New evidence of cloud invigoration from TRMM measurements of rain center of gravity



[1] Much of the recent focus regarding aerosol effects on convective clouds and rain patterns has been on the possible invigoration of convective clouds by aerosols. Here we approach the issue using new methods, as rain vertical profiles' center of gravity (RCOG) and spread (RS) from TRMM are combined with aerosol optical depth (AOD) measurements from MODIS to examine this effect in various marine regions worldwide. Careful attention is also given to the meteorological, spatial and temporal variance in each region, as we try to extract and isolate the regional aerosol effect out of the dominant dynamic forcing. We show that for the majority of cases, high AOD values are correlated with higher RCOG and larger RS, indicating significant invigoration of the rain vertical distribution by aerosols.

1. Introduction

[2] Aerosol can impact clouds via two main pathways: radiative and microphysical [Kaufman and Koren, 2006; Koren et al., 2008]. The radiative path refers to aerosol interaction with radiation that can change the environmental temperature and humidity profiles and by that influence cloud processes. Absorbing aerosol warms the aerosol layer and cools the layers below. This can suppress shallow clouds (inside or below the aerosol layer) by stabilizing the atmospheric layer below the clouds and suppressing moisture and heat fluxes from the surface [Feingold et al., 2005; Koren et al., 2004]. In the base of the microphysical effect lies the fact that aerosols serve as Cloud Condensation Nuclei (CCN) [Twomey, 1977] and as Ice Nuclei (IN). By changing the size distribution and number of clouds drops/ice particles (the microphysical path) the aerosols will affect condensation/evaporation rates, latent heat release, collision coalescence efficiency and related clouds properties, such as: cloud lifetime [Albrecht, 1989; Jiang et al., 2006], cloud dimensions [Kaufman et al., 2005], cloud reflectivity [Twomey, 1977], particles phase, precipitation patterns and more. The magnitude and sign of these effects depend on the cloud type and on the environmental conditions [Khain, 2009; Rosenfeld et al., 2008; Sorooshian et al., 2009; Stevens and Feingold, 2009].

[3] In cases of convective clouds microphysics and dynamics are tightly coupled. A high CCN number yields nucleation of more and smaller cloud droplets that might consume the available water vapor more efficiently. Furthermore, smaller droplets exhibit smaller terminal velocities and therefore a given updraft will lift the smaller droplets higher than the larger droplets [Koren et al., 2010a]. Consequently, more droplets are able to reach above the freezing level and release freezing latent heat, enhance cloud ‘effective’ buoyancy and as a result increase cloud top height and lifetime [Khain et al., 2005]. Smaller droplets with smaller size variance collide less efficiently and therefore the conversion of cloud droplets into rain drops is delayed [Rosenfeld, 2000]. An amplification of such feedbacks occurs above the freezing level as freezing processes are less efficient for smaller droplets [Rosenfeld and Woodley, 2000]. The addition of more IN to the cloud involves many other complicated processes that are not well known yet and cannot be evaluated properly [Teller and Levin, 2006; van den Heever et al., 2006]. Several other physical mechanisms have been suggested as players in cloud invigoration as well [Lebo and Seinfeld, 2011; Lee et al., 2010; Rosenfeld et al., 2008; Tao et al., 2007].

[4] The expected consequences of invigoration are deeper and larger clouds [Koren et al., 2005]. For deep convective clouds we can expect also larger hail, more thunderstorms and lightning flashes [Altaratz et al., 2010], and widespread anvils [Koren et al., 2010b]. Moreover, it was recently suggested that the delay in warm rain and redistribution of water and ice in higher levels in invigorated convective clouds will eventually result in significantly stronger rain rates [Koren et al., 2012]. Observational evidence for deep convective invigoration has been seen in many regions globally, such as the biomass burning areas in the Amazon [Andreae et al., 2004], tropical Atlantic Ocean [Jenkins et al., 2008; Koren et al., 2010a; Koren et al., 2005], the relatively aerosol free tropical Pacific Ocean [Yuan et al., 2011], and the South East US [Bell et al., 2009; Rosenfeld and Bell, 2011]. It is important to stress that these findings do not necessarily imply that more aerosols results in larger total amount of precipitation in a given region. Some studies actually report a negative correlation between aerosol loading and cumulative precipitation [Givati and Rosenfeld, 2004; Rosenfeld et al., 2001; Teller and Levin, 2006]. Although evidence supporting the theory of invigoration has been piling up in the recent years, the conditions, magnitude, and impacts of invigoration remain open questions [Khain, 2009; Lebo and Seinfeld, 2011]. The non-linearity of the aerosol-cloud interactions, uncertainties in remote sensing measurements, and difficulty to decouple the aerosol net affect from the general meteorological forcing makes the study of aerosol effects on convective clouds among the most challenging tasks in the field [Koren et al., 2010a]. In this work, with the use of two independent remote sensing instruments, we combine AOD (Aerosol Optical Depth) measurements with rain vertical profiles in order to examine the possible effects aerosols may have on the vertical structure of rain within and below the cloud.

2. Methods

[5] Rain vertical profiles for the years 2002–2009 were collected from the Tropical Rainfall Measurement Mission (TRMM) level 2 product 2B31 [Haddad et al., 1997; Kummerow et al., 2000]. TRMM covers 35°N to 35°S in a non-sun-synchronous orbit, providing spatial and temporal statistics for the tropics and sub-tropical regions. 2B31 product combines both Precipitation Radar (PR) and Thermal Microwave Imager (TMI) data to obtain the precipitation vertical profile estimation. The product output is adapted to the PR vertical profile with a ∼5km footprint. The total swath is 250km and the vertical profile ranges from the surface up to 20km above the earth's ellipsoid, with a resolution of 250m (i.e., total of 80 vertical levels).

[6] Using the center of gravity (COG) analysis [Grabowski et al., 2006; Koren et al., 2009], each rain vertical profile was translated into rain center of gravity (RCOG) height and rain spread (RS) values. Rain center of gravity (equation (1)) and rain spread (equation (2)) are defined as:

display math
display math

Where i is a running index between all 80 vertical levels, Riis the measured rain-rate at that vertical level, andHiis the actual height of the vertical level above the earth's ellipsoid. The RCOG can be considered as the physical height above sea-level that represents the weighted rain rate distribution of the column and RS represents the variance of the rain rate column around that height. The main advantage of this tool is that we focus on the distribution within the vertical profile which is insensitive and cancels out consistent rain-rate biases throughout the vertical profile. This is opposed to absolute value parameters such as surface rain-rate or total column rain mass which are regionally dependent and may contain many biases (such as Z-R relationship biases due to different drop size distributions [Rosenfeld and Ulbrich, 2003]). Level 3 AOD at 550 nm data was collected from MODIS (MODerate resolution Imaging Spectroradiometer, Level 3 algorithms available at [Remer et al., 2005; Salomonson et al., 1989]. The data includes daily measurements of global AOD on a 1° grid from both Terra (overpass at 10:30 LST) and Aqua (overpass at 13:30 LST) satellites.

[7] Per day, all RCOG and RS measurements were sorted into a MODIS corresponding grid and given an AOD value (i.e., the average of Aqua and Terra measurements from that day), therefore our sensitivity is limited to variations of AOD on a daily or longer temporal scale. Pixels with AOD values above 0.5 (or 1 for highly polluted regions, see Sect. 3), or cloud fraction above 80% were discarded to minimize cloud contamination.

2.1. Regions of Interest

[8] We narrowed the analysis to oceanic regions due to the fact that RCOG is highly correlated with topographic height (which would be difficult to later decouple), and both the TRMM TMI and the MODIS AOD algorithms are considered superior over the sea because of the dark homogeneous background [Remer et al., 2005]. Both clean and polluted environments were analyzed. Moreover, the regions are no larger in length than the typical synoptic scale of around 1000 km. The Regions of Interest (ROIs) are shown in Figure 1a, all located in areas with sufficient rain events for large statistics. Most regions lay within the latitude belt of 25°–35° (N and S) because of the increased statistics TRMM has in those latitudes (twice as many overpasses compared to the equator), the proximity to continental pollution sources and ambient convective conditions.

Figure 1.

(a) Study regions of interest (ROIs: South East US (SEUS), South Atlantic (SA), North Atlantic (NA), Central Africa (CAF), Eastern Mediterranean (EM) and Eastern China (ECH)). Land pixels in the ROIs were discarded. (b) RCOG anomaly as a function of AOD for South East US, May until October. Data is restricted by the VVEL at 350 mb, and divided to three regimes: Updrafts (red lines), Neutral (green lines) and Downdrafts (blue lines). Solid lines respresent the data sorted by AOD into five bins of equal RCOG counts, and dashed lines represent the weighted (by standard deviation magnitude) linear least squares fit done for each five bin set. (d) Standard deviation magnitudes for the corresponding bins in Figure 1b. (c and e) Same as Figures 1b and 1d, but for the RS anomaly. Both RCOG and RS show positive slopes for all meteorological regimes. Meteorological regime boundary values, counts per bin (CPB), and linear fit equations appear in the figure legends.

2.2. Meteorology

[9] To reduce the chance of meteorological causality (when the meteorology is the driver for both aerosols' and clouds' properties) addressed in other works [Koren et al., 2010a], we used NOAA-NCEP Global Data Assimilation System (GDAS) reanalysis data (for details seeParrish and Derber [1992] and Saha et al. [2010]). This meteorological data was used for division of the rain and aerosol days to subsets according to the meteorological conditions in order to test the AOD effects per specific meteorology. For each region, surface rain-rates obtained from the TRMM 3B42 product (details given byHuffman et al. [2007]) were correlated with GDAS meteorological parameters at pressure levels from 1000 mb to 100 mb (a total of 157 GDAS parameters). The 3B42 estimates were interpolated from a 0.25° grid to a 1° grid that fits the GDAS data. Two parameters most correlated to rainfall were chosen to represent each region's meteorological forcing. Analysis of all the ROIs reveals these parameters to be vertical velocity (VVEL) and relative humidity (RH) at different pressure levels (according to the specific location and season), in agreement with other studies [Koren et al., 2010a; Takayabu et al., 2010].

2.3. Anomalies

[10] Spatial and temporal variations of RCOG and RS within the ROIs and for specific season can create artificial correlations. Therefore, the RCOG and RS anomalies with respect to the mean values are taken per region per month, as can be seen in equation (3) for RCOG (the same is done for RS):

display math

Where m is the month of the rain measurement, x and y indicate the 1° × 1° grid coordinates in which the rain measurement is located, and subscript i is the rain measurement index. ΔRCOGi(mxy) is the anomaly value, RCOGi(mxy) is the original value, and inline image is the mean monthly value for the grid box. To further minimize temporal variations we divided each year of data into two (one in the case of Eastern Mediterranean) main seasons. The division was based on the analysis of RCOG seasonal variations and supported by 850 mb and 500 mb wind patterns from GDAS reanalysis.

3. Results and Discussion

[11] A representative example of the performed analysis for ΔRCOG can be seen in Figure 1b. This example focuses on the South-East USA (SEUS) ROI during May-October, while using vertical velocity (VVEL) at 350 mb as the restricting meteorological parameter. Rain COG anomaly values were first divided into three meteorological regime sets: 1) Updrafts (negative) values: −80 mb/hr to −1 mb/hr. 2) Neutral values: −1 mb/hr to 1 mb/hr. 3) Downdrafts (positive) values: 1 mb/hr to 40 mb/hr. We tried to equalize the number of rain counts for each meteorological regime. The three sets were then sorted into five equal count bins as a function of the corresponding (same day) AOD values. The data points (connected by solid lines) inFigure 1b and lines in Figure 1d represent the mean value and standard deviation of each bin, respectively. Finally, a weighted linear least squares fit (dashed lines) was applied to the data points, where the inverse of the variance in each bin was taken as its weight. Fit details appear in the figure legend.

[12] Positive correlations between ΔRCOG and AOD can be seen for all meteorological regimes in SEUS during May-October (seeFigure 1b). For low (high) AOD, the RCOG tends to be below (above) the mean RCOG value. The shift between RCOG values for different meteorological regimes (i.e., RCOG in updraft conditions is higher than in neutral conditions, and so on) is consistent with the choice of the VVEL as a determining parameter for cloud development and the rain vertical profile extent. The standard deviation values in Figure 1d reflect the large natural variance we expect to see in all rain profiles. An increase of standard deviation values with increasing AOD is seen.

[13] We conclude the example by showing similar results for the RS anomalies (and their corresponding standard deviation per bin) as a function of AOD in Figures 1c and 1e. Larger ΔRS values correspond to higher AOD values. The standard deviation values of ΔRS are not dependent on AOD. Meaning that the variance of the spread of the precipitation vertical profile is consistent for the whole AOD range. The results suggest that the rain profiles are not only “concentrated” at higher levels, but also exhibit larger vertical spread around that height as the AOD increases. This can be translated to an increase in rain initiation height for polluted clouds and implies cloud invigoration.

[14] An identical process was performed for all ROIs. The slopes of the linear fits for both ΔRCOG and ΔRS were collected for each case. The results of inline image slopes vs. inline image slopes are summarized in Figure 2, and the full list of linear fit correlation values and meteorological parameters information is presented in Table S1 in the auxiliary material. It is important to note that we increased the AOD limit value to 1 for ECH and CAF regions due to the lack of sufficient rain data in “clean” (AOD < 0.5) conditions and the highly polluted environmental conditions there. Moreover, we acknowledge that the linear fits do not apply to all data sets and may be an oversimplification of the ΔRCOG and ΔRS dependence on AOD and therefore show slope values (either for ΔRCOG or ΔRS) which are statistically significant with a high level of confidence (P-Value < 0.05).

Figure 2.

Weighted linear least squares fit slope values concentrated on ΔRCOG slope vs. ΔRS slope plots. All of the presented slopes obtained from fits that are statistically significant with P-Value<0.05. Colors are attributed to a region per sub-season (e.g., Black–East China during April–September) in the lower legend. The two meteorological parameters most correlated with rain (per region per season) appear in the lower legend as well. Meteorological parameters and regimes are divided by shape as can be seen in the upper legend.

[15] All the slopes (except one CAF slope) in Figure 2 lie in the 1st quadrant (positive slopes for ΔRCOG and ΔRS derivatives), indicating invigoration effects in all but one ROI. A clear positive linear dependence between ΔRCOG and ΔRS derivatives as a function of AOD is seen. Every ROI (except CAF) shows significant invigoration for at least one meteorological regime during a rain season. The invigoration effect seems to be larger and more robust at times when the rain clouds are deeper (see Table S2 in the auxiliary material), e.g., ECH during monsoon season, SEUS throughout the year. In those clouds the aerosol effect on the microphysical and dynamical processes can alter the cloud's vertical distribution of water and released energy substantially. Based on the regions selection, the cloud fraction <80% filter, and further analysis of rain profile characteristics, we can assume that the typical rain profile in all ROIs is most probably of convective nature and in the majority of meteorological regimes reaches well above the freezing level.

[16] The Central Africa (CAF) region is clearly anomalous compared to the rest. Negative aerosol effects on cloud development and precipitation in the region have been reported in other studies as well [Huang et al., 2009; Koren et al., 2012]. We can speculate that this can be due to the region's aerosol composition. The CAF region is known to be abundant with biomass burning throughout the year [Haywood et al., 2008] and Saharan mineral dust aerosols mainly during the boreal winter and spring [Ben-Ami et al., 2012]. Both types of aerosols are highly absorbing [Johnson et al., 2008; Malavelle et al., 2011; Osborne et al., 2008] and can have a large effect on the thermodynamic properties of the environment beside their effect on microphysical processes, resulting in cloud suppression and atmospheric profile stabilization as previously seen in the Amazon basin [Davidi et al., 2009; Koren et al., 2008]. Moreover, due to lack of clean pixels in this region, we allowed for higher AOD values (AOD < 1). This can further support the stabilization idea, as in this range absorption depends on the AOD almost linearly while microphysical effects are expected to saturate in much smaller AOD values [see Koren et al., 2008].

[17] An alternative way to view the results is by plotting ΔRCOG and ΔRS as a function of the meteorological parameter, while dividing the data according to three aerosol loading classifications: High, Medium and Low. Examples of this analysis for ΔRCOG over the Eastern Mediterranean and Eastern China are presented in Figure 3. The variance of the data is large, but it is clear that meteorology has the dominant effect on ΔRCOG values. The transition from positive to negative VVEL values corresponds to a transition from low to high ΔRCOG values. The same can be seen in transition from low to high RH values.

Figure 3.

Examples of RCOG anomalies and their corresponding standard deviation per bin as a function of meteorological parameters. (a–d) Eastern Mediterranean (EM) during November-March. (e–h) Eastern China (ECH) during April–September. (left) Vertical velocity (VVEL) at selected pressure levels, 650 mb for EM and 450 mb for ECH. (right) Relative humidity (RH) at selected pressure levels, 650 mb for EM and 500 mb for ECH. Low, medium and high AOD datasets correspond to the blue, green and red lines, respectively. The AOD classification boundary values and counts per bin (CPB) appear in the figure legends. Notice that the Y-axis are scaled differently for each region, according to their different rain profile characteristics and variances.

[18] However, the division to different AOD classifications yields clear translations (shifts) of the baselines of the meteorological trends. For any given meteorological state (any line parallel to the Y axis), high AOD points tend to be consistently higher than low AOD ones. Or in other words, for the majority of given meteorological settings, increased AOD results in an increase in ΔRCOG and invigoration. Similar results were obtained in other regions, with CAF being the sole exception. As seen in Figure 1d, standard deviation values per ΔRCOG bin generally increase with increasing AOD.

4. Summary

[19] In this work we examined the effects aerosols have on rain profiles for various oceanic regions. Using TRMM satellite, we analyzed rain center of gravity (RCOG) height and rain spread (RS) about that height. Results show that for many regions and meteorological conditions high aerosol loading tends to increase both the RCOG and RS of the rain columns independently of the local meteorological forcing. Although we see an indication of invigoration in all regions, the results are more prominent in cases favoring deeper convective clouds, such as the South East US (SEUS) and Eastern China (ECH) during summer months. An opposite trend is seen in the Central Africa (CAF) region, where aerosols seem to have a negative effect on the rain profile extent. This could be due to the unique aerosol composition or methodological limitations in that region.

[20] Future work which characterizes the environmental conditions in higher resolution, aerosols according to their size, composition, physical and chemical properties and clouds properties can help us gain more insight regarding the conditions most favorable for aerosol invigoration of convective clouds.


[21] This work was supported in part by the Israel Science Foundation (grant 1172/10) and NASA's Radiation Sciences Program and Interdisciplinary Studies.

[22] The Editor thanks two anonymous reviewers for assisting in the evaluation of this paper.