A multisatellite analysis of deep convection and its moist environment over the Indian Ocean during the winter monsoon



[1] The aim of this paper is to characterize the deep convective systems over the Indian Ocean during Indian Ocean Experiment (INDOEX) and their relationship to cloudiness and to the Upper Tropospheric Humidity (UTH) of their environment together with the relevant longwave radiation fields. Multisatellite analyses are performed (Meteosat, Scanner for Radiation Budget (ScaRaB), and Special Sensor Microwave Imager (SSM/I)) to measure these environmental parameters. The use of Meteosat water vapor (WV) channel appears very efficient not only for estimating UTH but also for separating high level cloudiness, including thin cirrus, from clear sky and low clouds. The Meteosat infrared (IR) and WV channels are also used for correlating Meteosat and ScaRaB measurements, allowing to retrieve continuously the longwave radiative flux. The longwave flux is used to compute the cloud radiative forcing as well as the clear-sky greenhouse effect. Spatial relationships between upper level cloudiness and UTH are established. A strong positive linear relationship is found suggesting a local moistening of the upper troposphere by convection. The temporal analysis reveals that during the active phase of the intraseasonal oscillation, the longwave cloud radiative forcing reaches a mean value up to 40 W m−2 over a large region in the open ocean, while the average clear-sky greenhouse effect is in excess of 180 W m−2. These radiative parameters are strongly correlated with the upper level cloudiness and upper level moisture, respectively. The temporal variability of UTH explains up to 80% of the greenhouse effect variability. The structure of the convective cloud systems is then studied. The observed population of systems spans a wide spectrum of area from 100 to 1,000,000 km2. The contribution to the high level cloudiness of the systems with a strong vertical development is dominant. These systems, with at least one convective cell reaching the highest levels (below 210 K), present indices of overshooting tops and are the most horizontally extended. The largest system exhibits an average longwave radiative forcing of around 100 W m−2. Their contribution to the cloud forcing over the Indian Ocean is overwhelming. The spatial and temporal variability of the systems is finally related to the UTH and to the clear-sky greenhouse effect. Strong correlations are found indicating that these organized convective systems at mesoscale play a leading role in the Indian Ocean climate. The analysis suggests that deeper convection is associated with larger cloud desks with larger cloud radiative forcing. It is also associated with a moister upper troposphere and a larger clear-sky greenhouse effect. These two effects would provide a positive feedback on the surface conditions.

1. Introduction

[2] The Indian Ocean Experiment (INDOEX) Intensive Field Phase (IFP) took place from 15 February to 31 March 1999 in the tropical Indian Ocean. This multiagency international effort aims at better understanding the role of anthropogenic aerosols on the climate (J. B. Coakley Jr. et al., General overview of INDOEX, submitted to Journal of Geophysical Research, Special Issue, 2000, hereinafter referred to as Coakley et al., submitted manuscript, 2000). Scientific goals include the direct radiative effect of aerosols, the so-called indirect effect of aerosols on cloud radiative properties and the role of the Intertropical Convergence Zone (ITCZ) in redistributing pollutants in the tropical troposphere. We here focus on the latter scientific goal. Indeed, a detailed knowledge of the convective activity over the Indian Ocean during the winter monsoon is required to better understand the role of the ITCZ in the chemistry of the troposphere. The ITCZ related cloudiness plays a fundamental role in the tropics radiative budget. Furthermore, the Indian Ocean cloudiness radiative properties seem to depart substantially from other tropical regions [Rajeevan and Srinivasan, 2000], indicating the need for their detailed characterization. Moreover, first results from the INDOEX IFP suggest that the integrated aerosol radiative effect could slow down the hydrological cycle by blocking sunlight before it heats the ocean, hence diminishing surface evaporation [Satheesh and Ramanathan, 2000]. Then, the basin wide atmospheric hydrological cycle over the Indian Ocean during the winter monsoon also deserves scrutiny. At last, the relationship between convective activity and its moist environment is an important source of uncertainties in current climate models projections of climate change [IPCC, 1995]. Documenting these aspects of the tropical climate is not possible with any single current satellite. A multisatellite approach has to be undertaken in order to relate clouds, moisture and radiation at the adapted scale. Such an effort is proposed in this study.

[3] The organization of the tropical convection in the form of mesoscale convective systems (MCS) is of major importance to the tropical climate. Spatial and temporal evolution of these systems play a fundamental role in the energy exchange cycle and in the radiative budget; in particular over the Indian Ocean [e.g., Sikka and Gadgil, 1980; Webster and Stephens, 1980]. These MCSs have been much studied in most of the tropical regions [e.g., Machado et al., 1992; Mapes and Houze, 1993; Chen et al., 1996]. The Indian Ocean region is less documented mainly because of the lack of satellite data [Laing and Fritsch, 1993; Roca and Ramanathan, 2000; Wilcox and Ramanathan, 2001]. Reviews of tropical convection studies allows to design a typical MCS resulting from different scales interaction [Houze and Betts, 1981; Redelsperger, 1997]. It is composed of a convective core where heavy rainfall takes place at typical scale of 10–100 km. This convective core is composed of individual convective cells of scales between 1–10 km. A stratiform anvil associated with lighter precipitation as well as nonprecipitating cirriform cloudiness of typical scale of 100–1000 km is attached to the convective core. Such a scale interaction is currently not included in General Circulation Models (GCM) [Moncrieff and Klinker, 1997].

[4] The interaction between organized convection and its moist environment is central to the much debated moistening or drying of the upper troposphere. Numerous observational studies emphasized the local positive correlation between tropical upper tropospheric relative humidity (UTH) and deep convection induced cloudiness [e.g., Soden and Fu, 1995]. The role of cirrus cloudiness in the distribution of humidity in the tropical troposphere is also important. It has been proposed as a mechanism that could explain tropospheric drying and is directly related to the mesoscale nature of tropical convection [Sun and Lindzen, 1993]. The interaction between tropical convection, large-scale dynamics and the water vapor distribution has also received considerable attention [e.g., Pierrehumbert, 1995; Sherwood, 1996; Salathé and Hartmann, 1997; Pierrehumbert and Roca, 1998; Soden, 1998]. These studies all point toward a better understanding of the source of moisture for the tropical troposphere and the associated mechanisms which are strongly linked to the organization of convection at mesoscale. The relatively poor representation of this interaction in GCMs [Soden and Bretherton, 1994; Salathé et al., 1995; Roca et al., 1997; Salathé and Hartmann, 1999; Roca, 2000] motivates the need of dedicated observational insights into the moisture source issue.

[5] The motivation of the present study is to relate the large-scale characteristics of the ITCZ and the Indian Ocean atmospheric moisture content to the synoptic characteristics of convection at the convective system scale. The approach consists in investigating the mechanisms that relate convection to its associated cloudiness and moist environments. The impact of these mechanisms on the radiative budget is also evaluated thanks to the merging of multisatellite information. Specifically, two mechanisms of importance to the tropical climate are emphasized:

  1. The individual convective systems relationship to upper level cloudiness. Using ISCCP data, Chou and Neelin [1999] show that the cirrostratus and cirrocumulus cloudiness is tightly connected with deep convection cloudiness. They furthermore propose a cirrus-detrainment-temperature feedback that reads as follows: as the surface warms, the convective clouds grow deeper and reach colder temperature; this colder detrainment level is associated with an increase of cirrus fraction. The combined radiative effect of these cirrus (both shortwave and longwave) together with the radiative properties of the deep convection cloudiness suggest a positive feedback onto the tropical climate. Using INSAT data over the Indian Ocean, Roca and Ramanathan [2000] investigated the relationship between the spatial extent of individual convective systems (convective core + stratiform and cirriform shield) and the minimum temperature reached within a cloud. The latter being a useful proxy for convection detrainment temperature. They show that, as the cloud detrainment temperature becomes colder the individual convective system area increases. This would suggest a modified detrainment-temperature-cirrus cloudiness feedback that would read as follows: as the surface warms, convection is reaching deeper height associated with the tropopause and the detrainment temperature becomes colder. Owing to the cloud detrainment temperature-spatial extent relationship, this deeper and cooler cell is associated with a more extended mesoscale convective system. This increase of the whole area of the mesoscale convective cloud system could provide the previously mentioned cirrus fraction increase. This organized convective systems based mechanism is investigated in section 6 and the longwave cloud radiative forcing component of the proposed mechanism is estimated.
  2. The individual convective systems relationship to the upper tropospheric humidity (UTH). Using few days of GOES observations, Udelhofen and Hartmann [1995] investigated the distribution of high level cloudiness and UTH over the tropical Eastern Pacific and Americas. Positive spatial and temporal correlation is found between cloudiness and UTH. They suggest that the tropical organized convective systems are responsible for this local moistening and stress the importance of the convection detrainment height in the UTH issue [see also Salathé and Hartmann, 1997]. The role of organized convection and detrainment levels in the tropical upper tropospheric moisture distribution is investigated as well in this paper; and the impact of such a relationship on the clear-sky radiative budget is evaluated.

[6] The approach we follow here consists in documenting the large-scale Indian Ocean cloudiness and moist environment together with the relevant longwave radiation fields. Then we scale down to the cloud scale and analyze the individual convective systems structural and radiative properties. Finally, the spatial and temporal evolution of the properties of these mesoscale organized convective systems is related to the large-scale environment: the longwave cloud radiative heating, the upper tropospheric humidity and the associated clear-sky greenhouse effect.

[7] The paper is organized as follows. Section 2 presents the data and the methods used for the cloudiness analysis. The humidity, precipitation estimates and sea surface temperature data are introduced in section 3. In section 4, the methodology and the results of the longwave flux estimation from Meteosat and ScaRaB data are given in details. The large-scale analysis of the ITCZ and its environment is provided in section 5. Section 6 introduces the technique used for extracting individual convective systems from the imagery, as well as the convective systems structural and radiative properties. The individual convective cloud properties are then related to their moist environment. Finally, summary and conclusions are offered in section 7.

2. Data and Method for Cloudiness Analysis

[8] In this section, the data related to cloudiness are presented. A cloud classification scheme is also introduced and compared with the results of other cloud analysis.

2.1. The Meteosat INDOEX Imagery

[9] In support of the field campaign, the European Organization for Meteorological Satellites EUMETSAT agreed to move one of the Meteosat satellites from its back-up position over Africa to the Indian Ocean. Meteosat-5 journeyed from 10°W to 63°E in spring 1998. After a 90-day travel at a 0.5°/day rate, the satellite was stopped and the full commissioning was successful. This important effort yields a unique opportunity for high quality observation of the Indian Ocean accessible to the research community. Meteosat images the Earth in three spectral bands. The visible channel spans 0.3–1.05 μm. The atmospheric window infrared channel (IR) measures the outgoing radiance in the 10.5–12.5 μm band. The third channel is the so-called water vapor channel (WV) centered around 6.3 μm. The latter spectral region corresponds to a rotation-vibration band of water vapor where strong absorption takes place. Hence, this channel is mainly sensitive to the humidity content of the mid-to-upper troposphere. More details are given about the WV channel in the next section. At nadir, the footprint of the observations are around 2.5 × 2.5 km for the visible channel and around 5 × 5 km for both infrared channels. The satellite acquires an image of its full field of view every half-an-hour. A vicarious calibration is performed for both the infrared channels [e.g., Schmetz and Turpeinen, 1988; Van de Berg et al., 1995]. This vicarious calibration relates the raw numerical count measured by the radiometer to computed radiances obtained from forward radiative transfer computations and radiosondes. Note that since July 2000, the blackbody on board calibration system of the nominal Meteosat is activated. First comparisons with the vicarious calibration of the WV channel suggests that the former technique was of good quality [See http://www.eumetsat.de]. In the case of the INDOEX Meteosat data, the water vapor channel is intercalibrated with Meteosat-7 in a procedure similar to the one undertook for Meteosat-3 when shifted over the Americas [de Waard et al., 1992]. The quality of the present calibration is difficult to estimate. Detailed comparison with in-situ radiosondes during the campaign could be a useful test but is out of the scope of the present paper. Incidentally, the comparison with ScaRaB performed in section 4 indicates that the calibration of the Meteosat infrared channels was stable all over the INDOEX IFP (see Table 3). The INDOEX subdata set of Meteosat observations is composed of all the three-channel imagery over the region 30°E–110°E, 35°S–35°N offering a full coverage of the tropical Indian Ocean. Half-hourly images are available from 1 January to 30 April 1999. Due to the Sun-satellite geometry, some images are missing around local midnight during the so-called eclipse period. The preceding and following periods are covered with a three hourly rate. The data were archived at the Laboratoire de Météorologie Dynamique (LMD), Palaiseau, France, thanks to the CLIMSERV database that can be accessed through Internet at http://www.climserv.polytechnique.fr. A regular grid was preferred to the original pixel projection. The full resolution data were reprocessed at LMD by the CLIMSERV data team on a 0.04° × 0.04° grid using a closest neighborhood algorithm. This data set was then converted into netCDF format for facilitating interportability and manipulation.

2.2. The EUMETSAT Cloud Analysis Product

[10] Together with the multichannel imagery, EUMETSAT provided the INDOEX community with so-called operational products. Indeed several algorithms have been developed over the last ten years to retrieve geophysical parameters from the raw radiation measurements [http://www.eumetsat.de]. Most of the EUMETSAT products are available in the native EUMETSAT format over the so-called segments that corresponds to 32 × 32 IR pixel boxes. The Cloud Analysis (CA) product is mainly composed of the raw results from the initial preprocessing step common to all the products. This preprocessing step corresponds to the scene identification and is based on a multispectral dynamic clustering approach and comparison with a set of typical radiances associated with different types of cloudiness and geotypes [Francis, 1996]. For each segment, up to three clusters (in the multichannel space), are defined and corresponding multichannel radiation statistics are archived. These statistics encompass the number of pixels forming the cluster, the estimated cloud top temperature and cloud top pressure of each cluster. It should be noted that special attention is paid to the cirrus cloudiness and the cloud top physical temperature and cloud top pressure are corrected for semitransparency effects along the lines of [Szejwach, 1982]. A full account of the CA product algorithm is given by Francis [1996]. The vertical and horizontal information is used to achieve cloud cover estimates assuming fully overcast pixels for different types of cloudiness. The Cloud Analysis products are available every three hours starting at 0000 Z.

2.3. A Bispectral IR/WV Cloud Classification Scheme

2.3.1. Rationale for the Cloud Scheme

[11] A cloud classification scheme is applied onto the raw Meteosat imagery data in order to (1) isolate the mid-to-upper level cloudiness including semitransparent cirrus from the total cloudiness and (2) to provide clear-sky detection needed to estimate clear-sky OLR based radiation budget terms (section 4). Over the Indian Ocean, low-level cloudiness is mainly attributed to stratocumulus and trades cumulus clouds that are encountered at the top of the marine boundary layer in the subtropical as well as tropical regions [Bony and Collins, 2000] (G. Sèze and H. Pawlowska, Cloud analysis from METEOSAT-5 during INDOEX, submitted to Journal of Geophysical Research, Special Issue, 2000, hereinafter referred to as Sèze and Pawlowska, submitted manuscript, 2000). Some large low stratocumulus decks are also present off the West coast of India during the first hours of the morning as revealed by careful inspection of the cloudiness during the field campaign. Note that most of the low-level cloudiness is composed of subpixel clouds at the Meteosat scale. Mid-to-upper level cloudiness include the convective towers, the high and thick stratiform clouds, the midlevel cumulus clouds and thick to semitransparent cirrus. The former three general types of clouds are readily discriminated from clear sky or other lower levels clouds with the use of the single IR channel data. Their signature on the radiation field is indeed substantially colder than the surface. (Given that this study does not aim at estimating cloud cover per se in difficult conditions, like low-level trade-cumulus fields, we will consider all the Meteosat full resolution pixels as totally filled and no partial filling is allowed for.) On the other hand, the thin cirrus clouds may have brightness temperature very warm in the IR channel owing to their low effective emissivities, yielding the discrimination from low-level cloudiness using the IR channel alone, illusive. Convective activity over tropical oceanic regions, and in particular over the Indian Ocean, is characterized by a strong diurnal modulation, with a maximum of activity during nighttime—early morning [Roca and Ramanathan, 2000]. According to these previous comments, the rationale for an efficient cloud classification scheme for convection induced cloudiness analysis should include an improved detection of thin cirrus. It should as well be available at day and nighttimes, precluding the use of the visible channel.

2.3.2. The Cloud Classification Scheme

[12] The Meteosat water vapor channel (5.7–7.1 μm) sensitivity in cloudy sky has been shown to allow cirrus cloud signature to be retrieved. Indeed, experimental observations of cirrus emissivities in both broad band channels (IR and WV) [Szejwach, 1982] as well as radiative transfer computations [Ebert and Curry, 1992] show that cirrus emissivities are almost identical in the two channels. Owing to semitransparency, clear-sky radiation complements cloud radiation in the measurement over a given pixel. The IR contribution function clearly peaks down enough for warm ocean surface to contribute significantly to the signal while the WV contribution peaks in the colder mid-to-upper troposphere. As a result, the semitransparent cirrus clouds appear colder in the WV channel than in the IR one. This very fundamental aspect of WV channel radiation in cloudy sky was successfully used for high cloud retrieval from Meteosat [Desbois et al., 1982] as well as for operational temperature semitransparency corrections. In the present study, we use the WV channel to improve thin cirrus cloud detection with a simple cloud classification scheme. Limitations of using the WV channel for cloud detection purposes arise from the existence of pure upper tropospheric water vapor structures. Indeed these mesoscale (50–100 km) water vapor features are commonly seen in the upper levels of the tropical troposphere and at the edges of the subtropical jet-streams [Desbois et al., 1996; Pierrehumbert and Roca, 1998; Soden, 1998]. It is usually considered that these cold water vapor structures would pollute any water vapor channel based spatial coherence approach for cloud detection [Soden and Bretherton, 1993]. As detailed in the next paragraph, the WV channel test is here applied only to the IR predetected cloudy pixels (no spatial coherence is applied on the WV channel). This prevents misinterpreting most of the pure upper tropospheric water vapor structures for clouds. Nevertheless, upper tropospheric pure moist features over low-level cloud fields, as expected over the subtropical trade-cumulus fields might be poorly classified. The selection criterion used in the present scheme minimizes these problems as discussed next, where the results of the present scheme are found in very good agreement with other independent cloud classification algorithm. In order to investigate the wide diversity of cloudiness encountered during INDOEX, and to account for the important issue of the warm cirrus clouds detection, a simple cloud scheme has then been developed. It is based on Meteosat observations in both the IR window and the water vapor channel to classify the entire cloudiness during day and nighttimes.

[13] The cloud classification scheme relies on simple thresholds and three cloud types are considered: clear sky, low-level cloud, mid-to-upper level cloud and thin cirrus. The first step consists in delineating the clear-sky region in the image. The clear-sky detection makes use of the IR channel only and relies on the sound fact that clear sky is warmer and more homogeneous than cloudy skies over tropical oceans. Homogeneity is estimated by computing the standard deviation of the 3 × 3 neighboring pixels along the lines of the classical spatial coherence technique [Coakley and Bretherton, 1982]. The pixel is classified as clear sky if the IR brightness temperature is warmer than the clear-sky threshold and if its standard deviation is less than a given threshold. The second step consists in a simple partitioning of the cloudy pixel into two categories. If the pixel brightness temperature is colder than 270 K, it is classified as mid-to-upper level thick clouds. In the opposite case (Tir > 270 K and cloudy), a third and last test is applied using the WV channel. If the WV channel brightness temperature is colder or equal than 246 K then the cloudy pixel is associated with the thin cirrus cloud type; if not, it is classified as low clouds. Under the preceding assumptions, the low cloud type includes water clouds, some being small towering cumulus as well as low-level stratocumulus and trade cumulus. Similarly, the warm cirrus cloud type includes all the warmest part of semitransparent high clouds, other semitransparent clouds with effective temperature colder than 270 K being included in the mid-to-upper level cloud types. Table 1 summarizes the threshold used and the different tests. An example of the scheme classification is presented in Figure 1. The IR and WV images (Figures 1a and 1b) reveal cold areas over Indonesia, Madagascar and south of the tip of India. Smaller clusters of cold areas scatter all over the Ocean as well. Warmer brightness temperature areas are present in the southeastern part of the region. No particular signature of this warm patch is seen in the WV image. The result of the classification scheme in shown in Figure 1c. The thin cirrus are clearly found in the surrounding of the convective systems. Low cloudiness is found in the warm (in terms of IR temperature) subtropical regions.

Figure 1.

Example of the cloud classification scheme and the individual convective systems retrieval on 1 March 1999 at 0730 GMT (slot 16). (a) The Meteosat IR brightness temperature. (b) The Meteosat WV brightness temperature. (c) The result of the cloud scheme. Clear sky: white, low-level cloudiness: blue, mid-to-upper cloudiness: pale blue, thin cirrus cloudiness: orange. (d) Class I individual cloud. An arbitrary random color is associated with each convective system. (e) Same as (d) for Class II. (f) Same as (d) for Class III. See section 6 for details on the last 3 images of the figure.

Table 1. The Bispectral Cloud Scheme and Thresholds and Cloudiness Classes Definitions
Cloudiness classBispectral test
Clear skyTIR > 282 K and σIR ≤ 0.5 K
Cloudy skyOtherwise
Mid-to-upper levelsTIR ≤ 270 K
Low levelsCloudy and TIR > 270 K and TWV > 246 K
Semitransparent thin cirrusCloudy and TIR > 270 K and TWV ≤ 246 K

2.4. Intercomparison With Other Satellite Cloud Products

[14] Validation of such an algorithm is a complicated task due to the lack of truth cloudiness observations. Nevertheless, comparisons with other algorithms as well as scrutinized reading by trained operators are indirect means of getting some insights into the quality of the technique. The scheme discussed above was ran during the field campaign on the locally acquired Meteosat observations and was proven to be a powerful interpreting tool for now-casting and flight missions preparation (Coakley et al., submitted manuscript, 2000). Moreover, the classification scheme went through a careful checking by the operational headquarter scientists and meteorologists and successfully passed this friendly subjective test. Intercomparing different algorithm results is another way for estimating the classified images. Sèze and Pawlowska (submitted manuscript, 2000) offer Meteosat cloud classification analysis based on dynamic clustering interpretation of bidimensional histograms of the visible channel and IR measurements. The use of the visible channel allows this classification to be an almost independent estimate to compare with the present one. Similarly, EUMETSAT provides INDOEX participants with their operational Cloud Analysis products offering a third party for intercomparison. The first of March 1999 0730 GMT was chosen as a test case. Indeed, this slot corresponds to local noon at nadir condition and is best suited for the visible/infrared clustering technique. This rapid intercomparison was performed over 5 different regions spanning different upper tropospheric water vapor background, from the moist deep tropics to the dry subtropical regions, as well as a full set of cloudiness conditions. The cloud cover estimates for the three algorithms for each region are shown in Table 2. The percentages in the table should be understood as the relative amount of pixel for a given cloud type. And it should not be associated with cloud cover, owing to the existence of partially filled pixels, especially for the low-level cloud type.

Table 2. Intercomparison of the Cloud Cover Between the VIS-IR Classification Scheme, EUMETSAT Cloud Analysis (CA), and the Present IR-WV Schemea
RegionClear skyLow-level cloudsMid-to-upper level cloudsThin cirrus
  • a

    Only oceanic regions are considered. The mid-to-upper level cloudiness here includes the thin cirrus classes. Units are %.

All 35S35N,40E110E40.865.244.124.311.021.034.923.834.917.117.3
Tropics 10S10N,40E105E32.460.334.716.47.413.251.232.352.124.224.4
Bay of Bengal 10N20N,80E100E89.492.
Arabian Sea 10N20N,60E75E96.810096.
Madagascar 30S10S,40E80E32.759.543.446.910.634.320.429.922.37.911.1
Subtropics 30S10S,60E105E33.165.732.126.921.527.740.012.840.219.812.0

[15] The EUMETSAT analysis overestimates the clear-sky fraction compared to both the other algorithms over all regions. The two other classifications give more restricted clear-sky areas which is favorable for measuring the clear-sky flux (see next section). Mid-to-upper cloudiness amount of pixel from the EUMETSAT classification is lower than the other schemes. Despite this bias, the EUMETSAT products temporal variability (not shown) is in good agreement with the other schemes and monthly means of mid-to-upper cloudiness is a consistent indicator of convective activity. The IR/WV scheme and the VIS/IR clustering schemes agree within 1% with similar regional deviations. The comparison for the thin cirrus class is performed only for the IR/WV and VIS/IR schemes (last two columns of Table 2). The thin cirrus clouds are mainly found in the Tropics. There, good agreement between the two techniques is found (less than 1% difference in the amount of pixel flagged cirrus). In the subtropics, but for the Madagascar region, the IR/WV scheme slightly underestimates the amount of thin cirrus when present when compared to the VIS/IR scheme. The use of the WV channel is shown to offer extended accessibility to the semitransparent cloudiness that would otherwise (say with the IR channel only) be impossible. The good behavior of the current scheme in the Tropics is a major improvement and allows to extract the entire cirriform shield of the convective systems over day and nighttimes (section 6). A more detailed intercomparison exercise is currently being conducted, including more cloud classification algorithms over a larger data set. Dedicated pixel-to-pixel intercomparison statistical analysis are performed in order to better understand the small discrepancies between the techniques.

[16] In summary, it is emphasized that the IR/WV scheme behaves well for mid-to-upper level cloudiness with respect to the VIS/IR algorithm, indicating that the overall approach is successful in achieving the goal it was designed for. The scheme provides clear-sky detection results similar to the VIS/IR scheme that allows to estimate clear-sky OLR from the Meteosat measurements after further processing that uses ScaRaB radiation measurements (section 4).

3. Data for Humidity, Precipitation, and Sea Surface Temperature Analysis

[17] This section introduces the data set related to humidity estimates used in this study. The precipitation climatology and the sea surface temperature are also briefly presented.

3.1. Humidity Data

3.1.1. Precipitable Water

[18] The precipitable water (PW) is the vertically integrated pressure weighted water vapor content of an atmospheric column. Owing to the exponential decrease of specific humidity with altitude, the PW estimates over the tropical oceans mainly reflect the boundary layer moisture content and complement the UTH measurements. The PW data set used here was acquired at Remote Sensing Systems, Santa Barbara, CA, USA. They were derived from the microwave measurements of the Special Sensor Microwave Imager (SSM/I) on board the Defense Meteorological Satellite Program satellites. The algorithm for the PW retrieval from the SSM/I measurements is fully discussed by Wentz [1997]. The comparison with radiosondes yields a root mean square error of 1.2 mm. The bias is very small around 0.6 mm for most of the water vapor burden conditions. Over the moistest regions (PW > 62 mm), the retrieval slightly underestimates the precipitable water by 1.7 mm. Measurements are restricted to the oceanic region. During the 1999 winter, three platforms were available at the same time offering a relatively frequent coverage of the Indian Ocean region. Monthly means were constructed from all the available passes. The data are available over a regular grid of 1° × 1° resolution and were averaged onto a 2° × 2° regular grid in order to allow the comparison with the other data set.

3.1.2. The Upper Tropospheric Relative Humidity

[19] The second EUMETSAT operational product used in this paper concerns the Upper Tropospheric relative Humidity (UTH). It corresponds to the mean relative humidity over a layer extending from 600 hPa to 300 hPa in clear sky. The retrieval of this parameter relies on the use of the 6.3 μm band. The radiation measured by the water vapor channel of Meteosat can indeed be related to different humidity quantities. The brightness temperature in the WV band can be thought of as the temperature of the isosteric specific humidity surface where the optical depth (in the Meteosat 6.3 μm channel) is equal to 1 [Ramond et al., 1981]. It also can be interpreted as the mean relative humidity of a wide tropospheric layer [Schmetz and Turpeinen,1988]. More recently, it was shown that the GOES 6.7 μm brightness temperature (which are very similar to Meteosat water vapor channel) is a linear function of the natural logarithm of the ratio of UTH by the cosine of the viewing angle [Soden and Bretherton, 1993]. The way the linear relationship coefficients are fitted is crucial to the quality of the retrieval. Recently, the EUMETSAT retrieval algorithm was updated along these lines from its original version [Schmetz and Turpeinen, 1988]. The scheme was improved by incorporating the local temperature lapse rate into the statistical coefficients computations [Schmetz et al., 1995]. The computations of UTH are restricted to the scene with neither high nor middle level cloudiness. Indeed, the contribution function of the water vapor channel prevents any low-level cloudiness effect. The algorithm is fully described by Schmetz et al. [1995]. Briefly, the forecasts European Center for Medium Range Weather Forecast provide the temperature profiles. Below 600 hPa the forecasted humidity profiles are used. Above 300 hPa, climatological values of specific humidity are used which make a very small impact on the radiance computation [Takayama, 1992]. Then, two forward radiative computations including the geometry of view effects are performed with the radiation code designed for operational calibration. The first one assumes a 5% relative humidity in the 600–300 hPa layer and the second one a 40% relative humidity. These two values are used to fit the linear relationship between brightness temperature and the logarithm of UTH divided by cos θ. This local look up table allows to convert the observed clear-sky radiance into UTH. The original product was available on the segment grid and was averaged onto a regular 2° × 2° grid. The UTH maps are available every hour, starting at 0000 Z.

3.2. Climatological Precipitation Estimates and Sea Surface Temperature

[20] Monthly means of precipitation rate from the Global Precipitation Climatology Project version 2× are used. This product relies on three sources of information: available rain gauges and ship of opportunities, infrared satellite measurements (both geostationary and polar platforms) as well as satellite based microwave measurements from the SSM/I instrument on-board the DMSP satellites [Huffmann, 1997; Huffmann et al., 1997]. Estimates of the uncertainties associated with this product are difficult to establish for the Indian Ocean and it should be borne in mind when quantitative interpretation of the results are discussed in section 5. The monthly mean surface precipitation rate (mm day−1) are further converted into atmospheric latent heat release (W m−2) or vertically integrated latent heating, by simply multiplying the later by the latent heat of condensation and the water density [e.g., Peixoto and Oort, 1992; Wilcox and Ramanathan, 2001]

[21] The Sea Surface Temperature (SST) used here are obtained from the Reynolds optimum interpolation scheme [Reynolds, 1988]. This blended analysis yields weekly means of SST at a 1° × 1° resolution where ships and satellite measurements are integrated. The SST are used to compute the clear-sky greenhouse effect as described in the next section.

4. Data and Method for Longwave Radiation

4.1. Radiation at the Top of the Atmosphere From ScaRaB

[22] Two models of the Scanner for Radiation Budget (ScaRaB) radiometer have operated in space from February 1994 to March 1995 and from August 1998 to April 1999. The first ScaRaB was launched on board the Russian satellite Meteor-3/7 and provided one year of data on the Earth Radiation Budget [Kandel et al., 1998]. The second ScaRaB was launched on board the Sun-synchronous Resurs 01/4 whose descending node crosses the equator at 1015 am. However, the satellite data transmitter failed early and only five months of useful data from November 1998 to March 1999, with some gaps, are available from this second flight [Duvel et al., 2000]. Concerning the INDOEX IFP, the ScaRaB-2 coverage is almost complete: from 20 January to the end of March 1999, about 50 days of ScaRaB data are available and provide day and night observations.

[23] Like ERBE [Barkstrom et al., 1989] and CERES [Wielicki et al., 1996], the ScaRaB instrument is a scanning radiometer with two broadband channels: shortwave or SW (0.2–4.0 μm) and total (0.2–100 μm). The SW channel measures the reflected solar radiation. The terrestrial longwave radiation (LW) is derived from the total channel at night and from difference between total and SW radiances during the day. The CERES protoflight model was also flying on board TRMM satellite since January 1998, but stopped routine measurements in September 1998. Some data collection resumed for the INDOEX period and participated to CERES/ScaRaB intercalibration exercises [Haeffelin et al., 2001].

[24] Instantaneous flux and regional monthly means are derived from the raw radiances using an ERBE-type secondary processing adapted to ScaRaB [Viollier et al., 1995]. In order to minimize biases between the various time series, the same 12 scene classifications as ERBE [Smith et al., 1986] are used. The scene identification is based on the combination of five geotypes (ocean, land, snow-ice, desert and coast) and four cloud categories (clear, partly cloudy, mostly cloudy and overcast). The algorithm of scene identification is the Maximum Likelihood Estimation (MLE) [Wielicki and Green, 1989] which compares the measured (unfiltered) LW and SW radiances to a predefined set of radiances for the appropriate geometry of view and geographic zone [Suttles et al., 1988a, 1988b]. This algorithm allows to determine the anisotropic factor for radiance-to-flux conversion. The instantaneous regional means are then interpolated or extrapolated along the day to all local times using the same algorithms as ERBE [Brooks et al., 1986]. In the LW domain, the radiant flux interpolation is linear over ocean and snow-ice, but uses a daytime half-sine fit over land/desert and coastal scenes. In the SW domain, each regional instantaneous average is adjusted to the local half-hours, taking into account modeled directional albedo for each scene type and differences of Sun elevation. The time interpolation/extrapolation procedure is another important issue since satellite in a near polar orbit observes low and mean latitudes only twice a day. The estimation of uncertainties for various radiation data is a large problem. Thanks to in-flight internal calibration and various cross-checking operation [Duvel and Raberanto, 2000], the accuracy of the radiances reaches 2% in SW, 1% in LW. The propagation of any calibration error through the monthly averages however is not necessary linear. Calibration error may act to change the cloud identification, and then the spectral, angular and diurnal corrections, all depending on the scene identification. Individual misclassification of scenes can result in erroneous angular corrections [e.g., Diekmann and Smith, 1989; Ye and Coakley, 1996], and even if the scene classification is correct, angular correction makes use of a model. Indeed, uncertainties of about ±15 W m−2 and ±5 W m−2 are expected on instantaneous regional means for SW and LW, respectively [Barkstrom et al, 1989]. If one assumes the individual observations to be statistically independent, these errors in the monthly means should be considerably reduced, but the time sampling issue introduces uncertainties of more than 5 W m−2 in both domains.

4.2. Estimation of Longwave Radiation at the Top of the Atmosphere From Collocated ScaRaB and Meteosat Infrared Channels

4.2.1. Methodology

[25] One recurrent approach in estimating longwave radiation at the top of the atmosphere from operational geostationary satellites consists in using collocated narrow band infrared radiances from geostationary spacecraft and from broad band longwave flux estimates from dedicated radiation instruments flown on board polar-orbiter platforms to fit a statistical relationship [e.g., Schmetz and Liu, 1988; Minnis et al., 1991; Li and Trishchenko, 1998, 1999; Trishchenko and Li, 1998; Dewitte et al., 1999]. The longwave (4–100 μm) flux (W m−2) corresponds to the hemispheric integration over all the solid angle of the longwave radiance and is measured with a small uncertainty due to calibration by instruments like ScaRaB. This uncertainty is typically about 5 W m−2 (see above). Geostationary satellites commonly, but not only, images narrow band radiances (W m−2 sr−1) in the 10.5–12.5 μm window region. Converting the latter in the former hence requires to handle the angular integration as well as the spectral integration. Unlike shortwave measurements [e.g., Duvel et al., 2000], the longwave domain is not much sensitive to the anisotropy of the radiation field indicating that the radiance-to-flux conversion step is not the limitative step. We here follow such a statistical approach proposed by Chéruy et al. [1991] that takes into account the availability of the narrow band 6.3 μm channel of Meteosat to yield better spectral integration than with a single classical window channel. Using the observations of the Earth Radiation Budget Experiment (ERBE) and collocated measurements of Meteosat-1, Chéruy et al. established equation (1) from multiple regression analysis. This study indicates that root mean square error of less than 10 W m−2 in the estimation of the instantaneous longwave flux can be achieved. This equation links the Meteosat radiance observations in the window and in the water vapor band to the longwave flux:

equation image

where FLW is the longwave flux (W m−2) estimated by ScaRaB. LIR and LWV are the radiances (W m−2 sr−1) in Meteosat infrared window and water vapor channels, respectively. The ais, I = 0 to I = 5, are the coefficients of the statistical fit. θ is the zenithal viewing angle of Meteosat. Quadratics and cubic terms in the relationship arise from the multiple regression procedure used and were shown to be the statistically significant higher order terms even though their physical meaning is unclear [Chéruy et al., 1991]. The Meteosat radiances at full resolution (about 5 km × 5 km at nadir) are averaged over the ScaRaB/Resurs footprint (about 50 km at nadir) before the coefficients are computed. Only the ScaRaB observations taken during the Meteosat image acquisition process are retained in the computation. This yields to a maximum time difference between the measurements of 15 min. Night and day data are obviously included and the entire ScaRaB passes falling within the 30°E to 110°E, 35°S to 35°N region, over both land and ocean are used. Note also that such an approach circumvents a possible bias due to calibration uncertainties in the Meteosat channels.

4.2.2. Result of the Regression

[26] Figure 2 shows the scatter diagram between the Meteosat estimated and the ScaRaB measured longwave flux. This plot includes all the passes from 20 January to 31 March 1999 and one point out of four from the images is used yielding more than 600,000 points for the comparison. The correlation coefficient is high (0.97) and the overall root mean square error less than 10 W m−2. The linear regression slope between estimated and measured flux is almost 1 and the intercept is 0.55. The Meteosat estimation seems to underestimate slightly the ScaRaB measurement for the coldest points (below around 120 W m−2). When using shorter time frame for the comparison, the results are still in good agreement. For instance, using the data for one single day, that is over 40,000 points, the correlation coefficient is 0.98 and the rms. is 8.3 W m−2. Note that if the water vapor channel were not used, the overall agreement of the regression would fall down, with correlation and RMS of 0.94 and 12.7 W m−2, respectively. The scatter diagram in the latter case (not shown) also indicates that the slight underestimation over the coldest point would be even worse without using the water vapor channel measurement in the computation. On a day to day basis, the time series of the correlation and RMS (not shown) indicates no particular trend over the January–March period. The coefficients of the regression have been estimated for all the available data in January, February, and March as well as for the whole period and are reported in Table 3. Overall, the regression coefficients are stable during the INDOEX campaign suggesting a rather stable calibration of the Meteosat infrared channel with time. The Meteosat estimated longwave flux agrees well with the ScaRaB observed flux with a RMS of the order of 10 W m−2.

Figure 2.

Scatter diagram of the Meteosat estimated longwave flux (W m−2) and the ScaRaB measurements over the Indian Ocean, from 20 January to 31 March 1999. One point out of four is retained (620,803) and every group of five points is averaged on the plot. The correlation coefficient is 0.97. The root mean square error is 9.45 W m−2. The dashed line corresponds to the regression. Slope is 0.998 and intercept is 0.549.

Table 3. Regression Coefficients to Convert Meteosat Radiances Into Longwave Fluxa
 a0a1a2a3a4a5R2S (W m−2)Nobs
  • a

    The six first columns correspond to the coefficients. The squared correlation, root mean square errors, and the number of colocated points are given in the last three columns. Note that for the January–March period, one point out of four is included in the computation.


4.3. Longwave Radiation Products

[27] In the following of the paper, Outgoing Longwave flux (OLR also expressed as F) refers to the OLR estimated using Meteosat observations and the fourth set of coefficients for the whole data set presented in Table 3. Several longwave radiation products are derived from these OLR estimates. The Clear-Sky Greenhouse Effect (Gclear) over the ocean is defined as:

equation image

where σ is the Stefan-Boltzmann constant, Tsurf is the sea surface temperature and Fclear the clear-sky outgoing longwave radiation. Computation of Gclear is restricted to the ocean and an emissivity of one is used. Assuming the ocean surface to be a perfect blackbody yields a small (1%) difference with respect to including the real slight departure from blackbodyness [e.g., Inamdar and Ramanathan, 1998]. Owing to the SST frequency, only weekly mean Gclear maps are built with a spatial resolution of 2° × 2°. Fclear is obtained by averaging the Meteosat pixels that are flagged clear in the cloud classification process (section 2). The last radiation budget term we use concerns the longwave Cloud Radiative Forcing (LWCRF) [Charlock and Ramanathan, 1985; Ramanathan et al., 1989]. LWCRF is defined as:

equation image

[28] Cloud being generally colder than surface, LWCRF is usually positive and is associated with radiative heating of the atmosphere by clouds. Monthly mean LWCRF are built at a 2° × 2° resolution as well as daily mean over large regions and are used in section 5. In this case, F corresponds to the all-sky OLR and includes both the cloudy and clear-sky contribution. LWCRF is also estimated for individual cloud systems in section 6. In this latter case, F corresponds only to the radiation measured over the cloudy pixel.

5. The Intertropical Convergence Zone and Its Moist Environment During INDOEX

[29] The ITCZ is here associated with the region of high upper level cloudiness or maximum cloud zone [e.g., Sikka and Gadgil, 1980]. A dynamical formulation of the convergence zone in the lowest levels of the atmosphere and its characteristics during the campaign can be found in the work of Verver et al. (Overview of the meteorological conditions and atmospheric transport processes during INDOEX 1999, submitted to Journal of Geophysical Research, 2000, hereinafter referred to as Verver et al., submitted manuscript, 2000). An set of hydrological cycle related parameters is used in order to document the main characteristics of the Intertropical Convergence Zone during the experiment. These parameters are the Outgoing Longwave Radiation (OLR), the Sea Surface Temperature (SST), the Upper Level Cloudiness (ULC), the Upper Tropospheric Humidity (UTH), the Latent Heat release in the atmosphere due to Precipitation (LHP), the Precipitable Water (PW), the Longwave Cloud Radiative Forcing (CRF) and the clear-sky greenhouse effect (Gclear). Such an ensemble of parameters allows to document the tropical convection as well as its moist environment.

5.1. Monthly Mean Fields

[30] Figures 3 and 4 show the monthly means of the set of parameters for February and March 1999, respectively.

Figure 3.

Monthly mean of (a) Outgoing longwave radiation (W m−2), (b) Sea surface temperature (K), (c) Upper level cloudiness (%), (d) Upper tropospheric relative humidity (%), (e) Vertically integrated latent heating due to precipitation (W m−2), (f) Precipitable water (mm), (g) Longwave cloud radiative forcing (W m−2), and (h) Clear-sky greenhouse effect (W m−2) for February 1999.

Figure 4.

Same as Figure 3 but for March 1999.

5.1.1. February

[31] Figure 3a shows the OLR map. The ITCZ is characterized by low values (260 W m−2 and colder) and forms a slightly tilted zone, about 15° to 20° wide, crossing the full Indian Ocean from 40°E–15°S to 110°E–0°N. Two major centers of convection with very cold OLR are found over the Eastern flank of Africa off Madagascar, and over the maritime continent where the OLR is as low as 190 W m−2. Over the central Indian Ocean, the average OLR is around 250 W m−2. A cold tongue is found on the South Western part of the region with values around 240 W m−2. The ITCZ is surrounded by two warm regions in both hemispheres. These radiative sinks are characterized by maximum values up to 310 W m−2 over the African Horn and the Arabian Sea and correspond to the subsiding branches of the Hadley–Walker circulation. The Upper Level Cloudiness (Figure 3c) is defined as the cloud cover associated with all the non-low-level cloud detected in section 2. It is a measure of the total convective induced cloudiness (convective towers, stratiform anvils and thin cirriform shields). Maximum values (90–100%) of ULC corresponds to the active convection centers. Over the Himalayas, a local maximum of upper level cloudiness is found and should be interpreted carefully. Indeed, over there, cold temperature may be associated with the elevated terrain rather than with cloudiness. In order to avoid these problems, in the following of the paper, the region north of 25°N will be excluded from the analysis and the focus is drawn on the tropical cloudiness over both the ocean, eastern Africa, Madagascar and Indonesia. The estimate of cloudiness overall agrees well with monthly mean upper cloudiness obtained with different techniques (Sèze and Pawlowska, submitted manuscript, 2000). The vertically integrated latent heating map (Figure 3e) confirms the maximum of convective activity over southeastern Africa and Indonesia where maxima of precipitation are found. There the associated heating exceeds 300 W m−2 while it is negligible over the subtropical regions. The central Indian ocean ITCZ is characterized by less intense precipitation on the average. As expected the ULC spreads over larger spatial region, unlike precipitation and OLR, indicating a contribution from thinner and nonprecipitating cloudiness to the total convective induced cloudiness. The longwave Cloud Radiative Forcing maps (Figure 3g) indicates that over the major convective regions, the atmospheric longwave radiative heating is close to 80 W m−2 in agreement with climatological values over this region [Rajeevan and Srinivasan, 2000]. Over these major centers of convective activity, the ratio of the latent heat forcing to the longwave radiative forcing is roughly about 3 in agreement with previous estimations for the winter monsoon [Wilcox and Ramanathan, 2001]. Overall, these different indicators of convection show a large internal consistency. Figure 3b shows the monthly mean sea surface temperature map. The maximum temperature of 302 K are found around Madagascar, close to Indonesia and in the central Indian Ocean. These maxima are embedded in a much larger pool of SST greater than 301 K that covers most of the region where deep convection is located. The Arabian Sea is characterized by a Southeast Northwest gradient of a few Kelvin. Minimum temperature as low as 291 K are found over the south eastern part of the ocean, where a large region (70°E–110°E:35°S–20°S) is colder than 298 K. Figure 3d shows the monthly mean UTH map. The moister regions with 70% humidity are located over the active convection region, while dryer regions are found in the subtropical Arabian Sea and South East ocean. The UTH and the large-scale dynamics are well correlated at the monthly scale: subsidence is associated with the dry areas and moist areas correspond to mean upward motions [Picon and Desbois, 1990]. The Hadley Walker subsiding branches are hence further identified in Figure 3d and associated with the radiative sinks of Figure 3a. The precipitable water map (Figure 3f) indicates a well shaped moist region surrounded by two drier areas. The maximum reaches up to 55 mm over a large longitudinal band centered at 5°S. Owing to the exponential decrease of specific humidity with height in the tropics, the precipitable water mainly reflects the boundary layer water vapor loading. Accordingly, over the coldest part of the Indian Ocean, the columnar burden of moisture is diminished with respect to the moister areas by almost a factor of two. Nevertheless, for similar SSTs, the northern subtropical regions appear dryer than in the Southern hemisphere. This is likely due to the low-level advection of dry continental air by the monsoon trades which are northeasterly during the winter (Verver et al., submitted manuscript, 2000). The monthly mean clear-sky greenhouse effect, noted Gclear, is shown in Figure 3h. Gclear integrates the radiative effect of both the moisture in the lower level estimated through PW but also the UTH contribution. Accordingly the high Gclear regions correspond well with the two moisture estimate maps and maxima are found over the moister region with values in excess of 200 W m−2.

5.1.2. March

[32] The SST increases all over the region and the 302 K isothermal area extends now over the full longitudes (Figure 4b). The cold OLR pool over Indonesia spreads over a larger area and the minima decreases to below 200 W m−2 (Figure 4a). Consistently, ULC shows an overall increase marked over the central open Indian Ocean and over the Maritime continent with respect to February. The African convection is shifted to north in agreement with the seasonal migration of the ITCZ over this region [Picon et al., 1995]. The tilt of the ITCZ is weaker than in February. Precipitation (Figure 4e) also increases and regional shifts are in agreement with the other convection indicators. The longwave cloud radiative forcing is accordingly stronger and the associated structures follow the ULC evolution (Figure 4g). On the other hand, the northern subtropical subsidence branches of the Hadley–Walker circulation are intensified and spread over an extended area (Figures 4a and 4d). The asymmetry between the northern and southern hemisphere columnar water vapor burden is reinforced (Figure 4f). The February–March UTH differences (Figure 4d) indicates moistening of the ITCZ area. The signature of both the precipitable water and the UTH on to the radiative budget are integrated in Gclear (Figure 4h) and it indicates a global increase of the clear-sky greenhouse effect. This positive perturbation of the hydrological cycle in March with respect to February is related to the passing of intraseasonal oscillation over the Indian Ocean during the campaign. The consistency of our multisatellite data set is further investigated by considering the regional relationship between the cloudiness and radiation parameters and the moisture and clear radiation parameters next. The cross relationships between convection and upper tropospheric moisture are also investigated.

5.2. Regional Relationship Between Upper Level Cloudiness, the Moist Environment, and the Radiation Field

[33] Figure 5a shows the scatter diagram of the upper level cloudiness versus the total sky greenhouse effect for February and March. The two quantities are well correlated and almost linearly related. Region with low monthly mean upper level cloudiness (20%) corresponds to greenhouse effect around 160 W m−2 while over overcast areas the longwave energy trapped in the atmosphere reaches 300 W m−2. Figure 5b shows the scatter diagram of ULC and LWCRF. The correlation is 0.93 and the scatter is small (rms. of 6.4 W m−2). Maximum value of around 70–80 W m−2 for overcast areas with a slight departure from linearity. This relationship agrees well with the Summer Monsoon results deduced from ISCCP and ERBE presented by Rajeevan and Srinivasan [2000]. As one can expect, the precipitable water and the clear-sky greenhouse effect are well correlated (R = 0.93) over the Indian Ocean (Figure 5c). The relationship between UTH and clear-sky greenhouse effect (Figure 5d) is nonlinear and the two observed fields are well correlated. This nonlinearity is stronger in the region of UTH below 20%. An increase of UTH from 10% to 20% is accompanied by an increase of Gclear from around 155 to 170 W m−2 while a change in UTH from 50% to 60% yields only a minor 5 W m−2 increase in Gclear. It is the signature of the well known OLR nonlinear sensitivity to moisture perturbation in a dry background with respect to moist environment [Udelhofen and Hartmann, 1995; Spencer and Braswell, 1997].

Figure 5.

Scatter diagram of the total sky greenhouse effect and upper level cloudiness (a). Scatter diagram of the longwave cloud radiative forcing and upper level cloudiness (b). Scatter diagram of the clear-sky greenhouse effect and precipitable water (c). Scatter diagram of the clear-sky greenhouse effect and upper tropospheric humidity (d). Dot corresponds to February and plus to March. Dashed lines correspond to the regression curve.

[34] Figure 6 shows the relationship between upper tropospheric humidity and upper level cloudiness. In Figure 6a, the upper level cloudiness from all high level cloud type but for the thin cirrus is considered. Strong correlation (R = 0.89) and rms. of 6.2% is found. This indicates that UTH is a linear function of cloudiness with a slope of 0.67 and an intercept of 13.4. The equivalent relationship for the tropical Americas obtained by Udelhofen and Hartmann [1995] is overlaid on the graph (long dash curve). The agreement is excellent. Interestingly enough, the same kind of relationship seems to hold between upper level cloudiness without thin cirrus over the Indian Ocean and over central Americas. This is of major importance to General Circulation Models evaluation; if such a relationship is indeed similar at the global scale it could provide a useful and easy way to implement validation test. This agreement also strengthens the consistency of our data set on both the clouds and moisture aspect. Figure 6b shows the UTH-ULC scatterplot. The thin cirrus contribution is now included in the upper level cloudiness. The correlation is stronger (r = 0.92) and the scatter smaller (rms. of 5.2%) suggesting that the full high level cloudiness has to be considered to understand the UTH distribution. The positive correlation between upper level cloudiness and UTH presented here confirms the idea of a local moistening by convection in the tropical regions [Soden and Fu, 1995] for the Indian Ocean region.

Figure 6.

Scatter diagram of upper tropospheric humidity and upper level cloudiness. For all but for the thin cirrus cloudiness (a). For all the upper level cloudiness in (b). Dot corresponds to February and plus to March. Dashed lines correspond to the regression curve. The long dash line in (a) is the equivalent linear relationship of work by Udelhofen and Hartamann [1995]. See text for details.

5.3. Temporal Variability of Upper Level Cloudiness, Upper Tropospheric Humidity, and Longwave Radiation

[35] The temporal variability of the cloudiness, moisture and radiation fields is now examined. Two regions are considered. The first region corresponds to the full area of study limited to the 25°N latitude and encompasses the centers of strong convective activity. The second region is a restricted portion of the open central Indian Ocean spanning the 60°E–80°E:20°S–10°N area. This region is representative of the oceanic ITCZ conditions and exhibits strong variation between February and March. Daily means are built from the hourly UTH products and from the half-hourly cloudiness and radiation products. The SST background needed to compute the clear-sky greenhouse effect and the clear-sky OLR needed for the longwave cloud radiative forcing estimation are weekly averages. Figure 7 presents the daily time series of the different parameters and their radiative counterparts for the first region considered. Over land and ocean (Figure 7a), both ULC and UTH decrease during the two first weeks of February from 40 to 25% and 30 to 20% respectively. A peak appears in both fields around 22 February and slight variability is observed after with a small increase around 6 March. The UTH and ULC time series are correlated (R = 0.75) confirming the local upper tropospheric moistening by convection. Figure 7b shows the daily time series of the LWCRF. From the first week of February to the second week of March, the LWCRF increases from 17 W m−2 to 25 W m−2. As expected, the variability of ULC and the LWCRF time series are well correlated (R = 0.91). When only oceanic regions are included in the daily average (Figures 7c and 7d), the variability is stronger and the correlation between the cloudiness, UTH and radiation increases. On 6 March, an abrupt increase is associated with the passage of the intraseasonal convection perturbation over the Indian Ocean. Using OLR anomalies analysis, Rasch et al. (Understanding the Indian Ocean Experiment INDOEX aerosol distribution with an aerosol assimilation, submitted to Journal of Geophysical Research, April 2000) show that the intraseasonal perturbation starts impacting the INDOEX region around that day. The correlation between ULC and UTH is similar to the full region result. The relationship between UTH and Gclear is weak (R = 0.4). This is to be expected considering the importance of the large-scale dynamics in the subtropical regions free tropospheric humidity which are here added to the moist tropical band [Sherwood, 1996; Pierrehumbert and Roca, 1998]. As shown next, over the smaller ocean region, where only the tropical ITCZ region is considered, the correlation improves. The last two panels of Figure 7 shows the ULC, UTH and LWCRF time series only over land. The plot reveals a variability of the island based and African convection during the period weaker than over the oceanic region. On 22 February, a peak is seen over land but the Madden Julian oscillation perturbation is smoothed out. A detailed regional analysis separating African convection from the Maritime convection is out of the scope of the present paper but would provide the further analysis needed on that point.

Figure 7.

Daily time series over the 30E–110E:35S:25N region. For upper level cloudiness (plain line) and upper tropospheric humidity (dotted line) in (a), (c), and (e). For longwave cloud radiative forcing (plain line) and clear-sky greenhouse effect in (b), (d), and (f).

[36] Figure 8 shows the same analysis for the central open Indian Ocean region. The impact of the intraseasonal oscillation is clear from 8 to 28 March in the ULC time series and coincides with a similar perturbation in UTH (Figure 8a). More than 90% of variability of the UTH series is explained by the ULC variability (R = 0.96). The ULC and LWCRF are strongly correlated (R = 0.91) and the UTH variability over this region is well related to the Gclear variability shown in Figure 8b with a correlation coefficient of 0.92. From 15 February to 18 March, ULC increases from around 15% to 65% and the UTH from 15% to 46%, respectively. The LWCRF increases from 3 to 33 W m−2 and the clear-sky greenhouse effect from 165 to 187 W m−2. This analysis of the temporal variability reveals strong positive relationships between cloudiness, UTH and longwave radiation. The local moistening of the upper troposphere by convection is here confirmed over the Indian Ocean. The tight correlations, both spatial and temporal, between cloudiness and longwave forcing as well as between UTH and clear-sky greenhouse effect have been quantified. In the next section we investigate the mechanisms relating these quantities together considering the properties of the mesoscale organized convective systems.

Figure 8.

Daily time series over the 60E–80E:20S:10N region. For upper level cloudiness (plain line) and upper tropospheric humidity (dotted line) in (a). For longwave cloud radiative forcing (plain line) and clear-sky greenhouse effect in (b).

6. The Convective Systems Forming the ITCZ

[37] The focus of the present section is on the individual convective systems forming the ITCZ and their environment. First, the method for retrieving the individual convective systems is briefly presented. Then, the systems population is described. The structural and radiative properties of each system categories are characterized. The regional and temporal variability of the systems are presented and related to its moist environment and to the radiation fields described in section 5.

6.1. Retrieval of the Individual Convective Systems

[38] As stated in the Introduction, a simple conceptual model of the individual convective systems is used here. It reads as follows: surrounding a convective core where violent updrafts and heavy precipitation are taking place, stands the extended stratiform anvil composed of different type of cloudiness from thick to thin cirrus. The aim of this section is to present the algorithm exploited to extract the convective systems from the satellite measurements following this simple cloud model.

6.1.1. Extraction of the Convective Systems: A Modified Version of the Detect and Spread Algorithm

[39] The spatial structure of the convective systems are obtained by using the Detect And Spread (DAS) algorithm [Boer and Ramanathan, 1997] that allows to extract individual cloud systems from the satellite imagery. It is a multithreshold multistage approach to delineate the cloud clusters from the satellite IR imagery and can be thought of as a generalized cloud clustering method. This technique has been successfully used for Western Pacific analysis [Boer and Ramanathan, 1997], Indian Ocean monsoonal convective systems study [Roca and Ramanathan, 2000, hereafter referred as RR] as well as for high resolution regional model validation [Zhang et al., 1999]. In a previous study, RR introduced a modified version of the technique dedicated to the upper level cloudiness. Owing to the limited INSAT data set available at that time, the upper cloudiness was defined with a single threshold in IR brightness temperature. Here, we introduce a further modified version of the method, where the DAS algorithm is applied on the preclassified image presented in section 2. This allows to take into account, in the clustering approach, the full upper level cloudiness including the warm semitransparent pixels. The modified DAS is run in four steps. The first step concerns the convective core detection. All the contiguous pixels with brightness temperature colder or equal than 240 K are associated with the cloud convective core. This threshold is classically associated with active convection and/or precipitation [e.g., RR]. After this detection step, the pixels surrounding the convective core are searched for. If the brightness temperature does not exceed 260 K, the new pixels are attached to the core. This is the spreading stage. The process is reiterated two more times with thresholds of 255 K and 270 K for the detection steps. The clouds are spread to 275 K and to clear sky, respectively. Note that the spreading step is limited to the cloudy pixels classified as mid-to-upper and thin cirrus cloudiness. The 5 K overlap between the last spread threshold and the next detection step allows to grow all the existing clusters to its clear-sky edge. Each step is run in multiple substeps to avoid clouds to growing around an existing cloud (for details, see Boer and Ramanathan, 1997, Annex 1). It should be mentioned at this point that some slight modifications were drawn on the original code. The original hit-and-miss loop over the image as been replaced by a dilatation operator [Serra, 1988]. The dilatation operator is applied on the detected clusters and is restricted to the preclassified cloudy pixel space. The criterion on brightness temperatures range still constrains the dilatation. This modification preserves the original philosophy of the approach. A number of parameters both structural and radiative are estimated for each cloud systems. These include the coordinates of the system centroid and its local time. Structural cloud properties that are archived include the total area of the cloud. The areal cloudiness is further split into the area colder than (if any) 210 K, 220 K, 240 K, 255 K and 270 K within a system. The area of the system formed by the thin cirrus cloudiness is also archived. The cloud parameters are completed by the convective system radiative properties. The latter include the mean IR brightness temperature averaged over the full cluster, the mean OLR, as well as the minimum and maximum brightness temperature in both the IR, the WV channel. The minimum brightness temperature is used to form cloud system categories (see next subsection). The box that surrounds any given cluster is also archived by keeping track of the minimum and maximum longitude and latitude encountered over a cluster. This is done in order to identify the region where the LWCRF is estimated. Finally, the difference between the WV and IR brightness temperature over the coldest pixel in terms of IR temperature, of the convective system is measured together with the area within a cloud where this difference exceeds 5 K. These last WV/IR parameters are used in as convective overshoot first order indicators [Schmetz et al., 1997].

6.1.2. Convective Systems Classification

[40] This last step of the processing aims at estimating the vertical characteristics of the cloud systems to complete the analysis. Once the individual cloud systems or clusters horizontal structures have been identified, they are further classified into three major categories according to their vertical extent. The minimum temperature encountered within a cloud cluster is used as a criterion for building the categories. At the Meteosat resolution (about 5 km), this coldest pixel is associated with the deepest convective cell present in the convective core. These major convective cloud types have been further split into subcategories. The two first classes (class Ia and Ib) include active convection that operates in the mid-to-upper tropospheric region. The two subsequent classes (Class IIa and IIb concern convection reaching the lower to mid levels as well as the detached decaying anvils, so-called convective debris [e.g., Websters and Stephens, 1980]). These types of clouds are mainly found surrounding the class I systems. Finally the cloud systems that only include thin cirrus cloudiness (that is their minimum IR brightness temperature is warmer than 270 K) form the class III and can be associated with anvil debris. These convective systems categories are summarized in Table 4.

Table 4. The Five Categories of Convective Systemsa
Cloud categoryTemperature criterionComments and naming convention
  • a

    TIRmin is the minimum infrared brightness temperature of the cloud system.

Class IaTIRmin < 210 KVery deep convection
Class Ib210 K ≤ TIRmin < 240 KDeep convection
Class IIa240 K ≤ TIRmin < 255 KMidtropospheric convection and debris
Class IIb255 K ≤ TIRmin < 270 KLow-to-mid troposphere and debris
Class IIITIRmin ≥ 270 KDetached thin cirrus

[41] Figures 1d, 1e, and 1f show an example of the individual cloud retrieval. An arbitrary random color is associated with every single cloud system. Class I clouds are found in the cold patches seen in the IR imagery (Figure 1a). The cold large area centered over Indonesia has been split into three individual systems. The Davina cyclone has been well detected (90E–12S). Class II and III correspond to warmer cloud systems and are found over most of the Indian Ocean. The class II cloud systems are found in the surroundings of the large Class Ia systems. The class III clouds are mainly composed by very small cluster of a few pixels scattered over the ocean.

6.2. Convective Systems Population and Associated Cloudiness

[42] Figure 9a shows the frequency of occurrence of the convective systems as a function of the convective cloud area for the full period February–March. The convective cloud area refers to the full spatial extent of the systems including the convective core, the stratiform anvil and the cirriform contribution to the cloudiness shield. The analysis is restricted to the cloud encompassing at least 10% of their extent over ocean. This is done in order to eliminate the Northern Indian continental regions from the statistics and as well to include the convective systems present over the island regions. Furthermore, we here restrict the analysis to the clouds larger than 5 pixels (80 km2) given the uncertainty attached to smallest clouds [Boer and Ramanathan, 1997]. Note that their contribution to cloudiness and radiative forcing is negligible. Under these considerations, the spectrum of detected cloud systems spans a wide range of scale from 102 to more than one million of squared kilometers. The detached thin cirrus cloud category (Class III) are mainly found in the smaller scale range and never exceeds 2.105 km2. The Class Ib and II cloud systems have a similar frequency of occurrence at all scales. The peak is around 103 km2 and the upper limit of these systems is 106 km2. The class Ia systems, i.e., the very deep convective systems, span the full spectrum of scale with significant occurrence over 106 km2. The class II and III systems dominate the smaller scales (area <50,000 km2) by several orders of magnitude when compared with class I while class I dominates the larger scales at most by an order of magnitude. The present distribution appears slightly shifted toward the larger scales with respect to the results of Roca and Ramanathan [2000] which is consistent with the inclusion of the thin cirrus into the convective cloud systems. Complementary to the frequency of occurrence, Figure 9b shows the cumulated contribution to the upper level cloudiness as a function of the convective cloud area. In spite of a high frequency of occurrence, Class Ib and II cloud systems only contribute between 5 and 15% to the upper level cloudiness. Class III clouds cumulated contribution is less than 1%. The less frequent Class Ia systems nevertheless dominate the upper level cloudiness with total contribution in excess of 65%. Up to around the 2.105 km2 scale, Class II cumulated contribution are larger than the very deep convective Class I systems. In the larger part of the spectrum, contribution from Class Ia clouds is overwhelming.

Figure 9.

The frequency of occurrence of the convective systems (a) and cumulated contribution to the cloudiness (b) as a function of the convective cloud area. The Class Ia clouds are in plain, Class Ib in dots, Class IIa in dash, Class IIb in dash-dot-dash, and Class II in dash-dot-dot-dot-dash style.

6.3. Characterization of the Individual Convective Systems

[43] The structure of the systems is first discussed in terms of brightness temperature partitioning. The signature of this partitioning on the radiative properties of the convective systems is then derived. Finally, the environment above the coldest cells of the convective systems is further investigated using the WV channel to complement our convective system characterization. Properties are averaged over the full February–March period and are presented for all the classes as a function of the convective cloud area. Only the area bins containing more than fifty occurrences of a given cloud system are included.

6.3.1. Structural Properties

[44] The fractional contribution of cloudiness colder than a given threshold to the total cloudiness of a given convective system is an important structural parameter. It allows to document the distribution of the different cloudiness type, with different optical properties, within a convective system. The IR brightness temperature provides a useful indicator of this optical variation. Figure 10 shows the fraction of the systems colder than several IR thresholds as a function of the convective cloud area. By definition (see Table 4), class III systems are only composed of thin cirrus. Hence, they do not appear on the Figure 10. The fraction of thin cirrus within Class II clouds increases with the system area and asymptotes toward 100%. This indicates that the largest Class II clouds are associated with extended warm cloudiness likely to be associated with decaying anvils or convective debris [Websters and Stephens, 1980]. The fraction of cloudiness colder than 270 K within the smaller Class II clouds can be as high as 55%; for Class IIa clouds, 20% of their area is colder than 255 K. This suggests that Class II small clouds are not pure detached anvils and can also include some convective systems reaching the mid troposphere. The percent of cold cloudiness within a Class I system increases with its area for the systems larger than 104 km2, while it decreases for the smaller scales. A typical large (area > 106 km2) Class Ia systems is composed of 7% of thin cirrus cloudiness, 61% of cloudiness colder than 240 K, the remaining 32% being associated with stratiform anvil of temperature comprised between 240 and 270 K. The very deep convective cells colder than 210 K form around 10% of its total area. The signature of the structural properties of the systems on their longwave radiative properties is investigated next.

Figure 10.

Fractional distributions of temperature colder than a threshold within a system as a function of the convective cloud area. Thresholds are 210 K (a), 220 K (b), 240 K (c), 255 K (d), and 270 K (e). The fraction of thin cirrus within a system as a function of the convective cloud area (f). Linestyle is the same as Figure 9.

6.3.2. Radiative Properties

[45] Figure 11 summarizes the radiative properties of the convective systems forming the ITCZ during the INDOEX campaign. The scale dependence of the IR brightness temperature averaged over the cloud (Figure 11a) indicates that class II clusters tend to have warmer average temperature as the convective cloud area increases. Class III cloud exhibits no significant variation of their mean temperature with size. On the other hand, the deep and very deep convective systems mean temperature increase with size for the smaller scale range (area smaller than around 104 km2). On the upper range of cloud sizes, the trend is reversed and the cloud mean temperature cools as the cloud area increases. Similar trends are evidently present in the OLR (Figure 11b). The average OLR of class Ia systems larger than one million of km2 is as low as 160 W m−2. The impact of these variations of the OLR with the systems spatial extent is measured by the longwave cloud radiative forcing scale dependence presented in Figure 11c. The Meteosat derived clear-sky OLR is used as a background for the computation of the forcing; weekly averages are used. The bounding box of the convective system under consideration is a parameter retrieved from the cloud identification process that defines the localization of the cloud. The forcing is obtained by subtracting the instantaneous cloud average OLR to the average of the weekly mean clear-sky OLR over the bounding box. Shorter timescale for the background OLR have not been considered owing to the difficulty of finding clear sky in the convective area. Larger timescale, like monthly means of the clear-sky OLR can be used [e.g., Wilcox and Ramanathan, 2001]. As shown in section 5, significant intraseasonal variability is observed in the clear-sky greenhouse effect. Using weekly average for the clear-sky background radiation allows to consider this variability in the forcing computation. The half-hourly sampling of the Meteosat ensures that all of the Indian Ocean has been sampled in clear-sky conditions over weekly timescale. Figure 11c shows that Class III systems average LWCRF do not exceed 10 W m−2. Class IIa (b) cloud systems LWCRF decreases from 60 (40) W m−2 for the smaller scale to around 20 (15) W m−2 at the end of the area spectrum in consistent manner with respect to their structural scale dependence properties. The Class I systems have the LWCRF values larger than 60 W m−2 at all scales. In particular, the Class Ia clouds are associated with the largest forcing with an average LWCRF for systems larger than 106 km2 of 110 W m−2. The longwave cloud forcing normalized to the individual clouds integrates the different type of cloudiness (optically speaking) that forms each individual system. The combined effects of the cloud frequency of occurrence and the cloud spatial extension together with the individual cloud radiative properties on to the winter monsoon longwave forcing are shown in Figure 11d. Class III clouds high frequency of occurrence in the small scales does not balance their weak radiative effect and their cumulated contribution to the longwave forcing is negligible (less than 1%). The Class IIa(b) has a nonnegligible contribution to the total forcing of 17 (12) %. The main contribution nevertheless comes from the Class I clouds systems and reaches 70% with 38% and 32% from Class Ia and Class Ib respectively. The former are more important in the upper range of cloud area. At scales smaller than 105 km2, the forcing of the class II systems is more important than the Class Ia. At larger scale, the convective and very convective cloud systems dominate the longwave radiative forcing. In summary, the Class I system appear as the most important cloud class to the radiative budget. Their deep convective systems nature being made clear, we shall further characterize their properties with emphasis on the convective overshoots.

Figure 11.

The convective systems radiative properties as a function of the convective cloud area. IR average temperature (a), average outgoing longwave radiation (b), longwave cloud radiative forcing (c), and cumulated contribution to the longwave forcing (d). Linestyle is the same as Figure 9.

6.3.3. Convective Overshoot Properties

[46] In order to better describe the very deep convective systems, the class Ib clouds (cloud with a minimum temperature colder or equal than 240 K and warmer than 210 K) is split into two classes. Clouds with a minimum temperature colder or equal than 220 K form class Ib while Class Ic refers to the remaining part of the cloud population (Minimum temperature colder than 240 K and warmer than 220 K). Figure 12a shows the minimum temperature encountered within a convective system for class Ia, Ib and class Ic. This parameter was previously used to build of the cloud classes. Owing to the very high spatial resolution of the Meteosat observations (5 × 5 km at nadir), this minimum temperature pixel can be associated with the deeper individual convective cell within the convective systems. For all the classes, as the convective cloud area increases, the deeper convective cell deepens. This trend is well marked for the class Ia systems. Individual convective cells as deep as 190–185 K, are found only in organized convective systems with areas exceeding one million squared kilometers. These temperatures are characteristic of the tropopause temperature over Indonesia during February and March [Shimizu and Tsuda, 2000], suggesting that the tropopause temperature would be driven by organized convective systems as advocated in RR.

Figure 12.

The convective systems overshoot characteristics as a function of the convective cloud area. Minimum IR blackbody temperature within a cloud (a). Difference WV-IR equivalent blackbody temperature at the location of the minimum (b). Absolute area of overshoots (c). Fractional area of overshoots. Note that the line corresponds to Class Ia (plain), Class Ib (dotted), and Class Ic (dash). See text for details.

[47] Recently, Schmetz et al. [1997] proposed an approach for determining the convective overshooting from the Meteosat measurements in both the IR and the WV channel. This technique relies on the brightness temperature difference between the WV and the IR channels (TWV − TIR). Positive differences up to 6–8 K are indeed commonly observed over tropical Africa and the Atlantic Ocean. The two channels sensitivity to the cloud microphysics, neither the Planck function nonlinearities at the subpixel scale, nor the vicarious calibration of the WV channel can explain such a large positive difference [Schmetz et al., 1997]. They hence suggest that positive difference results from the presence of water vapor above the cloud tops. This stratospheric moisture would absorb the upwelling cold radiation from the cloud top. Assuming a concomitant lapse rate sign change at the tropopause, the humidity molecules would radiate toward the satellite at warmer temperatures. The 6.3 μm channel is strongly influenced by moisture and would finally measure a warmer brightness temperature than the IR window channel. They link this moisture with convective overshooting and moisture injection in the stratosphere. The presence of these so-called warm water vapor pixels is confirmed over the Indian Ocean. The difference between the 6.3 μm and the window channel (TWV − TIR) over the coldest pixel of the convective systems, that is the TWV measured over the deepest cell of the convective systems minus the TIR of the cell, has been computed for each cloud systems. This is chosen as an indicator of convective overshooting. Figure 12b shows this difference as a function of the convective cloud area for Class Ia, Ib, and Ic. For the Class Ic cloud systems, the overshoot indicator is always negative, increasing from −7 K to almost zero for the larger systems. This class deepest cell indeed never reaches temperature colder than 230 K (Figure 12a). Class Ib systems overshoot is around 3 K over most of the scale except for the smaller (up to 7.5 K for the cloud systems with area below 103 km2). Above 105 km2, the Class Ib difference increases from 3 to 6 K. By definition these clouds deepest cell temperature is colder than 220 K and is shown (Figure 12a) not to be colder than 208 K over the whole spectrum of scale. Class Ia differences are larger than 5 K over the whole spectrum and reach up to 14 K for the largest system for which the temperature of the cloud deepest cell is 183 K.

[48] The winter monsoon climatological tropopause temperatures average around 190 K over Indonesia [Shmizu and Tsuda, 2000] and around 200 K over the Indian Ocean [Highwood and Hoskins, 1998]. These tropopause temperatures are hence reached by the deepest cell of the organized convective systems of class Ia where the overshoot value is larger than 5 K. Recall that climatological values are monthly means while our cell temperature are instantaneous. Further work is needed to explain this 5 K limit, but the WV channel calibration based on cross satellite intercomparison might be biased. This could explain the large value of the overshoot presented here. Consistently, we define an overshoot by the region where the difference (TWV − TIR) is larger or equal to 5 K.

[49] We now focus only on class Ia systems, that is the systems with at least one pixel colder than or equal to 210 K and where the so-defined overshoot exists. Figure 12c shows the absolute area of overshoot within a cloud as a function of the convective cloud area. It increases monotonically with the convective cloud size, indicating that a larger cloud system overshoots over a larger area. The fractional overshoot area (Figure 12d) confirms the breaking scale of around 104 km2 in the convective cloud properties. Below the breaking scale, the fraction of overshoot decreases with the cloud area, while it increases with the scale for the systems larger than 104 km2. At the breaking scale, only 4% of the system area overshoot. The areal overshoot never exceed 12% of the class Ia system over the full spectrum of scale.

6.4. Spatial and Temporal Relationship Between Convective Systems and Their Upper Tropospheric Moisture Environment

[50] According to the preceding analysis, the main difference in cloud scale properties takes place around a critical area of around 104 km2. This scale of 100 km is usually associated with mesoscale organized convection [e.g., Redelsperger, 1997]. In this subsection, we investigate the regional relationship between these organized convective systems (scale larger than 100 km) and their large-scale moist environment. Monthly means are considered. Figure 13a relates the upper level cloudiness induced by the convective systems larger than 104 km2 (all classes are included) to the upper tropospheric humidity. As expected, the correlation is good (R = 0.9) and the rms. is similar to the one found in section 5. This suggests that the positive spatial relationship between upper level cloudiness and upper tropospheric relative humidity is driven by the individual cloud system larger than 104 km2. Figure 13b further presents the scatter diagram of the mean number of systems of these organized systems per slot versus UTH. The frequency of occurrence is computed for each 2° × 2° grid box. A system spreading over more than one grid box is counted in each of the box it covers. This reinforces the idea of the larger convective clouds being spatially well related to upper level moisture. This also indicates that not only the cloud size is an important parameter but also outlines the role of the frequency of occurrence. Hence, the results suggests that the previously mentioned local moistening of the upper troposphere by convection at the regional scale is to be attributed to organized mesoscale convective systems over the Indian Ocean.

Figure 13.

Scatter diagram of the upper tropospheric humidity and the upper level cloudiness induced by the systems larger than 104 km2 (a) and the upper tropospheric humidity and the frequency of occurrence of the systems larger than 104 km2 (b). Dot corresponds to February and plus to March.

[51] We now investigate the temporal relationship between convective systems and upper tropospheric humidity at the convective cloud scale together with their radiative impact. Owing to their overwhelming role within the organized convective systems, the emphasis is put on the more vertically developed convective systems, i.e., the Class Ia clouds. Figures 14 and 15 represent the daily mean time series of the total number of systems for the two regions introduced section 5. Over the whole region, the intraseasonal variability in the upper level cloudiness shown in section 5 is well related to the frequency of occurrence variability of the Class Ia organized systems (Figure 14a). Indeed, the frequency explains 55% of the ULC variance. This agreement is mainly due to the clouds over the oceanic region where up to 70% of the variance of the ULC is explained by the organized systems frequency of occurrence shown in Figure 14b. While over the land regions (Figure 14c) the correlation is poor. Accordingly, the LWCRF variability presented in Figures 7b, 7d, and 7f is related to the system frequency with correlation of 0.85, 0.87 and 0.25, respectively. As far as the cloud moist environment is concerned, the correlation between the systems frequency and UTH is overall lower: 0.23, 0.45, and 0.07 over the whole area, the ocean and the land, respectively. Accordingly, over the ocean, the frequency of large Class Ia systems explains 47% of the variance of Gclear shown in Figure 7d. The fact that the relationship over this large region between the convection related fields and the UTH is weak, is due to the inclusion of the subtropical regions in the spatial average. Indeed it was shown that subtropical free tropospheric moisture is controlled by large-scale dynamics complementing the convection source [Pierrehumbert and Roca, 1998; Soden, 1998]. Over a smaller region over the open ocean that excludes the subtropical low humidity areas, all the temporal correlations improve (Figure 15). The frequency of occurrence of the class Ia systems explains up to 84% and 77% of the variability of UTH and Gclear over this region. The cloud and cloudy radiation parameters have similar correlation that the ones for the larger region.

Figure 14.

Daily time series for the region 30E–110E:35S:25N. The plain line corresponds to Class Ia mean minimum IR temperature and the dotted line to the frequency of occurrence of Class Ia systems. (a) Over the whole region. (b) Oceanic regions only. (c) Continental regions only. A 7-day running mean has been applied on the raw daily curves.

Figure 15.

The plain line corresponds to Class Ia mean minimum IR temperature and the dotted line to the frequency of occurrence of Class Ia systems. A 7-day running mean has been applied on the raw daily curves.

[52] The mean minimum temperature of the Class Ia systems daily variability is also plotted on Figures 14 and 15. It indicates the changes in the mean penetration height of these systems during the INDOEX campaign. From the first week of February and the second week of March for instance, the mean deepest cell temperature of the systems drops from 201 K to 196 K. This parameter is anticorrelated with the frequency of occurrence. The impact of the property of these systems on the moist environment and radiation indicates that the colder and deeper systems are associated with a moister upper troposphere, a larger Gclear and a larger LWCRF. Over the Central Indian Ocean, the correlations between the mean minimum temperature and UTH, Gclear and LWCRF are −0.7, −0.84, and −0.77, respectively. This suggests that the intraseasonal perturbation of convection observed over the Indian Ocean in the upper level cloudiness, the UTH and the radiation fields is not only due to the modification of the frequency of occurrence of the Class Ia systems, but is also associated with a deepening of the later convective systems.

[53] The more delicate interpretation of these relationships over the land regions and the weaker correlation may be related to the difference between land and oceanic convection in their moistening of the upper troposphere. Over the Eastern Pacific and Central America region, it was shown that land based convection relationship with UTH were different from the oceanic convection [Salathé and Hartmann, 1997]. In particular the diurnal cycle convection and UTH appear to be in phase with a slight lag of 2 hours over the land and outphased by 12 hours over the ocean [Udelhofen and Hartmann, 1995; Soden, 2000]. Relating the above mentioned differences over the land and oceanic region to their respective diurnal cycle for the Indian Ocean deserves further work and is a topic of future research.

7. Summary and Conclusions

[54] An overview of the convection activity and its moist environment during the INDOEX campaign is provided through a satellite perspective. A simple yet efficient bispectral IR/WV cloud scheme is introduced. The quality of the thin cirrus detection is shown to be enhanced by the use of the WV channel. Statistical relationships relating the two Meteosat infrared channels radiances to the ScaRaB longwave flux are obtained. The Meteosat derived flux agrees well with ScaRaB measurements with a RMS of 10 W m−2. Together with other data set and satellite derived meteorological products, an extended ensemble of convection related parameters is achieved. The OLR, upper level cloudiness, latent heat release and the longwave radiative forcing all show strong spatial and temporal consistency. The moist environment is described using upper tropospheric humidity and precipitable water together with the clear-sky greenhouse effect and exhibits coherent relationships as well. Owing to the high spatial and temporal resolution of Meteosat, the convective cloud systems are analyzed. Structural and longwave radiative properties at the cloud scale are established. Using the WV channel, the environment above the deepest turrets within the systems are further documented. Finally, spatial and temporal characteristics of the organized convective systems are related to the upper tropospheric moisture distribution and to the radiation fields. Specifically, the study shows that over the Indian Ocean:

  1. The Upper Level Cloudiness and the Upper Tropospheric Humidity show a linear positive relationship in average over the whole Indian Ocean.
  2. The signature of an intraseasonal perturbation on the upper level cloudiness, upper tropospheric humidity and longwave radiation is estimated: from the inactive to the active phase of the perturbation, the central Indian Ocean ULC increases by 10%, the UTH by 20%, the LWCRF by 10 W m−2 and the Gclear by 9 W m−2.
  3. The mesoscale convective systems population forming the ITCZ spans a wide range of scale from 100 to 1,000,000 km2. The organized systems (area greater than 104 km2 and minimum IR temperature colder than 240 K), although less numerous than the smaller ones, have a dominant contribution on the upper level cloudiness.
  4. The largest systems contain the deepest convective cells reaching the tropopause temperature. Comparisons of the difference between the WV and the IR channel of Meteosat suggest the presence of water vapor above these turrets. These largest systems LWCRF can reach up to 100 W m−2 and their contribution to the integrated longwave cloud radiative forcing over the Indian Ocean is overwhelming (72%).
  5. The active phase of the intraseasonal perturbation is associated with an increase of the vertically developed systems (Class Ia). During this phase, the average Class Ia are reaching deeper heights and have a larger spread than during the inactive phase.

[55] The overall strong correlations between organized convective systems, UTH, longwave radiative forcing and clear-sky greenhouse effect allow to highlight the role of the organized convection in the feedback mechanisms mentioned in the Introduction. The observed deepening of the coldest convective cell within the convective systems is accompanied by an increase of the whole cloud system area. The radiative properties of the system indicates that the cloud system area increase yields to an increase in the longwave cloud radiative forcing. The UTH-convective systems relationship further suggests that this deeper, larger, more radiatively active cloud systems are associated with a moister environment, which translates into an increased greenhouse effect. So, a deepening of convection implies an increase in both the cloudy and clear-sky radiative forcing that can feedback on convection. These radiative signatures of the cloudiness and moist environment are rooted in the fact that convection is organized in the form of convective systems. The organization of convection at mesoscale appears crucial to the longwave radiative feedbacks, both cloudy and clear, over the Indian Ocean.

[56] A number of perspectives arise from this first analysis of the Indian Ocean convection and moist environment. Work will be pursued along two main lines. The first concerns the integration of the previously mentioned organized convective systems characteristics along their life cycle. This will allow to study the energetic cycle of convection and to complement the present effort toward a better understanding the impact of the organized systems on the upper tropospheric moisture and the diurnal cycle scale. The study will be extended to the shortwave component of the radiative budget and the interest shifted to relating the different mechanisms highlighted in this study to the surface condition (the SSTs) in order to provide an estimation of the associated climatic feedbacks.


[57] Integrating this amount of satellite observations would not have been possible without the efficient and much appreciated support from Jean-Louis Monge and the IPSL Climserv database server (http://www.climserv.polytechnique.fr). Enlightening monsoon discussions with Robert Sadourny provided the major motivation for this study. A number of individuals helped through exciting discussions to shape up the present study. Among them, specials thanks are due to Erwin Boer, Jean Philippe Duvel, Hanna Pawlowska, Geneviève Sèze, Carlsten Standfuss, and Eric Wilcox. The French INDOEX team is acknowledged for providing an exciting scientific work environment. RR would like to thank Pr. V. Ramanathan from C4 for making his participation to INDOEX a reality as well as providing exciting discussion about the convective cloud issues. In the field discussions with D. Sikka and the forecast team are very much appreciated. The EUMETSAT important support to the INDOEX community and in particular Dr. Soerensen for his dedicated efforts are gratefully acknowledged. The ScaRaB program is a cooperative program of France, Russia, and Germany. The GPCP data were acquired at NASA Goddard SFC, Maryland, USA. The PW estimates from SSM/I were from Remote Sensing Systems, Santa Barbara, CA, USA. This study was funded by the French Programme Atmosphère Océan à Moyenne échelle. RR benefited from a Centre National d'Études Spatiales postdoctoral Fellowship. We would like to also thank the two anonymous reviewers for their comments that helped to clarify the presentation of our results.