Journal of Geophysical Research: Atmospheres

Nocturnal convective cloud formation under clear-sky conditions at the eastern Andes of south Ecuador



[1] The formation of nocturnal convective clouds at the eastern Andes of south Ecuador and the adjacent Peruvian Amazon basin was investigated in a numerical model study. Their formation is expected to be an interactive procedure of nocturnal downslope flows in the Andean terrain, which forms a concave drainage system in the target area. Satellite imagery were used for both the identification of a sample case with a nocturnal cold cloud appearance and for the verification of the simulated results. The cloud patterns were distinguished on the basis of IR temperatures. A comparison of the data demonstrated the occurrence of a cold cloud shield in the target area, although the modeled cluster is significantly smaller. Further analysis of the development of the convective cells confirmed the assumed underlying processes. A strong current in the lower atmosphere, presumably a drainage flow, was recognizable in association with strong moisture convergence using a cross section through the cluster. Their presence was confirmed on the basis of their characteristic features and the surface energy fluxes as the driving force for thermally induced downslope flows.

1. Introduction

[2] The formation of convective clouds in the tropics over land is dominated by the diurnal cycle of solar radiation with a rainfall maximum in the late afternoon [Mapes et al., 2003; Poveda et al., 2004]. In regions such as south Ecuador, which is located in the northwest of South America, the diurnal course of precipitation is modified by the complex terrain of the Andes. The high mountains induce modifications in the atmospheric circulation (e.g., effects of windward and lee side) and regional circulation systems such as mountain-valley breezes. This results in higher spatiotemporal variability in rainfall occurrence [Rickenbach, 2004; Sato et al., 2009]. In our study area, the eastern Andes of south Ecuador (see Figure 1), precipitation measurements with automatic climate stations and local area weather radar (LAWR) revealed an unexpected early morning rainfall peak at the Estacion Cientifica de San Francisco (ECSF at latitude 3°58′18″ S, longitude 79°4′45″ W, altitude 1860 m asl) [Bendix et al., 2006]. An analysis of cloud-top temperature distributions from corresponding Geostationary Operational Environmental Satellite (GOES) imagery showed the nocturnal occurrence of mesoscale convective systems (MCS) in the region of the Andean foothills southeast of southern Ecuador in the Peruvian Amazon basin [Bendix et al., 2009], which might be related to the formation of nocturnal rainfall. The generation of the MCS was hypothesized to be a consequence of an interaction between nocturnal drainage air from the Andean slopes and valleys and the warm moist air of the Amazon basin. In doing so, the katabatic flows act like a local cold front, resulting in a destabilization of the lower atmosphere.

Figure 1.

South Ecuador and the adjacent Peruvian Amazon basin with (left) the nested domain configuration, (upper right) GOES-E image and location of the LAWR, and (lower right) terrain configuration of the study area.

[3] The correlation between nocturnal convective clouds and low tropospheric flow systems has already been demonstrated by other authors. Lopez and Howell [1967] was the first to reference katabatic flows in the tropical Andes and describe the effects of katabatic winds on the eastern slopes, as well as an associated convection initiating hydraulic jump. Garreaud and Wallace [1997] observed that nocturnal rainfall presumably occurs due to enhanced low-level convergence resulting from nocturnal circulation between the Andes and the Amazon region. Angelis et al. [2004] described the convergence of cold air drainage from the Andes and warm, moist air from the Amazon, where the katabatic flows induce low-level instability by acting as a cold front.

[4] Although several phenomenological studies have been conducted on MCS formation and the correlation between convective cloud and cold air drainage flows, evidence for MCS formation in the target area of southern Ecuador and northern Peru induced by nocturnal cold air drainage flow is hitherto lacking. In this context, Trachte et al. [2010] investigated the impact of the terrain on the dynamic behavior of thermally induced katabatic flows regarding their confluence due to concave geometry. Her work was based on an idealized case study with a simplified terrain which reduced the main features of the Andes to an extensive drainage system directed into a wide basin. In a second idealized modeling study (K. Trachte and J. Bendix, Katabatic flows and their relation to the formation of convective clouds—Idealized case studies, submitted to Journal of Applied Meteorology and Climatology, 2010), the importance of sufficient moisture in the atmosphere for atmospheric instability in such a situation was examined, showing that the formation of a convective cloud cluster is possible due to the topographically induced confluence.

[5] The study at hand encompasses the analysis of the development of such nocturnal convective clouds southeast of southern Ecuador in the Peruvian Amazon basin. Its aim is to describe the nocturnal cell formation in the context of a representative case study observed using satellite data. Previous studies showed that infrared (IR) satellite images are effective in analyzing the occurrence of MCSs [Maddox, 1983; Cotton et al., 1989]. Machado et al. [2002] investigated the diurnal cycle of tropical convection using satellite images. Yang and Slingo [2001] also used brightness temperatures to construct a climatology for the diurnal cycle in convection and cloudiness for the tropics. Furthermore, several studies showed that the use of satellite data is beneficial for the validation of numerical models [Chevallier and Kelly, 1997; Morcrette, 1991; Sun and Rikus, 2004; Trigo and Viterbo, 2003]. For this reason, the satellite data in this study are used to verify the simulated cloud occurrence by means of calculated brightness temperatures.

[6] The paper is structured as follows. Section 2 gives an overview of the model setup and the satellite data that was used. The model is validated by comparing observed and modeled brightness temperatures in nocturnal clouds. Information on the reasons for nocturnal cloud formation is provided by an analysis of the area's thermodynamics. In section 3 an analysis of the development of the cloud cluster using a cross-section provides first evidence for the existence of downslope flows as the driving force of convective initiation. In section 4 the formation of convective clouds in conjunction with katabatic flows is discussed.

2. Data

[7] Geostationary satellites provide atmospheric data with a high temporal and spatial resolution and are especially useful for cloud verifications of numerical model results. This is especially true for regions like the target area in the Ecuadorian/Peruvian Amazon (see Figure 1), where no other observational data are available.

[8] In the current study, GOES-East data are used to compare the observed and modeled brightness temperatures. The most important wavelengths are in the infrared (IR) spectrum (GOES-E channel 4 at 10.2 μm–11.2 μm), which is sensitive to clouds. They provide a spatial resolution of 4 km at nadir, which is sufficient to observe cloud features such as MCS.

[9] GOES-E originally provides image data with a repetition rate of 30 min. The data for this study were downloaded at the Marburg satellite station in the framework of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Broadcast System for Environmental Data (EUMETCast) distribution system (J. Bendix et al., The Marburg satellite station, paper presented at Meteorological Satellite Users' Conference, EUMETSAT, Dublin, UK, 2003). The temporal resolution is 3 h with available time slots for Ecuador at 1900, 2200, 0100, 0400, 0700, 1000, 1300, 1600 LST (LST Ecuador = UTC − 5 h). To complete lacking time slots, additional imagery are taken from the NOAA's Comprehensive Large Array-data Stewardship System (CLASS). For the presented case study, we used night images from 12 October to 13 October 2009 (sunset at 1900 LST and sunrise at 0700 LST) in the target area) because the imagery reveals a remarkable example of a MCS in the target area, which is representative for the case study.

3. Model Setup

[10] In the current study the Advanced Regional Prediction System (ARPS) from the Center for Analysis and Prediction of Storms (CAPS) at the University of Oklahoma is used. ARPS is a fully compressible, nonhydrostatic numerical model with a generalized terrain-following coordinate system and vertical stretched grid. For more details, see Xue et al. [1995, 2000, 2001].

[11] For our study we used multiple one-way nested grids with an increment of 36 km with 57 × 33 points, 12 km with 99 × 72 points, 4 km with 150 × 150 points, and finally 1 km with 300 × 300 grid points (see Figure 1 and Table 1). The first domain (D1 in Figure 1) covers Ecuador, southern Colombia, northern Peru, and the western part of Brazil. Domain 2 (D2) and 3 (D3) cover south Ecuador and the adjacent Peruvian Amazon. The smallest domain (D4) was focused on the cloud formation area identified from the satellite data. The vertical grid of both domain D3 and D4 has 55 layers with an average spacing of 350 m. Near the ground the grid is stretched with a hyperbolic function to a minimum of 20 m vertical resolution for the lowest 500 m to capture the development of thermally induced downslopes in the planetary boundary layer (PBL). The 2.5° × 2.5° resolved NCAR/NCEP DOE Reanalysis II data [Kanamitsu et al., 2002] were used to initialized the outer domain. The terrain of each domain is represented by the USGS GTOPO30 data, which have a horizontal resolution of approximately 1 km. For the inner domain (D4) a nine-point smoother was applied 10 times to reduce outliers occurring in the terrain data, e.g., point to point height difference of 1000 m.

Table 1. Grid Configurations
Domaindx, dydzavgdzminLarge Time StepSmall Time StepTerrain
D136 km600 m300 m12 s3 s30″
D212 km500 m150 m6 s3 s30″
D34 km350 m20 m1 s0.5 s30″
D41 km350 m20 m1 s0.5 s30″

[12] ARPS was applied with the Deardorff closure scheme physics package for turbulence parametrization and the implementation of 1.5-order turbulent kinetic energy (TKE) [Deardorff, 1972]. The surface fluxes are responsible for the mass and heat exchange with the atmosphere. In ARPS they are computed by a stability and roughness-length dependent surface-flux model [Businger et al., 1971; Byun, 1990]. The fluxes are solved on the basis of the similarity theory by Monin and Obukhov. In the present simulations, the surface fluxes are calculated with stability-dependent drag coefficients and the bulk Richardson number as the stability parameter. For more details, see Xue et al. [1995, 2001].

[13] A force-restore two-layer soil and vegetation model was used to provide soil-model general surface characteristics [Noilhan and Planton, 1989]. The input data for the soil type and the vegetation are derived from the USGS global input data with a horizontal resolution of approximately 1 km. Radiative processes are considered by the Goddard shortwave and longwave atmospheric radiation transfer parametrization [Chou, 1990, 1992; Chou and Suarez, 1994].

[14] The convective parametrization is disabled in both domain D3 and the inner domain D4 due to the high horizontal resolution sufficient to initiate deep convection [Xue and Martin, 2006] so that convection is calculated directly as a response to the dynamics. An explicit microphysical parametrization scheme was used for the simulation of the cloud properties, which solves six prognostic equations for water vapor, cloud water, rain water, cloud snow, cloud ice, and grauple [Lin et al., 1983].

[15] The brightness temperatures of the model are derived using the Community Radiative Transfer Model (CRTM) [Chen et al., 2008] using the same lookup tables as for GOES Variable format (GVAR) [Weinreb et al., 1997]. The implementation of this retrieval is adapted from the Weather Research and Forecasting Post Processor software.

4. Results

[16] The multiple nested domain setup presented above was used to examine the development of nocturnal convective clouds in the eastern Andes of south Ecuador. The simulated results are compared to observational data on the basis of brightness temperature distributions. Further analyses of the atmospheric conditions show the impact of the horizontal and vertical resolution on the described nocturnal cloud formation mechanism. The development of convective clouds and the presence of a strong low-level flow as the driving force is illustrated with a cross section through the cloud formation area. Their specific features will be discussed in the context of the energy fluxes in order to demonstrate the occurrence of a downslope flow.

4.1. Convective Cloud Patterns

[17] We begin with an evaluation of the mesoscale convective cloud appearance in the target area. Generally, MCS are defined as an ensemble of convective elements with a time scale of at least 3 h Houze [1995]. They differ from an individual cumulonimbus by their duration and spatial extension. A typical feature of MCS are their upper cirriform cloud shield combining several convective cells into an organized cloud system. They consist of both convective regions with intense vertical velocities and stratiform regions with a more uniform texture, which leads to more complex dynamics [Houze, 2004]. MCS are divided into different classes, such as the mesoscale convective complexes (MCC), by IR temperature [Bluestein and Jain, 1985, 1987; Jirak et al., 2003]. MCCs are quasi-circular, long-lasting systems with an IR blackbody temperature TBB < −52°C over an area of 50,000 km2 [Maddox, 1980]. More definitions and thresholds are summarized in Table 2 as defined by Maddox [1980], which are used to identify the nocturnal cloud cluster in the target area.

Table 2. Mesoscale Convective Complex Definitionsa
 Physical Characteristics
Sizecloud shield with continuously low IR temperatures ≤ 220 K must have an area ≥50,000 km2
Shapeeccentricity (minor axis/major axis) ≥ 0.7 at time of maximum extent
Durationsize definition must be met for a period ≥ 6 h
Initiatesize definition is first satisfied
Terminatesize definition is no longer satisfied

[18] Figure 2 and 3 display GOES-E (10.2–11.2 μm) IR brightness temperature images and modeled data from 12 October 2009 to 13 October 2009 at different time intervals beginning at 2015 LST. The images show the nocturnal evolution of a MCS southeast of southern Ecuador. The occurrence of a convective cluster can be confirmed with cold cloud temperatures. The satellite data in the target area (Figures 2a, 2d, and 2g) shows that low IR temperatures can be used as an indicator for the development of convective cells. They start in the early nighttime (2015 LST) and grow throughout the night (2215 LST) to a size of 5000 km2. Several close cells can be seen to possess minimum brightness temperatures of 220 K denoted by the white contour of 220 K. During the night, the cells merge to an organized quasi-circular system (0115–0315 LST, see Figure 3), reaching brightness temperatures of 220 K over an area of 10,000 km2, which characterizes a cirriform cold cloud shield. Inside this shield further areas can be differentiated, revealing considerably colder temperatures below 220 K. These cold cores represent higher atmospheric regions, indicating intense convective regions with overshooting tops. At 0215 LST the interior of the cloud system reaches its maximum size, with a horizontal extension of 12,000 km2. The clear-sky environment surrounding the clusters, which are sustained over the whole night, and the stationary appearance of the cold cloud are particularly noteworthy.

Figure 2.

(a, d, g) GOES-E brightness temperatures (10.2–11.2 μm, K), (b, e, h) ARPS brightness temperatures (K) (domain D3), (c, f, i) ARPS brightness temperatures (K) (domain D4) with a white contour (220 K) for 2015 LST, 2115 LST, and 2215 LST.

Figure 3.

(a, d, g) GOES-E brightness temperatures (10.2–11.2 μm, K), (b, e, h) ARPS brightness temperatures (K) (domain D3), (c, f, i) ARPS brightness temperatures (K) (domain D4) with a white contour (220 K) for 0115 LST, 0215 LST, and 0315 LST.

[19] A comparison of the thresholds in Table 2 shows that not all of the physical definitions for a MCC are fulfilled. Despite the low IR temperatures, the cold cloud shields extend over an area of 12,000 km2, which is too small for an MCC. Additionally, the duration of this shield is less than 6 h and thus fails to fulfill the definition of an MCC. Therefore the pattern is an MCS consisting of an ensemble of convective elements [see Houze, 1995] but not a more severe MCC, as defined by Maddox [1980].

[20] At 0415 LST the cluster begins to weaken (see Figure 4), which is observable by the increasing brightness temperatures (0615 LST). By now the values are around 230 K. The break up of the cloud is related to extensive rainfalls and the lack of sufficient strengthening through latent heat. By the morning hours around 0915 LST, the cloud system has completely disintegrated.

Figure 4.

(a, d, g) GOES-E brightness temperatures (10.2–11.2 μm, K), (b, e, h) ARPS brightness temperatures (K) (domain D3), (c, f, i) ARPS brightness temperatures (K) (domain D4) with a white contour (220 K) for 0415 LST, 0615 LST, and 0915 LST.

4.2. Comparison of Observed and Simulated Cloud Patterns

[21] The spatiotemporal development of the identified convective cloud cluster depicted in Figures 2, 3, and 4 is used to validate the simulation results by comparing satellite-observed data and model derived brightness temperatures. Only domain D3 and D4 are displayed, since domain D1 and D2 show no additional information and are not relevant for the current considerations (see Figure 1).

[22] For the time steps between 2015 LST and 0915 LST, Figures 2b, 2e, 2h, 3b, 3e, 3h, 4b, 4e, and 4h show that the simulated data show no distinguishable convective cells in the target area, as visible in the satellite data. ARPS produces neither deep convection nor cloudiness during the night in the region of investigation, except at the eastern side of the domain where warmer clouds (250 K) occur. Analyses regarding the reasons for this lack are discussed in the following sections.

[23] In contrast, an inspection of the inner domain (D4) with a resolution of 1 km (Figures 2c, 2f, 2i, 3c, 3f, 3i, 4c, 4f, and 4i) confirms the formation of a growing cold cloud pattern southeast of Ecuador featured by the contour of 220 K. A convective cell develops in the target area at the beginning of the night, which is characterized by low brightness temperatures (Figure 2c). Although this cell is smaller (800–1000 km2) than displayed in the satellite data and the GOES-E data show the generation of more than one cell, they show comparable temperatures (220 K). At 2115 LST a quasi-circular system is created reaching a spatial size of 6500 km2 but still remains less than the observed formation. Northwesterly a second cold cloud evolves with brightness temperatures of 220 K and a size of 800–1000 km2 creating an enlarged cluster of 8000 km2 until 2215 LST.

[24] In the ongoing simulation, the two convective clouds are maintained until 0115 LST with colder areas in the cloud shield, corresponding with the patterns visible in the GOES-E data. They are colder than 220 K indicating convective overshooting tops. While in the satellite-observed data the cells merge to one enlarged system reaching a size of maximum 12,000 km2, the simulated data show two separated and smaller cold clouds, respectively. The dissipation of the southerly cluster has already started by 0115 LST as indicated by the warmer IR temperatures, which characterize stratiform areas. Notable is that the simulated cold clouds are within an equal area visible in the satellite-observed data and that they have a comparable orientation.

[25] At 0115 LST new convective cells evolve to the north-east with values of 220 K. While these cells grow until 0415 LST to an ensemble of convective elements more comparable to the GOES-E data, the two circular systems of time period 2115 LST to 0115 LST have disintegrated. The dissipation of the observed as well as simulated nocturnal convective cluster takes place in the early morning hours (0615–0915 LST). Although the size of the simulated cluster lies below the observed pattern the development of a nocturnal convective cloud cluster occurs with similar brightness temperatures.

4.3. Analyses of Atmospheric Conditions

[26] In order to take a closer look at the reasons for initiation and absence of nocturnal convection, analyses of the development of the surface front and the thermodynamic structure of the atmosphere are presented. The interaction between the cold drainage of air from the Andean mountains with the warm-moist air of the Amazon, both for domains D3 and D4 (see Figure 5), is investigated in the first instance. The thermodynamics are examined based on the vertical structure of the atmosphere (see Figure 6 and 7). The influence of the terrain is displayed in Figure 8.

Figure 5.

Horizontal cross-section (xy-plot at z = 300 m asl) of the equivalent-potential temperature θe (shaded, K) of (a) subset of domain D3 equal to domain D4 and (b) domain D4 for 1900 LST.

Figure 6.

Vertical cross-section (xz-plot at x = 38 km and y = 35 km, x = 170 km and y = 215 km) of the equivalent-potential temperature θe (shaded, K) of (a) domain D3 and (b) domain D4 for 2015 LST.

Figure 7.

SkewT log P profiles (a) domain D3 and (b) domain D4 taken at −78.0° × −4.9° for 1900 LST.

Figure 8.

Horizontal cross-section (xy-plot at z = 300 m asl) of the divergence field (shaded, 1/s amplified by a factor of 1000) for (a) subset of domain D3 equal to domain D4 and (b) domain D4 for 1900 LST.

4.3.1. Surface Cold Front

[27] The development and location of the surface cold front is shown by the horizontal distribution of the equivalent-potential temperature θe, which is defined as

equation image

with θ as the potential temperature, Le as the latent heat of evaporation, cp is the specific heat content, TLCL is the temperature at the lifting condensation level (LCL), and wv as the mixing ratio for water vapor.

[28] Here θe represents a value which is composed of the temperature and the dew point. The greater the dew point, the higher an air parcel is able to rise due to the release of latent heat. Thus θe is an indicator for the energy content of the air. Its values are quasi-constant within an air mass.

[29] Figure 5a illustrates θe for the 4 km domain (D3) at 1900 LST for the lower troposphere. The differences between the cold Andean regions, with average values of 324 K, and the warm-moist Amazonian basin, with average values of 348 K, are clearly visible. The katabatic winds from the mountains drain cold air into the basin, generating a horizontal gradient, which is evident by the rapid increase in the equivalent-potential temperature field. This gradient is especially noticeable in the cloud formation region southeast of south Ecuador. The values rise by 14 K in a distance of approximately less than 40 km, indicating a boundary of air masses and thus the formation of the cold front.

[30] A comparison to the equivalent-potential temperature in the 1 km domain (D4), as displayed in Figure 5b, shows that there are no great differences in its horizontal distribution. On average, the lower θe values of 325 K dominate in the western part of the domain, and in the eastern part, the high values of the Amazon dominate, with an average of 344 K. Here a clearly marked zone of rapidly increasing values developed in the cloud formation area due to the cold drainage of air from the Andean mountains. A horizontal gradient of 14 K also occurs in D3 at a distance of approximately 40 km.

[31] However, the impact of the higher horizontal resolution on the individual processes and elements is visible in D4. Along the slopes of the Andes, the cold air propagates further into the Amazon basin. This is caused by stronger generated katabatic flows as a result of more accurately featured terrain: the GTOPO30 terrain data in domain D4 are despite their smoothing better defined affecting the occurrence of the downslope winds.

4.3.2. Atmospheric Stability

[32] A cross section through the front of both domains (D3 and D4) in the region of cloud formation gives further information about the reasons for the initiation and the lack of deep convection, respectively. Therefore the θe is used as well, since it demonstrates the potential instability of the atmosphere. A decrease of θe with height signifies a decrease in the absolute humidity.

[33] Figure 6a shows the vertical distribution of D3 at 2015 LST. A layer of warm-moist air in the lower troposphere is located to the east of the Andes in the Amazon basin, which reaches a thickness of approximately 2.5 km. It is characterized by equivalent-potential temperatures with average values of 346 K. Areas with high values of 354 K are arranged in this layer. They are often burst points for strong convective developments. Machado et al. [2002] confirm that θe values >356 K are mainly associated with an unstable boundary layer.

[34] Above this warm-moist air is a layer of colder and thus drier air. This layer is located between 2.5 km and 6 km and reaches minimum values of 328 K. That is, the equivalent-potential temperature decreases with height to a level of 6 km. The result is a potentially unstable atmosphere, which can become conditionally unstable if the air is lifted and additionally becomes saturated. In this process the mentioned burst points play an important role in the initiation of deep convection, since they can provide the essential energy. Above this layer the equivalent-potential temperature rises again.

[35] The vertical distribution of the equivalent-potential temperature in the 1 km simulation at the same location and time step (2015 LST) is demonstrated in Figure 6b. In the lower troposphere, warm-moist air from the Amazon is visible as well. The θe values reach a maximum 350 K and a thickness of approximately 2.5 km. A comparably colder layer is located above this air, as displayed in Figure 6a. However, in the current simulation, this cold layer is interrupted by several columns with high values (344 K). They characterize a vertical transport of warm-moist air which has reached saturation and becomes more unstable due to the release of latent heat. Hence this displacement of air masses indicates convective activities, which are already suggested by the brightness temperature patterns in Figures 3 and 4.

[36] Apart from the θe horizontal and vertical distribution further parameters are used to study the atmospheric conditions and to evaluate whether convective activities are induced (see Table 3). The basic parameters for these analyses are the convective available potential energy (CAPE) [Moncrieff and Miller, 1976] and the convective inhibition (CIN) [Colby, 1984].

Table 3. Environmental Parameters
 Domain D3Domain D4
CAPE872 J/kg964 J/kg
CIN−168 J/kg−146 J/kg
LCL869 hPa897 hPa
LFC683 hPa770 hPa

[37] The CAPE is the vertically integrated positive buoyancy of a parcel between the height of the level of free convection (LFC) (zLFC) and the height of the equilibrium level (zEL). It represents the maximum energy available to an ascending air parcel and is an indicator for the potential of convective initiation:

equation image

with Θpar as the potential temperature of the parcel lifted from the surface (zsfc) up to the LFC (zLFC) and Θenv the ambient potential temperature.

[38] In contrast, the CIN represents the energy required to lift a negatively buoyant parcel from the surface to the LFC and is defined as:

equation image

Figure 7 shows two atmospheric profiles by night (see Figure 7a and 7b). The amount of CAPE for domain D3 is 872 J/kg, which represents only a marginally unstable atmosphere and mean potential for deep convection. In the same time, the CIN reaches values of −168 J/kg, which is a great amount of negative buoyancy and has to be overcome to initiate convection. On the other hand, the Lifted Index (Li) of −3.1 indicates a moderately unstable atmosphere. This is also true for the K-Index of 31, which characterizes a potential for the development of a thunderstorm.

[39] The atmospheric profile for D4 seems not to differ much from domain D3: the environmental temperature profile and the temperature of an ascending air parcel are almost identical. Due to the same vertical resolution they both develop a surface inversion. The amount of CAPE (964 J/kg) is slightly greater and the convective inhibition of −146 J/kg is slightly smaller. Both profiles represent marginal atmospheric instability and require an explicit trigger to reduce the negative buoyancy and initiate deep convection. However, the K-Index, which with its value of 38 is greater than in D3, indicates a very good potential for the formation of a thunderstorm, as already illustrated in Figures 2b, 3b, and 6b. Additionally, a closer look offers some important details: the lifting condensation level (LCL) and the level of free convection (LFC) are significant lower than in the 4 km simulation. While in D3 the LCL is at the 869 hPa level, the LCL in D4 reaches 897 hPa. This is also observable in the LFC, which is reached in D3 at the 683 hPa level and in D4 at a height of 770 hPa. With a significant trigger function, presumably the downslope winds from the Andean mountains, the warm-moist air masses of the Amazon at the foothills seen in Figure 6b are able to reach the LCL and, finally, the LFC.

[40] The analyses of the D3 and D4 simulation demonstrate that both domains offer a potential instability and thus the probability to initiate deep convection and the formation of a thunderstorm. The results also show that a strong trigger is required to lift the air.

4.3.3. Horizontal Convergences

[41] It is assumed that the nocturnal convective activities in the target area are triggered by strong convergences of katabatic flows from the Andean mountains, which are developed by the shape of the terrain. The flows can be described as horizontal divergence, where negative values describe the convergences and positive values the divergent patterns.

[42] The results of domain D3 and D4 at 1900 LST are demonstrated in Figure 8. It shows the horizontal divergence patterns with an amplification factor of 1000 in the cloud formation area. Especially noteworthy is the consistent distribution of the convergence patterns in domain D3 with rather small areas of greater values (Figure 8a), which reach maximum values of 2.5 10−3 1/s. The convergence zones are organized along the slopes, reflecting the more coarsely rendered terrain due to the 4 km resolved grid size. These patterns also describe the regime of katabatic flows and their convergences at the foothills.

[43] The more highly resolved D4 simulation shows greater variability in the horizontal divergence distribution when compared with the D3 simulation (Figure 8b). The convergence patterns are organized in the same areas but this time more structured and strongly developed. Distinct convergences occur along the slopes with maximum values of 6.5 10−3 1/s, which is more than twice as strong as in domain D3. This is caused by terrain effects (see Figure 1, lower right), which are defined better in domain D4, despite the nine-point smoothing, due to the 1 km spatial resolution, which affects the katabatic flows. They propagate down the steeper slopes and converge at the foothills with a higher mass contribution in these areas. As demonstrated in the idealized case study [Trachte et al., 2010] the shape of the terrain strongly influences the pressure increase and convergences caused by the downslope winds. Those convergence zones are closely linked to vertical velocities because they are a consequence of the horizontal convergences and thus an increase of mass. The patterns herein are located in the region, where the cold clouds appear in the observational data and in the D4 simulation results but not in the coarser D3 data. The horizontal convergence demonstrates the influence of the terrain on horizontal dynamics and, subsequently, vertical characteristics. The mass contribution in D3 is not as strong due to the reduced terrain information resulting in a low increase of pressure. Considering the higher LFC of D3, those convergences are too weak to sufficiently lift the air and initiate deep convection. For domain D4, the convergences are stronger and the LFC lower. In the next section the development of the cloud cluster is examined and discussed in conjunction with katabatic flows.

4.4. Convective Cloud Development

[44] A cross section through the cluster in domain D4 is used to further examine the formation of the displayed cloud. Figure 9 shows the horizontal moisture convergence (shaded), the wind field in uw direction (vectors) and the total condensed water (solid line) at four time steps in a 15 min interval. In Figure 9 the development of a convective tower in its cumulus stage, which can be identified by means of the buoyant plume, is illustrated. It is characterized by a region of increasing wind velocities (w vectors increase with height), which reached maximum values of 13.6 m/s and a height of 14 km. Those accelerations occur when the LFC is reached. Strong low-level moisture convergence with maximum values of 40 (g/kg)/s forces the air parcels to ascend. Within the boundary layer a deep cloud is generated over the maximum of the low level moisture convergence (visible by the solid line of total condensed water) due to the release of latent heat. In the lower troposphere entrainment occurs as a result of the upward mass flux. Accordingly divergence branches in the upper troposphere as well as detrainment evolve at the lateral sides. The cloud remains in this region and grows deeper forced by an enhanced moisture convergence (Figure 9b). Cell regeneration takes place in this region, as indicated by the creation of a group of cells. They form as well over the maximum moisture convergence with values of 35 (g/kg)/s causing an acceleration of vertical velocities (5 m/s). With increasing time (Figure 9c) the convective cells merge to a multicell cluster in its mature stage. Dissipation begins at the rear, which is associated with local upward and downward motion causing extensive precipitation patterns. But at the frontside the new cells grow deeper. After a further 15 min the rearward cell has dissipated but is survived by upper level cirrus it generated. The convective cell at the leading edge in contrast has strengthened characterized by the increasing wind vectors with height. It shows similar features as the tower in Figure 9a. Additionally, downbursts of cold air in the lower levels occur initiating the formation of a new cell in the region of strong moisture convergence with maximum values of 27 (g/kg)/s. Those structures shown herein of dissipating and growing convective cells can also be observed in the satellite-observed cloud properties in Figure 2 and are typical for convective cloud clusters [Houze, 2004].

Figure 9.

Vertical cross-section of domain D4 (xz-plot from x = 71 km, y = 80 km and x = 170 km, y = 215 km) of horizontal moisture convergence amplified by a factor of 1000 (MC, shaded, (g/kg)/s), the wind field in uw-direction (vectors, m/s) and the total condensed water (TW, solid line), g/kg at (a) 2000 LST, (b) 2015 LST, (c) 2030 LST, and (d) 2045 LST.

[45] During the developmental stages of the cluster, a strong current in the lower levels of the atmosphere is recognizable, as can be identified by the vectors. This nocturnal flow, presumably the katabatic flow, advances cold air into the convective area. The interaction of the warm-moist air of the Amazon, which occurs in the region of the displayed frontal zone in Figure 5b, supports the ascension of air parcels and thus the formation and maintenance of the multicell cloud cluster.

4.5. Katabatic Flows

[46] Katabatic flows and their interaction with the specific terrain configuration in the target area are assumed to be the driving mechanism behind the nocturnal cloud formation. The flows develop in hilly regions on calm, clear nights through radiative surface cooling. A buoyancy deficit is generated as a result of a radiation divergence associated with the ground heat fluxes, which result from mechanically induced turbulent-kinetic energy (TKE) and latent (LE) and sensible (H) heat fluxes [Prandtl, 1942; Defant, 1949]. The soil temperature reduces to a value below the adjacent air temperature followed by the sensible heat flux (H) from the atmospheric boundary layer to the earth surface balancing the heat loss. The surface acts as a sink for the thermal energy. The result is a cooling of the lower atmosphere and a development of a positive temperature gradient [Trachte et al., 2010]. Figures 10 and 11 support the assumption that the low-level flow seen in Figure 9 is a thermally induced downslope. Figure 10 displays the TKE and a vertical profile, taken at an individual slope of the Andes at −78.5° × −5.2° of D4, of the wind velocity in u direction. The wind profile discloses a jet-like profile in the lower atmosphere, which is typical for downslope motions with maximum wind speeds of 20.0 m/s. Below this level, the velocity decreases as a result of ground friction; above it, the velocity is reduced by a decreasing positive temperature gradient. TKE occurs near the maximum of the mean wind speed with values of 1.6 m2/s2. In addition to mechanically induced turbulence the vertical wind shear is the second main cause for TKE.

Figure 10.

Profiles of (a) the wind vector (u, m/s) and (b) the turbulent kinetic energy (TKE, m2/s2) for 1900 LST and 2000 LST at −78.5° × −5.2°.

Figure 11.

Heat energy fluxes in W/m2 with the net radiation (Rn), the sensible heat flux (H), the latent heat flux (LE), and the ground heat flux (G) as a function of time between 12 October 1300 LST and 13 October 1300 LST taken at −78.5° × −5.2°.

[47] A further indicator for the development of katabatic flows is the heat exchange between the lower levels of the atmosphere and the surface. Figure 11 displays the surface fluxes and the net radiation (Rn) from 12 October 1300 LST to 13 October 1300 LST. They show the typical course of the transition from day to the night and back to day. A positive Rn is observed during the afternoon, which inverts to negative values between 1700 LST and 1800 LST due to the sunset and the associated begin of radiation loss. After a maximum peak of 100 W/m2 at 1800 LST, the radiation loss achieves nearly constant values of 70 W/m2 during the night. The surface fluxes change their direction correspondingly. As a result of the heating effect during the day, the sensible heat flux (H) has positive values but is reduced to almost zero at 1900 LST. The latent heat flux (LE) reduces as well, but keeps clearly positive until 0200 LST indicating a flux from the surface to the atmosphere due to evaporation. The ground heat flux (G) reach negative values as a result of the temperature deficit between the atmosphere and the surface layer. Therefore an energy flux from the PBL to the surface layer exists, which is caused by a constantly cooling atmospheric boundary layer as a result of a negative radiation budget which is typical for nighttime. This can lead to the development of a surface inversion, as already indicated by the amount of negative buoyancy, which induces a positive temperature gradient. The surface heat fluxes indicate the development of thermally induced katabatic flows, which are associated with the characteristic features of the downslope flow (Figure 10).

5. Discussion

[48] As expected, the nocturnal formation of the presented convective cloud southeast of Ecuador in the Peruvian Amazon basin is affected by well defined and strong downslope winds from the Andean slopes and valleys. Those katabatic flows induce a surface cold front at the foothills of the Andes in the Amazon basin. The special terrain configuration in the target area can cause the formation of a quasi-concave terrain line with strong convergences.

[49] The underlying processes resulting in the formation of a cold cloud are examined using a case study (12 October to 13 October 2009) diagnosed by GOES-E satellite images. This outlines the impact of the terrain and, consequently, of the katabatic flows on the cloud formation: the 4 km domain (D3) reveals the absence of any cloudiness in the target area at nighttime, but in the inner domain (D4) a convective cloud cluster was generated (Figures 2 and 3). The atmospheric stability of both domains regarding the frontal zone and the thermodynamic structure was examined. The development of the katabatic induced surface front was described by the horizontal distribution of the equivalent-potential temperature (see Figure 5). They show similar structures regarding the evolution of a horizontal gradient of θe in the cloud formation area. But even here, the impact of terrain and, subsequently, the downslope winds was visible. The cold drainage air in D4 propagated further into the basin, indicating a stronger flow. A cross section through the front revealed first distinctions (see Figure 6). While D4 demonstrated the vertical displacement of warm-moist air in the target area in one time step while the brightness temperature showed cold cloud patterns, D3 did not offer such structures. Although D3 has a potential instability and areas of high θe (354 K), which are sufficient for a destabilization of the atmosphere [Machado et al., 2002], convective activities are completely absent. An inspection of the environmental parameters gave more information concerning the reasons for the initiation and lacking convection, respectively (see Figure 7). Both domains are marginally unstable, with a great amount of negative buoyancy in the lower levels. However, they also show a good potential for the development of a thunderstorm on the basis of the Li and K Index (see Table 3). The LFC in D3 and D4 are significantly different. In D3, the LFC for an ascending air parcel is reached at 3359 m, which is 1000 m higher than in D4. In association with the great amount of convective inhibition, which must be overcome first, a strong trigger function must be present. This trigger mechanism is assumed to be the convergence of the katabatic flows from the Andean mountains in connection with the Amazonian warm-moist air. As well, D4 demonstrated stronger convergence patterns indicating the initiation of sufficient vertical motions (see Figure 8). In this process, the most accurate representation of the terrain, the surface fluxes as well and the resulting katabatic flows are important features. Laurent et al. [2004] described that MCS initiation in South America is mostly driven by topography. Calbo and Millan [1998] also show the role of the terrain and highlight its influence on the model grid size. The authors demonstrated the significant impact of horizontal resolution on the vertical structure of the model domain. They ascertain that high resolved simulations increase the ability to develop stronger vertical velocities and that the vertical fluxes are more strongly developed. The vertical fluxes can result in modifications in the surface temperature and humidity field, which are sensitive to thermodynamic activities [Mapes and Houze, 1992]. These surface fluxes are also the main driver for the development of thermally induced katabatic flows. However, the vertical resolution essential for the evolution of the flows of both domains (D3 and D4) is identical, but the horizontal grid size differs. By increasing the horizontal resolution, as in domain D4, the complexity of the digital elevation model (DEM) increases as well. Thus the terrain is more structured and the concave shape of the Andes in the target area is better revealed. The katabatic flows are stronger and the convergence in the target area increases. The result is a compressional lifting with markedly high vertical velocities due to a larger increase in pressure. In D4 sufficiently dynamic vertical motions occur, whereby the ascending air parcels reach the LFC. In contrast, the 4 km simulation generates weaker horizontal convergences with consequently ascending air parcels, which fail to reach their LFC (see Figure 9). Therefore the reason for the missing nocturnal cloudiness in domain D3 is that, in the current situation, the lower atmosphere requires a strong trigger to induce deep convection and the formation of a mesoscale cloud cluster. The chosen resolution of D3 is insufficient for this initiation mechanism, since it less represents the terrain, and thus the katabatic flows and the increase in mass, which are the main driver.

[50] In domain D4 ARPS underestimates the development of the generated convective cloud cluster, since the simulated cluster is smaller than the observed one (see Figures 2, 3, and 4). The simulated brightness temperatures at the beginning of the night (2115–0115 LST) showed two distinguishable systems with low IR temperatures (220 K) emerging at two separated regions different from the GOES-E data. But in these areas the convergence patterns, as a direct responds to the terrain (Figure 8b), are strongest causing the convective initiation. Unlike the observed data, the simulated data do not merge to one convective system. The reason is the lack of a stronger upper level wind field (see Figure 10) reaching values of approximately 1.5 m/s. However, new cells emerged at 0215 LST, which are more equal to the observed ones. They consisted of several organized cells within a cloud shield. These different occurring cells suggests two different initiation phases: one being the convergence due to the katabatic flows and the other in the second half of the night a cell regeneration due to the circulation dynamics of the convective cluster. Two different mechanisms of storm initiation in South America was also mentioned by Valesco and Fritsch [1987], although the authors describe one in the afternoon and one nocturnal.

[51] Aside from this deviation, there are several conformances. As in the observed satellite data, the cloud cluster develops in the evening, grows during the night and dissipates in the early morning hours. The cells gain similar brightness temperatures and exhibit typical interior areas with lower values (below 220 K) as well. Even if the location is not exactly collocated with the observed data, the orientation of the cells are similar to the GOES-E data and the error is marginal in comparison to the cloud physics. So for these conditions there is a good agreement between the satellite images and the modeled data.

[52] A closer look at the development of the cloud cluster demonstrates the influence of the surface convergence by an intermittent influx of cold moist air, indicating drainage flow (Figure 9). An analysis of an atmospheric profile regarding wind velocity, the TKE and heat exchange (Figure 10 and 11) contribute to this result. Although the inversion layer is deeper than in a previous study concerning the confluence of katabatic flows [Trachte et al., 2010], it shows the specific jet-like profile and TKE characteristics. The reason for the deviation of the katabatic layer is its dependency on the vertical resolution; in this study, a coarser one was used. The idea that a katabatic flow develops is further supported by the surface fluxes representing diabatic heating. The results in Figure 11 disclose a typical nocturnal course with net radiation loss and a corresponding energy flux from the PBL to the surface layer. It induces a buoyancy deficit, which is the driving force of the downslope flow. On the basis of these results and due to the fact that the 1 km domain (D4) used no cumulus parametrization scheme (see Table 1), a scale dependency for the initiation of these convective activities is assumed. Considering that katabatic flows are small-scale features, the demonstrated simulation results lead to the conclusion that they play a major role in the activation of convection. However, the thermally induced downslope winds act only on the lower levels of the atmosphere in the PBL. They cause the formation of a surface inversion, which produces negative buoyancy (−146 J/kg for D4). CIN, thus obtained in the PBL, has to be overcome first to reach the LFC with subsequent formation of cold clouds. Therefore an additional trigger mechanism must be available for strong and sufficient lifting of air. In the light of the concave Andean ridge line southeast of Ecuador, which forms an extensive drainage system directed into the Amazon basin, this specific terrain geometry is extremely important. Particularly the confluence of the cold drainage of air due to the shape of the terrain results in a compressional lifting, which effectively removes the negative buoyancy. In this process, a destabilization of the atmosphere is caused by the interaction with the warm-moist air of the Amazon, causing deep convection as shown in Figure 9 and a previous idealized study (Trachte and Bendix, submitted manuscript, 2010).

6. Summary and Conclusions

[53] In this study, the mesoscale model ARPS was used to examine the development of a nocturnal cloud cluster southeast of southern Ecuador in the Peruvian Amazon basin. The main subject of the investigation is the demonstration of the driving mechanism behind the initiation of nocturnal deep convection. We hypothesized that cold drainage of air from the Andean slopes and valleys confluence due to the concave shape of the terrain in the target area. There it forms a katabatic induced surface cold front in connection with the warm-moist air of the Amazon basin. Beyond that, this confluence leads to compressional lifting of air, which causes atmospheric instabilities and the formation of convective clouds.

[54] For the investigation, we used the night from the 12 to 13 October 2009, which was chosen based on GOES-E IR temperature images to identify an MCC in the target area. The data showed a quasi-circular pattern with minimum brightness temperatures of 220 K and an area of 12,000 km2. The clear-sky conditions were particularly noticeable in the wider area of the MCS, which is favorable for the development of katabatic flows.

[55] A comparison of the observed data with the modeled data was used to validate the simulations. First, we compared the 4 km domain with the observed data, but no convective patterns were observed. However, the 1 km domain revealed a convective cloud with minimum brightness temperatures comparable to the GOES-E data. Since the inner domain with the higher resolved terrain produced a multicell thunderstorm, we expect a scale dependency in convective activities concerning the DEM. The environmental parameters for both domain D3 and D4 reveal a marginally unstable atmosphere with a probability for deep convection. But the essential trigger mechanism, and thus the katabatic flow and its strong confluence, is only well presented and strong enough in the more highly resolved 1 km domain.

[56] A cross section through the simulated cloud cluster of the inner domain illustrated the initiation of convection and the formation of a multicell-cell thunderstorm. The data gave a first insight into the PBL and the occurrence of strong low-level flows. The assumption of a strong confluence leading to sufficient amounts of ascending of air was strengthened by the environmental parameters.

[57] Finally, the PBL was more closely examined in order to identify whether the katabatic flows are the driving force for convective initiation. The formation of the downslope flow was shown by a profile taken at an individual slope of the Andes directed into the basin. Furthermore, the surface fluxes indicated a net radiation loss with a subsequent energy flux from the PBL to the surface layer, which is the driving mechanism for the development of katabatic flows.

[58] The results of the presented study confirm the expected underlying processes resulting in convective activities with subsequent cold cloud formation. Moreover, they highlight that this specific cloud development is a local scale phenomenon which is not driven by mesoscale circulation.

[59] Hence they confirm our hypothesis that an interaction of katabatic flows with the concave shaped Andean drainage system in connection with the warm-moist air of the Amazon causes a confluence which is strong enough to cause compressional lifting and that this lifting can induce the formation of a nocturnal cloud cluster.


[60] The authors are indebted to the German Research Foundation (DFG) for the funding of the work in the scope of the Research Unit RU816 “Biodiversity and Sustainable Management of a Megadiverse Mountain Ecosystem in South Ecuador,” subprojects B3.1 and Z1.1 (BE 1780/15-1, NA 783/1-1). We thank three anonymous reviewers for useful comments on this manuscript.