100, intense southwesterly flows remain in some areas, but they either weaken or their flow direction changes, as on the southern slopes of the Sistema Central. In this area, the wind blows from the west or northwest (downslope). It produces a convergence area that moves southward, reaching the Greater Madrid Area and the Tajo Valley, as the daytime southwesterly flow weakens and the downslope winds accelerate. b. Surface wind comparison Time series comparison between observed and modeled (10 m) surface wind speed and direction at the six stations deployed by CIEMAT (Fig. 5) during the experimental campaigns was carried out to check the performance of the model in the area surrounding Madrid.

11 890 JOURNAL OF APPLIED METEOROLOGY FIG. 7. Same as Fig. 6 but for 16 Jul. All observed data are at 10 m AGL, except the CI station, which is at 80 m AGL (80-m wind simulations were used for this comparison). The Hoyo (HY) station (northern area of Madrid region) is representative of flows close to the slopes of the Sistema Central, but it is on the southern slope of a small hill. The San Martín (SM) station is located south of Madrid in a small valley with a north south orientation, and the El Encín (EN) station is east of the Greater Madrid Area close to a small ridge oriented northeast southwest. Last, the Villanueva (VI) station is west of the Greater Madrid Area, and the Majadahonda (MJ) and CI stations are representative of the flows influenced by the Madrid urban area. The CI station is located close to the summit (500 m away) on the western slope of a 100-m-high hill with an inclination of about 4%. 1) 14 JULY Wind direction is well simulated of most stations, especially the time of changes in flow direction in the early morning (Fig. 10). The most important differences are observed at the EN and SM stations, where effects caused by the local small-scale topography cannot be simulated by the smoothed topography of the simulations. For wind speed, overprediction occurs at the EN, SM, and VI stations, but the speed is well simulated at the remaining three stations located close to the Sistema Central or in the Greater Madrid Area for the nighttime period ( ). During morning, the most accurate speed predictions are at the EN, SM, and HY stations, and an underprediction is observed at the remaining stations, especially CI. In the afternoon, the wind speed

12 MAY 2001 MARTÍN ET AL. 891 FIG. 8. Same as Fig. 6 but for 15 Sep. is well simulated at every station, but after 1800 it is underpredicted at EN, MJ, and CI even when wind direction is simulated well. As seen in the evening wind fields (Fig. 6), an area of more-intense southeasterly flow is modeled just over and southeast of the Greater Madrid Area, where the EN and CI stations are located. However, the strong winds observed at the CI station can be due to the influence of the 100-m-high hill on whose western slope it is located. This hill is relatively small and is not represented in the 10 km 10 km topography of the simulations. 2) 16 JULY Unfortunately, because of the lack of observed data records, comparisons between observed and simulated wind through the day can only be done for three stations (Fig. 11). During this day, TVM results fit the wind speed and direction evolution relatively well at the HY station. Observed data for the CI station show a wide variability, especially during the daytime. In the early morning, observed winds in this station are mostly from the northeast as modeled. Then, an abnormal variation in the observed wind direction occurs, with some persistent northeast wind around noon, in contrast to the progressive evolution given by the model. In the afternoon, the observed wind is from the west or southwest, in contrast to the modeled south or southwest wind. Slight overprediction in the nocturnal wind speed also occurs. A similar behavior, but less erratic in the morning, is observed for the VI station. In spite of the lack of observed data at the other stations, it might be con-

13 892 JOURNAL OF APPLIED METEOROLOGY FIG. 9. Same as Fig. 6 but for 17 Sep. cluded that the morning wind evolution is not well simulated by the model in the Greater Madrid Area, and it seems that the northeasterly drainage flow is also overpredicted in this area. However, the wind evolution in areas close to the mountains is simulated well. 3) 15 SEPTEMBER Wind direction is simulated well by the TVM model for the nighttime (Fig. 12). The diurnal evolution is also simulated well at the EN and VI stations, at the HY station in the early morning and late afternoon, and at the CI station in the afternoon. The main differences are detected at the SM station, because the local circulations in its small valley cause significant disturbances on the mesoscale flow. The variability observed during the morning and early afternoon at the CI station is not reproduced by the TVM simulations. At the HY station, TVM gives a clockwise rotation of the wind direction from a 140 to 260 direction during the daytime, but the observed wind blows from a 150 to 190 direction, as in the July simulations. In this case, the HY station is an area of strong spatial variability in the simulated wind direction, because it falls in the modeled convergence area (see wind field discussion). Wind speed is generally simulated accurately, although underprediction of early morning wind speed at all stations has been found. At that time, wind speeds higher than 2 m s 1 were observed, but the model simulates almost calm wind. On the other hand, an overprediction was detected at Villanueva at nighttime and, to a lesser extent, during the day. During the late af-

14 MAY 2001 MARTÍN ET AL. 893 FIG. 10. Simulated temperature and wind velocity vs observed values for 14 Jul. ternoon or evening, some overprediction also exists at the SM station. The most important differences were found at the CI station, where a significant underprediction (about a factor of 2) is observed at night, and wind speed is overpredicted (by about a factor of 2) during the afternoon and evening. At midnight, the observed wind at CI is strongly accelerated, blowing from the northeast and reaching speeds of 7 m s 1. This acceleration can be related to a local perturbation of the northerly and northeasterly mesoscale drainage flows. This local perturbation could result from the effect of the hill on which the station is located. In addition, a good fit of model results to observations at the HY station on the southern slope of Sistema Central is seen at night, when the northwesterly downslope flow becomes important. 4) 17 SEPTEMBER The TVM model simulates fairly well the complete daily cycle of wind direction at every station (Fig. 13). However, the modeled change of wind direction is advanced 2h at the VI, EN, and SM stations. At the EN and SM stations, results are better for the daytime than

15 894 JOURNAL OF APPLIED METEOROLOGY FIG. 11. Same as Fig. 10, but for 16 Jul. at night, because of local drainage winds at night that were not simulated by the model. Observed and modeled wind speeds generally coincide with the general pattern of the daily cycle, which consists of very weak winds at night and strong accelerations during the day. Differences exist, however, in the magnitude of the maximum wind speeds. In all cases, the observed wind speed maximum occurs about 1500 and it is slightly underpredicted. At the VI station, the underprediction is higher (the maximum observed wind speed was 9.4 m s 1 at the VI station, and the model result was 5.2 m s 1 ). c. Surface temperature Simulated surface temperatures were compared with observations. Temperature was measured at 1.5 m AGL at all of meteorological stations, except for the CI station where it was at 10 m AGL. Model results for the 1.5-m- AGL temperature were obtained by interpolating between TVM results for the two lowest temperature levels, taking into account the formulation used in TVM for the SBL. Maximum temperature simulations fit the observations in many cases, especially the July cases (Figs. 10 and 11). Nevertheless, underprediction is notable at some stations, especially for the 15 September case (Fig. 12). In this case, the maximum temperature is underpredicted by 2 C at the VI, HY, and CI stations. The extreme underprediction (about 7 C) is at the SM station. In contrast, the nighttime temperature is generally overpredicted (especially for the September cases), except for the CI station. The amplitude of the daily oscillation of temperature is mostly underpredicted, especially for the September cases (Figs. 12 and 13). However, for the case of 15 September, the amplitude of measured daily thermal oscillation is very high, reaching 33 C at the SM station (much higher than at other stations), in contrast with the average thermal oscillation observed for the other studied cases. Best results are for the CI station, where observations and model results correspond to 10 m AGL. The cause of the more notable underprediction of the daily thermal oscillation for the September cases might be changes in the soil conditions after all the summer season. The specific results for the SM station (in which the daily thermal oscillation is always much higher than the other stations) might be due to the differences between the local soil where the station were installed and the land use type selected as representative for their 10 km 10 km cells. Modeled and observed daily cycle of temperature are synchronous for all stations, except VI and CI, where maximum temperature is observed later (about 1700). d. Vertical profiles Vertical TVM profiles of potential temperature and wind speed and direction have been compared with observed

16 MAY 2001 MARTÍN ET AL. 895 FIG. 12. Same as Fig. 10 but for 15 Sep. profiles taken from free soundings launched at the Villanueva station (Fig. 14). Two sets of profiles have been used to check the ability of the model to simulate the mixing layer growth. The first set of profiles correspond to early morning, and the second set is representative of noon or early afternoon. Observed profiles represent local atmospheric conditions at a certain moment of the day, but model results are a smooth representation of atmospheric features over a 10 km 10 km cell. Moreover, the lower vertical resolution of the TVM results as compared with the high detail of the observations increases differences between observed and modeled variables. Deepening of the mixing layer is simulated well by the model. Some discrepancies are detected in a datumto-datum comparison, because sometimes 2- or 3-K differences have been observed between observations and model results, especially in the early morning profiles. However, the stable layer height is simulated well by the model. The model overpredicts the mixing layer heating in early morning, except in the case of 17 September. This overprediction is connected to the slight overprediction in the surface temperature at the Villanueva station (Figs ). At this station, observed temperature growth is less intense than is modeled, giving a maximum temperature about 2h after the simulated maximum. Early-morning wind profiles show a more complex situation because of the flow decoupling observed between the mixing layer and upper level. In profiles of around 0700 or 0800 UTC (see profiles for 14 and 16 July in Fig. 14), some nocturnal characteristics remain, such as northeasterly drainage flows in the mixing layer well distinguished from the flow in the upper level. This feature is simulated well by the model. In the profiles made later (0930 UTC, especially for the 17 September

17 896 JOURNAL OF APPLIED METEOROLOGY FIG. 13. Same as Fig. 10 but for 17 Sep. case), modeled wind profile shows the start of the southerly upslope flows, which fits well with the observations. The upper-level results also fit the observations, but to a lesser extent for the 15 September case, when wind direction change is simulated to occur in a lower level than that observed and wind speed is underpredicted. In the late morning profile of the 14 July case, wind speed and direction are simulated well by the model in the mixing layer, but wind speed is underpredicted in the upper level. Best agreement between observations and model results occurs in the early afternoon profiles (see cases of 16 July and 15 and 17 September in Fig. 14), when the mixing layer is almost completely developed. 7. Cross-sectional results The y z plane distributions of potential temperature perturbation, vertical wind speed, TKE, wind speed, and wind direction have been computed for the four above scenarios. The plane runs from the south to the north and crosses the x Universal Transverse Mercator (UTM) coordinate at 449 km (see Fig. 3). The selection of this plane was done taking into account the Greater Madrid Area location and that it should be better to exclude peaks of the mountain range in order to have an average representation of mountain range effects. The selected cross section is just on the east side of the Greater Madrid Area and cross a saddle between peaks in the ridge. a. 14 July The synoptic wind blows from 150 with a speed of 4.0ms 1. The initial vertical temperature gradient is 2.5Kkm 1 from 1600 to 4100 m MSL. During the night, a hydrostatic mountain wave is observed in north-

18 MAY 2001 MARTÍN ET AL. 897 FIG. 14. Simulated and observed vertical profiles of wind and potential temperature. ward direction, as is clearly seen in (i) the undulation of the isentropic lines with propagation vertically but not horizontally, (ii) the distribution of updrafts and downdrafts, and (iii) the acceleration of the flow on the northern slope (Figs. 15a,c). This result is expected, because simulated stability is relatively strong in the layer between 1500 and 3000 m MSL, the synoptic winds are also weak but with a significant perpendicularto-ridge component, and the ridge is very wide. Under these conditions, buoyancy is dominant enough that the

19 898 JOURNAL OF APPLIED METEOROLOGY FIG. 15. Simulated values of perturbation (from initial surface state) of potential temperature (K; dotted lines) and vertical wind speed (m s 1 ; solid and dashed lines represent positive and negative values, respectively) at (a) 0300 and (b) 1500 UTC including (shaded) turbulent kinetic energy (J Kg 1 ). Simulated (shaded) horizontal wind speed (m s 1 ) and direction ( ) corresponding to (c) 0300 and (d) 1500 UTC. All information is in north south cross-sectional plane corresponding to a UTM x coordinate of 449 km for 14 Jul. vertical accelerations are inhibited and the flow is hydrostatic. The maximum vertical wind speed is about 0.1ms 1. The amplitude of the isentropic undulation is about 300 m. The maximum horizontal wind speed was 5.0 m s 1. In addition, katabatic flows are observed on the southern slope and from the northeast direction in the Greater Madrid Area and Tajo Valley. In both cases, katabatic flows are in a shallow layer, about 200 m thick, in agreement with depth of the surface inversion layer at 0300 UTC. During the day, differential heating of the surface produces thermally driven circulations with weak updrafts and downdrafts in the mountains. The nighttime mountain waves have disappeared. The turbulent kinetic energy can be strong, reaching almost 3.0 J kg 1, and the maximum mixing layer depth is about 2000 m (Fig. 15b). Flows are mainly from 120 to 150 in the Greater Madrid Area and Tajo Valley. Maximum wind speed (about 4.0 m s 1 ) is over the Sistema Central, as can be seen in Fig. 15d. b. 16 July The synoptic wind blows from 190 with a speed of 3.0ms 1, and the vertical temperature gradient is 2.3 Kkm 1 from 2000 to 4200 m MSL. In this simulation, mountain waves are not detected, because neither perturbation in the isentropic lines nor significant vertical flows are observed (Fig. 16a). In this case, the predicted

20 MAY 2001 MARTÍN ET AL. 899 FIG. 16. Same as Fig. 15, but for 16 Jul. stability between 1500 and 3000 m MSL is very weak. Most circulations are katabatic flows along mountain slopes and drainage flows over the Greater Madrid Area and the Tajo Valley. Katabatic flows on the northern slopes are favored by the synoptic wind (Fig. 16c). The northeasterly flow over the Greater Madrid Area and Tajo Valley is only in a 200-m-thick layer affected by a thermal inversion, similar to that estimated for the 14 July simulation at 0300 UTC. In the daytime, updrafts appear over the Greater Madrid Area (Fig. 16b). This could be related to some effect of either the high roughness or differential heating (i.e., convection) in the urban area with respect to the rural surroundings. The former hypothesis is related to the flow deceleration observed in the surface horizontal wind fields estimated by the model over the Greater Madrid Area at 1500 (see Fig. 7). This latter hypothesis runs counter to observations, because the difference in 1.5-m temperature between urban and rural regions in the Greater Madrid Area is less during the late morning and early afternoon (Crespí and Artíñano 1995). However, these updrafts may be specific to this case. Further studies must be done to determine the cause of this effect. The TKE patterns are influenced by many factors, such as TKE production by buoyancy or wind shear and TKE advection. Generally the effects of these factors are superimposed, and it is difficult to distinguish one from the others. However, for the 16 July case, some of them can be distinguished. In Fig. 17, the evolution of the TKE patterns can be observed. The TKE production by buoyancy is reflected by the general increase of TKE during the morning and early afternoon before Then, the TKE destruction starts. At 1000, the