The lifecycle of anvil clouds and the top-of-atmosphere radiation balance over the tropical west Pacific

Abstract

Observations from a geostationary satellite are used to describe the lifecycle of mesoscale convective systems (MCS), their associated anvil clouds, and their effects on the radiation balance over the warm pool of the tropical west Pacific Ocean. In their developing stages, MCS primarily consist of clouds that are optically thick and have a negative net cloud radiative effect (CRE). As MCS age, ice crystals in the anvil become larger, the cloud top lowers somewhat, and clouds with neutral and positive net CRE become more common. Shading from anvils causes cool anomalies in the underlying sea surface temperature (SST) of up to −0.6 °C. MCS often occur in clusters that are embedded within large westward-propagating disturbances, so shading from anvils can cool SSTs over regions spanning hundreds of kilometers. Triggering of convection is more likely to follow a warm SST anomaly than a cold SST anomaly on timescales of several days. This information is used to test hypotheses on why, over the warm pool, the average shortwave and longwave CRE are individually large but nearly cancel. The results are consistent with the hypothesis that the cancelation in CRE is caused by feedbacks between cloud albedo, large-scale circulation, and SST.

1. Introduction

Mesoscale convective systems (MCS) are fundamental to the radiation budget of the tropics. They contain extended canopies of high, cold, and bright anvil clouds that emanate from convective towers. Anvils efficiently trap outgoing thermal radiation and reflect sunlight, but over the warmest tropical oceans these heating and cooling effects nearly cancel one another. This cancelation can be seen in Figure 1, which shows the observed climatology of top-of-atmosphere cloud radiative effects (CRE). Over the western tropical Pacific and the Indian Ocean – regions of warm and uniform sea surface temperature (SST) and enhanced convection – the average shortwave (SW) and longwave (LW) CRE are about ±80 Wm⁻², while the net CRE is an order of magnitude smaller.

Figure 1.

Climatology of cloud radiative effects during June, July, and August of 2000–2014. In the bottom panel, the black box shows the study region and the gray dot shows the subsatellite point of Himawari-8. Data were derived from the Clouds and the Earth’s Radiant Energy System (CERES) satellite measurements and obtained from NASA.

The close balance between SW and LW CRE in the convective tropics is a surprising feature that was first detected in early satellite measurements of Earth’s radiation budget [Ramanathan et al., 1989]. A convincing explanation for this balance has remained elusive, but several hypotheses have been proposed:

1. The Coincidence Hypothesis

Kiehl [1994] argued that the balance in CRE is a coincidence resulting from two unrelated features of the tropical atmosphere. He argued that thick anvils determine the average CRE. Since very thick clouds have an albedo and emissivity that are essentially fixed, the LW CRE is set by the temperature of the upper troposphere, the SW CRE is set by the average insolation, and the cancelation between SW and LW CRE is just a coincidence. Cess et al. [2001] and Hartmann et al. [2001] showed that deep convection produces a variety of cloud types – many of which are far from radiatively neutral – so there must be another explanation. However, a weak version of the coincidence hypothesis could be true: even though deep convection produces a diverse ensemble of clouds, it could be a fortuitous coincidence that the ensemble-mean net CRE is small.

2. The Cloud-Circulation-SST Feedback Hypothesis

Ramanathan and Collins [1991] argued that over the warm pool, cloud albedo, atmospheric circulation, and SST are tightly coupled through two-way interactions. For instance, consider how a particular region would respond to a sustained period without convection. Within the region, the sea surface absorbs more solar energy than its surroundings, so it warms until it becomes unstable. At this point convection fires, generating a bright anvil that shades and cools the sea surface. Once the anvil dissipates, the sea surface warms from the renewed solar heating, and the process repeats. Thus, cloud albedo, atmospheric circulation, and SST regulate one another through negative feedbacks.

Hartmann et al. [2001] hypothesized that a cloud-circulation-SST feedback process could constrain the top-of-atmosphere net CRE to be uniform in the warm tropics. They devised an energy-balance model to illustrate the feedback process as simply as possible. The model includes a warm region with active convection, a cooler region of suppressed convection, and an overturning circulation that connects the two regions and transports energy between them. The circulation is assumed to have a horizontal length scale larger than individual convective systems, but small enough that the circulation is confined to the deep tropics. The small Coriolis parameter and near-neutral convective stability in the tropics cause weak horizontal temperature gradients and weak vertical gradients in moist static energy. As a result, SST gradients generate a strong circulation response that is relatively inefficient in transporting energy horizontally. Weak circulations nonetheless give large changes in the radiative properties of clouds since convection responds strongly to small mean vertical velocities. Within the convecting region it is assumed that the outgoing LW radiation is fixed by fixed emission temperature for the clouds, but that the albedo of the clouds would respond more sensitively to vertical motion. This causes the average top-of-atmosphere net radiation, and in particular the net CRE, to be similar in the regions of active and suppressed convection. Hartmann et al. [2001] estimated that the average net CRE is around −10 Wm⁻² in regions of suppressed convection within the warm pool, so the net CRE must be small in regions of active convection as well.

The Hartmann et al. [2001] model has not been verified with observations or with more realistic models. Hartmann and Berry [2017] recently pointed out that the cancelation in CRE occurs rapidly in nature, while the temperature of the ocean mixed layer changes more slowly. Therefore, they argue that another feedback mechanism that operates more quickly is needed, and the cloud-circulation-SST feedback hypothesis may not be a complete explanation. Others have also challenged the validity of the model [Chou and Lindzen, 2002, but see Hartmann et al., 2002].

3. The Radiative Heating Hypothesis

Hartmann and Berry [2017] showed that over the warm pool, thin cirrus are observed more frequently than thick anvils. Therefore, the balance in CRE results from a cancelation between thick anvils, which are less common but have a strong cooling effect, and thinner extended ice clouds, which are much more common but have a weaker warming effect. Based on this observation, Hartmann and Berry [2017] argued that the balance in CRE likely results from a process that increases the longevity of thin and medium anvils relative to thick ones. They then investigated if this process involves radiative heating of clouds. Radiative heating lengthens the lifetime of anvils by heating and lifting the cloud layer, and by preferentially heating the cloud base and encouraging turbulent mixing within the cloud. These effects are weaker in thick anvils because they have a lower and warmer cloud base, and emit a larger downward flux of LW radiation. Since radiative heating extends the lifetime of anvils, but the effect is stronger for thin and medium clouds than for thick ones, Hartmann and Berry [2017] hypothesized that radiative heating could cause the net neutrality of the cloud population. The importance of radiative heating for the longevity of thin and medium anvils has also been demonstrated in the modeling studies of Fu et al. [1995] and Harrop and Hartmann [2016].

It is important to determine which of these hypotheses is correct – if any – since the answer may inform how tropical clouds will change in the future. As the climate warms, anvils are expected to rise and maintain a nearly constant cloud-top temperature, causing a positive LW cloud feedback [Hartmann and Larson, 2002]. Additionally, anvil coverage may decrease [Bony et al., 2016], but it is unclear how the average cloud albedo will change. If the SW and LW CRE are currently balanced due to a coincidence, then they may not remain balanced in the future. On the other hand, if the balance in CRE results from a robust feedback within the climate system, then it opens the possibility that CRE will remain balanced. It is important to determine which scenario is more likely, since breaking the CRE balance would likely have a profound impact on Earth’s climate [Pierrehumbert, 1995].

Our goal is to study MCS clouds over the warm pool and to test the hypotheses on the cause of the balance in CRE. Because the CRE cancelation occurs rapidly in nature [Hartmann et al., 2001], we will focus on fast processes and describe the cloud evolution over the MCS lifecycle. To accomplish this we use observations from a geostationary satellite that include retrievals of cloud macrophysical, microphysical, and radiative properties. The observational data are described in Section 2. We use an algorithm to objectively track MCS, and this algorithm is described in Section 3. The evolution of the cloud properties and the large-scale environment over the MCS lifecycle are described in Section 4, and in Section 5 this information is used to evaluate the hypotheses for why the CRE is balanced over the warm pool. Conclusions and a summary are presented in Section 6.

2. Observational Data

We study the tropical west Pacific region – 130°−180°E, 20°S-20°N – during July 5 through August 31 of 2015, and June 1 through August 31 of 2016. The study domain is shown in Figure 1. Two satellite products are used: cloud properties and radiation retrieved by Himawari-8, and SST retrieved by the Advanced Microwave Scanning Radiometer 2 (AMSR-2).

2a. Himawari-8

Himawari-8, hereafter “Himawari,” is a geostationary satellite that orbits above 0°N and 140.7°E (Figure 1) [Bessho et al., 2016]. It houses a high-resolution multispectral imager with five channels that are useful for cloud monitoring: visible (0.64 μm), shortwave-infrared (3.9 μm), infrared (10.4 μm), split-window (12.4 μm), and CO₂-slicing (13.3 μm). These channels are used to retrieve cloud phase, optical depth, particle size, condensed water path, and cloud-top pressure via the NASA Langley Satellite Cloud and Radiation Property retrieval System (SatCORPS) algorithms. The cloud retrieval algorithms were originally designed for the Clouds and the Earth’s Radiant Energy System (CERES) experiment and applied to data from the Moderate-Resolution Imaging Spectroradiometer (MODIS), but were adapted for use with current geostationary satellites [Minnis et al., 2008a, 2011]. The dataset also includes preliminary estimates of top-of-atmosphere broadband albedo and LW flux; however, to ensure consistency with CERES, these parameters are re-derived for this study. Broadband SW albedo is derived from the observed visible radiances by applying narrowband-to-broadband conversion functions [Minnis et al., 2016] and then normalizing to remove any residual bias in albedo relative to CERES. The Edition 4 CERES Synoptic product (SYN1deg; [CERES, 2017b]) is used in the normalization step. Broadband LW flux is derived by applying a modified version of the radiance-based approach of Doelling et al. [2016] and then normalizing using the Edition 4 CERES Aqua Single Scanner Footprint product (SSF1deg-HOUR; [CERES, 2017a]). Finally, measurements of infrared brightness temperature are used to track MCS. Pixel-level data are analyzed, which have been sampled to ~8 km horizontal resolution and one hour temporal resolution [Minnis et al., 2008a; NASA, 2017].

Himawari provides a unique opportunity to study MCS for two reasons. The first is its sampling characteristics. Because Himawari follows a geostationary orbit, it views the entire lifecycle of individual MCS. The high space- and time-resolution of Himawari also makes it possible to resolve the distribution of cloud properties within MCS. Secondly, the SatCORPS cloud retrieval for Himawari is state-of-the-art. For example, the cloud retrieval of the International Satellite Cloud Climatology Project (ISCCP; [Rossow and Schiffer, 1991]) – the most commonly used cloud record based on observations from geostationary satellites – uses only the visible and infrared channels. The SatCORPS algorithms use the additional spectral information measured by current geostationary satellites to retrieve more cloud variables, and with higher accuracies.

Although Himawari has these advantages, it is still subject to certain limitations and uncertainties. Like all satellites, instrument degradation is a possible issue. To correct for this, Himawari radiances are matched with MODIS data and calibrated monthly [Minnis et al., 2008a]. The estimates of cloud properties are also sensitive to errors in the retrieval algorithms. Theoretical uncertainties in cloud optical depth, particle size, and condensed water path are 10%, 15%, and 17%, respectively [Dong et al., 2008]. These variables are subject to large biases when the solar zenith angle is large [Grosvenor and Wood, 2014] and are not reliably retrieved for optically thick clouds at night. To mitigate this problem, we restrict our study of these variables to scenes where the solar zenith angle is less than 70°. Furthermore, the retrieved particle size represents conditions in the top portion of the cloud. Anvils often contain small ice crystals near their tops, so the retrieved ice crystal size is typically smaller than the average over the entire depth of the anvil [Heymsfield and McFarquhar, 1996; McFarquhar and Heymsfield, 1996; Jensen and Del Genio, 2003]. Cloud-top height is often underestimated for high-clouds, particularly when other clouds are present below. This bias is around 1 km for clouds with optical depths greater than 3, and can be as much as 7 km for some thin cirrus [Minnis et al., 2008b; Smith et al., 2008; Holz et al., 2009].

The updated Himawari-derived broadband radiative flux retrievals are compared to coincident measurements from CERES to assess uncertainty. Instantaneous fluxes from the Edition 4 CERES Aqua and Terra SSF1deg-HOUR product are matched with Himawari-derived fluxes, similar to the validation procedure of Minnis et al. [2016]. The overlapping measurements are used to compute error characteristics of the Himawari data, and these are shown in Table 1. Upper bounds for the magnitude of the errors relative to CERES are a 0.1% bias and 2.5% root-mean-square deviation for LW flux, and a 0.6% bias and 1.1% root-mean-square deviation for SW flux. These correspond to errors of around 6 Wm⁻² or less.

Table 1.

Error characteristics of the Himawari broadband radiative flux retrievals. Himawari data are compared to coincident measurements from CERES onboard the Aqua and Terra satellites. The bias and root-mean-square deviation (RMSD) are shown. Note that error characteristics are computed from instantaneous measurements. Aqua crosses the equator around 0130 and 1330 local time, and Terra crosses around 1030 and 2230.

	Bias (Wm−2)	Bias [%]	RMSD (Wm−2)	RMSD (%)	Number of Observations
LW Flux Compared to CERES-Aqua	0.0	0.0	6.3	2.5	262,145
LW Flux Compared to CERES-Terra	−0.2	−0.1	6.1	2.5	265,672
SW Flux Compared to CERES-Aqua	−1.2	−0.6	2.0	0.9	130,913
SW Flux Compared to CERES-Terra	0.9	0.4	2.4	1.1	133,611

2b. Advanced Microwave Scanning Radiometer 2 (AMSR-2)

AMSR-2 is a microwave radiometer that is housed on the polar-orbiting GCOM-W1 satellite. We use the AMSR-2 L2 Version 3 Standard Product [Japan Aerospace Exploration Agency 2013; 2017b], which includes instantaneous, footprint-level measurements of SST. These measurements have a spatial resolution of 62 km × 35 km and represent the temperature in the top 1 mm of the ocean. Compared to buoy data, AMSR-2 measurements have a bias on the order of 0.01 K and root-mean-square error of 0.6 K [Japan Aerospace Exploration Agency, 2017a].

When studying the SST data, we remove the seasonal and diurnal cycles and analyze the anomalies. First, data are gridded into 1°×1° bins. Next, for each individual month of data, the monthly-mean is removed. Data are then sorted by hour of the day, and the mean of each hour is subtracted from the data.

AMSR-2 and Himawari data complement one another because of their different strengths and limitations. An important limitation of AMSR-2 is that SST is not retrieved when the rain rate is 0.5 mm/hr or more, which includes moderate and heavy precipitation. Also, since AMSR-2 is housed on a polar-orbiting satellite, it does not sample the evolution of individual MCS. However, the key strength of AMSR-2 is its ability to retrieve SST beneath clouds. This allows us to study SST near regions of deep convection.

3. Methods

3a. Estimating Cloud Radiative Effects

The cloud radiative effect (CRE) is a measure of how clouds alter the transfer of radiation through the atmosphere. It is defined as the difference between the top-of-atmosphere radiative fluxes during all-sky and clear-sky conditions. We estimate CRE by comparing the retrieved radiative fluxes to average clear-sky fluxes [Ramanathan et al., 1989]. The average clear-sky outgoing LW radiation (${\overline{\text{OLR}}}_{\text{clear}}$) is computed from all pixels in the domain that are over ocean and cloud-free, and the average clear-sky albedo (${\overline{\alpha }}_{\text{clear}}$) is computed similarly but is weighted by insolation. The LW CRE is computed as

$$\text{LW\hspace{0.17em}}{\text{CRE}}_{\text{i}}={\overline{\text{OLR}}}_{\text{clear}}-{\text{OLR}}_{\text{i}}$$

where the subscript “i” refers to an individual pixel. For SW CRE, we estimate the radiative effect under daily-mean insolation:

$$\text{SW\hspace{0.17em}}{\text{CRE}}_i=\left({\overline{\alpha }}_{\text{clear}}-{\alpha }_i\right){\overline{S}}_l$$

where ${\overline{S}}_l$ is the daily-mean insolation at the location of the measurement. Using daily-mean insolation ensures that variations in SW CRE are due to cloud properties only, not solar zenith angle. To avoid problems at large solar zenith angles, SW CRE is computed only for pixels with solar zenith angles less than 70°.

A potential source of error in this approach is that, compared to using instantaneous values of SW CRE, scenes with higher solar zenith angle are weighted more heavily, and vice versa. As a check, we compared the climatology of SW CRE computed using daily-mean and instantaneous insolation. The daily-mean estimate is 1.8 Wm⁻² more negative than the instantaneous estimate. This error source is comparable to the uncertainty from the retrieval (Section 2a).

3b. MCS Tracking

We use an image segmentation algorithm called Tracking Of Organized Convection Algorithm through a 3-D segmentation to objectively track MCS. The algorithm returns a “cold-cloud mask” that labels the pixels occupied by MCS clouds, with each MCS receiving a unique label. Fiolleau and Roca [2013] describe the algorithm in detail, so we will present an outline here.

The input of the algorithm is a timeseries of infrared brightness temperatures. The atmosphere is nearly transparent at these wavelengths, so photons that reach the top-of-atmosphere originate from either clouds or Earth’s surface. The coldest brightness temperatures are found above towering clouds near convective cores, and the warmest temperatures are found above clear skies. Above thin-to-moderately-thick anvils, the satellite views a combination of warm radiation emitted from Earth’s surface and cold radiation from the anvil, so intermediate brightness temperatures are observed. The signature of MCS typically includes a cold-core with incrementally warmer temperatures moving radially away (e.g., see Figure 1 of Houze [2004]). Because MCS have this distinct signature in the infrared, and because infrared measurements can be made during all hours of the day, infrared brightness temperature is useful for tracking MCS.

The MCS tracking algorithm is based on Boer and Ramanathan’s [1997] “detect and spread” algorithm. It has two components:

1. A “detect” step in which cold-cores are identified. Cold-cores are defined as groups of contiguous pixels that have very cold infrared brightness temperature.

2. A “spread” step in which the cold-cloud mask is incrementally expanded around each cold-core to include neighboring pixels that are slightly warmer.

The brightness temperature measurements occupy a three-dimensional space-time grid, and the “detect” step locates groups of contiguous pixels in this grid that have an infrared brightness temperature colder than 200 K. This includes the coldest 0.4% of high-clouds. When locating contiguous regions of cold cloud, two pixels are considered neighbors if they share a face or an edge (i.e. if one pixel belongs to an 18-connected neighborhood of the other). To be considered a cold-core, contiguous regions of cold cloud must contain at least 35 pixels and span at least two hours. This criterion is used to remove small, isolated cumulus so that only organized MCS are analyzed. Of all pixels colder than 200 K, 91% meet this criterion. Each cold-core is assigned a unique label in the cold-cloud mask.

Following the detection step, the cold-cloud mask is incrementally expanded from the cold-cores to neighboring pixels that are slightly warmer. Seed pixels that belong to the MCS, but are on the border, are compared to neighboring pixels outside the MCS. Neighbor pixels are added to the cold-cloud mask if they are sufficiently cold but are at least 1 K warmer than the seed pixel. This constraint stops the expansion of the cold-cloud mask when a local maximum in brightness temperature is reached – a condition that is found when multiple convective towers feed the same cold-cloud shield. In effect, this constraint assigns cold, cloudy pixels to the nearest cold-core. Note that the expansion happens in both space and time, so the tracked MCS generally include anvil cloud that is observed after the cold-core has dissipated.

The “spread” step is repeated in increments of 5 K up to a threshold of 235 K. The 235 K threshold is arbitrary, but it includes both precipitating and non-precipitating portions of the anvil [Yuter and Houze, 1998; Liu et al., 2007]. If a warmer threshold is used, then independent midlevel cloud can be spuriously assigned to the cold-cloud mask. If a colder threshold is used, then less of the anvil will be tracked. The threshold of 235 K is thought to be the best compromise between including as much of the anvil as possible, but minimizing contamination from midlevel clouds [Bouniol et al., 2016].

Missing data are rare. A total of 28 frames are missing from the five-month record, and no two missing frames are adjacent to one another. To allow tracking to continue when a missing frame is reached, the missing data are filled with infrared brightness temperatures from the previous time step. Filling of missing data applies only to brightness temperature – not to other cloud variables.

Furthermore, we restrict our study to MCS with space- and time-coverage that is at least 99% over ocean. MCS that touch the edge of the domain are not considered in the analysis to ensure that every MCS is observed throughout its entire lifecycle. The tracking algorithm identifies 5429 such MCS in the study domain.

An example of the tracking of two MCS is shown in Figure 2. The MCS are identified at 0300 and 0600 UTC – at the onset of deep convection. The clouds are initially distinct, but merge into a single cold-cloud shield at 0900 UTC. At this point the cold-cloud shield is shared between the masks of the two MCS, and is partitioned based on proximity to the cold-cores. As they are tracked forward in time, the two MCS remain distinct from one another, and from the convective activity to the west that fires at 0900 UTC. Tracking continues as the clouds dissipate, and ends at 1400 UTC.

Figure 2.

Example of MCS tracking. The upper panels show a timeseries of infrared brightness temperature. The lower panels are similar to the upper panels but also show the pixels associated with two tracked MCS – one in red and the other in blue. The arrows indicate the onset of deep convection. Frames are centered on 5°S, 154°E. Data are from June 24, 2016, and times are in UTC. The tracking algorithm identifies other MCS in this domain, but only two are indicated here.

This example highlights two strengths of the MCS tracking algorithm. First, the anvil boundaries are tracked accurately throughout the MCS lifecycle – from the onset of deep convection, through the merging of several clouds, to the dissipation of the cloud. Second, when several MCS merge, the algorithm partitions the cold-cloud shield based on proximity to the cold-cores. This minimizes artifacts from splitting and merging of MCS, and provides an optimal view of the MCS lifecycle.

Although the tracking algorithm has these advantages, it has one important weakness: it is unable to track optically thin cloud. Because tracking stops when the infrared brightness temperature reaches 235 K, clouds are not tracked once their optical depth drops to around 3–4. This limitation is important because thin cirrus account for a significant portion of the anvil. For instance, Protopapadaki et al. [2017] found that when a convective core is active, thin cirrus make up 10–30% of the MCS area. The fraction of thin cirrus becomes even larger after the convective core dissipates. Additionally, thin cirrus play a fundamental role in the tropical radiation budget [e.g. Berry and Mace, 2014; Hartmann and Berry, 2017]. In terms of their net radiative effect, thick anvils have a cooling effect, thin cirrus have a warming effect, and the crossover occurs at an optical depth of around 5 [Ackerman et al., 1988]. Thus, our analysis is biased towards medium and thick anvils, and underrepresents the warming effects of thin cirrus.

3c. Defining MCS Lifecycle Stages

The area of most tracked MCS follows a simple lifecycle. Cloud area grows to a maximum and then decays to zero, and both growth and decay are approximately linear in time [Roca et al., 2017]. Therefore, cloud area is a useful metric for defining stages of the cloud lifecycle. Cloud area and growth rate are used to discretize the MCS lifecycle into five stages, including two growing stages, a mature stage, and two dissipating stages. We refer to these as stages I-V, respectively. Lifecycle stages are distinguished by the times when the MCS reaches 40% and 80% of its maximum area. This is shown visually in Figure 3. Defining the MCS lifecycle by area and growth rate, rather than by time since initiation, allows MCS of varying lifetimes to be compared in a meaningful way.

Figure 3.

Demonstration of the MCS lifecycle stages. The area (A) of the blue cloud in Figure 2 is plotted as a function of time. The lifecycle stages are determined by the times when the cloud reaches 40% and 80% of its maximum area (A_max).

In some cases, MCS have a more complex lifecycle than the example shown in Figure 3, and have multiple pronounced local maxima in cloud area. These cases account for 3% of the number of tracked MCS and 5% of the total space- and time-coverage of MCS. In these cases, the secondary growth of the MCS is assumed to result from regeneration of the system, and the MCS is returned to an earlier lifecycle stage when regeneration begins.

3d. Tracking the Large-Scale Environment Surrounding MCS

The MCS tracking method described above is useful for illuminating processes that happen within individual MCS, but it does not show the evolution of the large-scale environment surrounding the storm. We study this with compositing analysis. For each cold-core identified by the MCS tracking algorithm, the time when the cold-core reaches its maximum area is identified. The centroid of the cold-core is computed at this time, and we refer to this location as the “storm center.” Composites are drawn for the region that spans 400 km in either direction of the storm center and for times within ten hours of the cold-core peak. A single MCS anvil is typically around 100–200 km across [Pope et al., 2008; Igel et al., 2014], but over the west Pacific, clusters of multiple connected MCS are common [Nakazawa, 1988; Mapes, 1993; Yuan and Houze, 2010; Protopapadaki et al., 2017]. The 800 km × 800 km region of the composite is large compared to a single MCS, but comparable to the size of large clusters of MCS.

Confidence intervals for the SST anomalies in these composites are computed as follows. A random error of 0.6 K in the SST measurements is assumed (Section 2b). The sample size is determined by counting the number of MCS that are used to generate the composite, and then reducing this value to account for the fact that multiple neighboring MCS are often observed within the domain.

3e. Estimating the influence of SST on the triggering of deep convection

We also wish to determine whether or not SST influences the triggering of convection. To this end, we identify scenes in which SST measurements are available and convection is not active, and then compute the likelihood that convection will fire over the next several days. As a measure of deep convection, we define the “cold-cloud fraction” as the fraction of pixels that have an infrared brightness temperature of 235 K or colder. A threshold of 235 K is used for consistency with the MCS tracking algorithm. The cold-cloud fraction is computed over all SST footprints, and then the footprints with zero cold-cloud fraction are selected. These data are assigned to a warm composite if SST > σ and a cold composite if SST < −σ, where σ = 0.54 K is the standard deviation of the SST anomalies. Finally, the evolution of the cold-cloud fraction over the following five days is examined. This analysis tests whether or not SSTs influence the triggering of deep convection on timescales of five days or less.

When computing confidence intervals we account for serial correlation in the data in the following manner. One degree of freedom is assigned for each AMSR-2 overpass. In other words, at a given instant in time, if a measurement of cold-cloud fraction from one location is used in the composite, then additional measurements from other locations are treated as if they add no additional information. This removes the effects of serial correlation in the spatial dimensions, so that only temporal serial correlation needs to be accounted for. Because AMSR-2 repeats coverage every 1–2 days, and because individual cold clouds usually have lifetimes shorter than one day [Roca et al., 2017], measurements of cold-cloud fraction associated with different AMSR-2 overpasses are assumed to be independent.

When studying the effects of SST on deep convection, we select a subset of the study region where deep convection is especially common: 12°S – 12° N, 150°E – 170°E [Hartmann and Berry, 2017]. About 97% of the SST measurements in this domain are warmer than 27.5°C – the threshold above which deep convection is possible [Graham and Barnett, 1987]. By focusing on this narrower domain, the calculation is not affected by the regions at the edge of the original study domain where SSTs are cooler and convection is less common.

4. Results

4a. Cloud Evolution over the MCS Lifecycle

We begin by discussing the MCS lifetime. Statistics of the MCS lifecycle are presented in Table 2. On average, MCS have a lifespan of 12.1 hours and spend between 1.6 and 3.0 hours in each lifecycle stage. The initiation and mature stages (I and III) are the longest, and the second growing stage (II) is the shortest. The mature stage includes about half of the space- and time-coverage of MCS. The first and last stages of the MCS lifecycle each account for 7% of the total coverage, but are useful to examine because they show the cloud evolution. Note that these values do not include cirrus with optical depths less than 3 or 4, which can linger for several hours to several days after convection has ceased [Luo and Rossow, 2004; Mace et al., 2006].

Table 2.

Time and area coverage during the MCS lifecycle stages. Fractional Coverage is the cloud area in the particular lifecycle stage divided by the cloud area summed over all stages. For each entry, the first number is the mean and the number in parenthesis is the standard deviation.

Lifecycle Stage	I	II	III	IV	V	All
Duration (hours)	3.0 (1.6)	1.6 (1.1)	3.0 (1.5)	1.8 (1.2)	2.7 (1.5)	12.1 (4.8)
Fractional coverage (%)	7 (4)	17 (9)	49 (14)	19 (10)	7 (4)	100 (0)

The evolution of cloud-top pressure and cloud optical depth (τ) throughout the MCS lifecycle is shown in Figure 4. Cloud tops range from 210 hPa to 100 hPa, and clouds of medium (3.6 < τ ≤ 23) and thick (τ > 23) optical depths are seen. During stage I, when convection is developing, the MCS includes a mix of medium and thick clouds. By stage II the highest and thickest clouds dominate the population, and these clouds remain the most prominent during stage III. Over the two dissipating stages, the cloud population shifts towards clouds with lower tops and with medium optical depths, and by the last stage, most clouds have tops between 150–210 hPa and more than half of the clouds have medium optical depth.

Figure 4.

The evolution of cloud properties over the MCS lifecycle. Joint histograms are shown for (top row) cloud visible optical depth and cloud-top pressure, (middle row) albedo and outgoing LW radiation, and (bottom row) SW CRE and LW CRE. The black lines in the bottom row show contours of net CRE from −100 Wm⁻² to +100 Wm⁻² in increments of 50 Wm⁻², and the heavy black line shows a net CRE of 0 Wm⁻². The first five columns show the MCS lifecycle stages, and the final column shows the histograms computed from the entire lifecycle. Bin widths are 2.5% for albedo, 2.5 Wm⁻² for OLR, 10 Wm⁻² for SW CRE, and 2.5 Wm⁻² for LW CRE.

The limitations of the cloud-top pressure retrieval should be considered here. Estimates of cloud-top pressure are biased towards lower altitudes, and the bias increases as clouds become optically thinner [Minnis et al., 2008b; Smith et al., 2008; Holz et al., 2009]. This could explain part of the apparent sinking of the anvil top with age. However, sinking of the anvil top with age is also seen in active retrievals, which provide a more accurate measurement of cloud-top height. Using satellite-based radar and lidar measurements, Bouniol et al. [2016] found that cloud tops lower as anvils age and dissipate, and Yuan et al. [2013] found that cloud-top height decreases as a function of distance from the cold-core. Even though the cloud-top pressure retrieval of Himawari has inherent bias, the sinking of the anvil top with age is probably a real feature. The sinking rate is likely determined by the competing effects of radiative heating and sedimentation [Ackerman et al., 1988].

The middle and bottom rows of Figure 4 show the evolution of cloud radiative properties over the MCS lifecycle. The MCS is composed of clouds with a wide variety of radiative effects. Albedo ranges from 38% to 73% and outgoing LW radiation ranges from 89 Wm⁻² to 147 Wm⁻². This corresponds to LW CRE ranging from +139 to +197 Wm⁻², SW CRE from −112 to −276 Wm⁻², and net CRE from −104 Wm⁻² to +42 Wm⁻². These values are the 5^th- and 95^th-percentiles of the data computed from the entire MCS lifecycle. Note that these values do not include the effects of thin cirrus, which have an important greenhouse effect. The lifecycle of cloud radiative properties is consistent with the cloud-top pressure and cloud optical depth evolution described above. In stage I, the cloud population has a modal albedo of around 68% and OLR of around 95 Wm⁻², but dimmer clouds with higher outgoing LW radiation are also seen. By stages II and III, the cloud distribution is strongly peaked at the coldest, brightest clouds that make up the modal values in stage I. These clouds have a net CRE of about −55 Wm⁻² – a substantial cooling effect. As the cloud ages, the population shifts to lower albedo, higher OLR, and a net CRE that is closer to neutral.

The evolution of cloud ice properties is shown in Figure 5. The top row shows ice water path, which is defined as the total mass of cloud ice above a unit area of Earth’s surface. The distribution of ice water path has a sharp peak with a modal value around 0.150 kg m⁻², and a long tail with values up to 4 kg m⁻². The largest values of ice water path distribution become increasingly more common during the growing stages, and then less common during the decay stages. The observations of ice crystal diameter, which represent conditions near the top of the anvil, show that over the MCS lifecycle the average ice crystal size increases by about 9 μm, which is 13% of the average crystal diameter during stage I. Heymsfield et al. [2005] suggested that very small ice crystals form by homogeneous nucleation in convective cores and are lofted to the top of the anvil. These small crystals are the first to be depleted, either by sublimation or by aggregating with larger crystals. Yuter and Houze [1998] have also suggested that in the stratiform region of anvils, ice crystals gently settle downward and grow slowly by vapor deposition. The shift towards larger ice crystals over the MCS lifetime could be a result of either of these processes.

Figure 5.

Evolution of cloud-ice properties over the MCS lifecycle. Histograms of (top row) ice water path and (bottom row) ice crystal diameter are shown. The first five columns show the lifecycle stages, and the final column shows the histograms computed from the entire lifecycle. Thin black lines show the histogram computed from the entire lifecycle. Bin widths are 20 gm2⁻² for ice water path and 1 μm for crystal diameter.

The average cloud properties during the different lifecycle stages are shown in Table 3. Confidence intervals for the mean are computed by using the random errors in the measurements described in Section 2a and by assuming that each MCS is independent, and therefore contributes one degree of freedom. These averages confirm that changes in the cloud population throughout the MCS lifecycle are statistically significant.

Table 3.

Average cloud properties for each stage of the MCS lifecycle. The first number is the mean and its 95% confidence interval. The median is shown in parenthesis. LW CRE is computed from all hours of the day, while SW and net CRE are estimated from daytime measurements (Section 3a).

	Lifecycle Stage
	I	II	III	IV	V	All
Optical Depth	54 ± 1 (29)	66 ± 1 (43)	67 ± 3 (46)	56 ± 2 (34)	41 ± 1 (24)	61 ± 1 (38)
Cloud-Top Pressure (hPa)	146 ± 2 (145)	133 ± 2 (125)	132 ± 4 (124)	146 ± 2 (145)	163 ± 1 (168)	138 ± 8 (133)
Albedo (%)	58.8 ± 0.1 (60.4)	61.5 ± 0.1 (63.7)	61.9 ± 0.2 (64.2)	60.1 ± 0.1 (61.8)	56.7 ± 0.1 (57.7)	60.7 ± 0.1 (62.7)
OLR (Wm−2)	121.0 ± 0.5 (123)	113.6 ± 0.8 (111)	114 ± 1 (112)	122.7 ± 0.8 (124)	132.5 ± 0.5 (136)	117.3 ± 0.4 (116)
SW CRE (Wm−2)	−197.7 ± 0.3 (−201)	−209.3 ± 0.4 (−215)	−208.9 ± 0.7 (−215)	−201.7 ± 0.5 (−206)	−187.3 ± 0.3 (−190)	−204.4 ± 0.2 (−210)
LW CRE (Wm−2)	164.9 ± 0.5 (163)	172.3 ± 0.8 (175)	172 ± 1 (174)	163.3 ± 0.8 (162)	153.5 ± 0.5 (150)	168.6 ± 0.4 (169)
Net CRE (Wm−2)	−33.8 ± 0.6 (−37)	−40.1 ± 0.8 (−44)	−39 ± 1 (−43)	−39.5 ± 0.9 (−43)	−34.9 ± 0.5 (−38)	−38.6 ± 0.4 (−42)
Ice Water Path (kg m−2)	0.98 ± 0.03 (0.51)	1.21 ± 0.05 (0.77)	1.28 ± 0.09 (0.87)	1.12 ± 0.06 (0.69)	0.83 ± 0.03 (0.48)	1.17 ± 0.02 (0.72)
Ice Crystal Diameter (μm)	67 ± 1 (68)	68 ± 2 (70)	71 ± 3 (73)	75 ± 3 (76)	76 ± 2 (77)	72 ± 1 (73)

4b. Evolution of the large-scale environment surrounding MCS

Having described the MCS lifecycle, we now investigate the evolution of the large-scale environment surrounding MCS. In this analysis, SW CRE data are treated as missing when the solar zenith angle is greater than 70°. To verify that this is not an issue, we first investigate the diurnal cycle of convection, which is shown in Figure 6. It is most common for cold-cores to reach their peak in the early morning between the hours of 0300 and 0900 local time, and least common for cold-cores to reach their peak in the afternoon between 1500 and 1800. Consistent with previous studies, the diurnal cycle of convection is apparent but not especially strong [Fu et al., 1990; Mapes and Houze, 1992; Chen and Houze, 1997]. Because cold-cores peak at all hours of the day, lead-lag composites with adequate sampling can be made from the daytime hours. This is more relevant for the SW CRE composites, since data from all hours of the day are used in the LW CRE composites.

Figure 6.

The diurnal cycle of convection. This histogram shows the local time when the cold-core reaches its maximum area.

The evolution of SW, LW, and net CRE within a large domain surrounding MCS are shown in Figures 7–9 respectively. Between 10 and 7 hours before the cold-core reaches peak coverage, SW and LW CRE are both around ±100 Wm⁻² and are nearly in balance. At this time a disturbance appears to the east of what will become the storm center. The disturbance propagates westward and eventually reaches the storm center when the cold-core reaches its maximum coverage. At this point a bright anvil is located in the middle of the composite. Near the storm center the average SW CRE is around −230 Wm⁻², LW CRE is around +180 Wm⁻², and net CRE is around −50 Wm⁻². The magnitude of the net CRE drops off sharply within ~100 km of the storm center. The disturbance continues to propagate to the west during the hours that follow. Near the center of the composite, the magnitude of the SW and LW CRE decline following the cold-core peak, and by 10 hours after the cold-core peak the SW and LW CRE are nearly in balance and back to their climatological values.

Figure 7.

The evolution of SW CRE over a large domain surrounding MCS. The composites are centered on the location of the storm center at the time when the cold-core reaches its peak coverage. The composites show SW CRE from 10 hours before the cold-core peak to 10 hours after. The lag relative to the time of the cold-core peak is shown in top left corner of each panel. Contours are in increments of 15 Wm⁻². Note that individual MCS are typically around 100–200 km in diameter, which is much smaller than this domain.

Figure 9.

The evolution of net CRE over a large domain surrounding MCS. The composites are centered on the location of the storm center at the time when the cold-core reaches its peak coverage. The composites show net CRE from 10 hours before the cold-core peak to 10 hours after. The lag relative to the time of the cold-core peak is shown in top left of each panel. Contours are in increments of 15 Wm⁻². Note that individual MCS are typically around 100–200 km in diameter, which is much smaller than this domain.

The westward-propagating disturbances seen here are consistent with previous studies. Nakazawa [1988], Hendon and Liebmann [1994], and Chen et al. [1996] showed that during the active phase of the Madden-Julian Oscillation, mesoscale variability is dominated by westward-propagating disturbances that contain clusters of enhanced convection. These disturbances have a spatial scale of around 1000 km, a phase speed of around 10–15 m/s, and a period of around two days. Takayabu [1994] argued that these disturbances are westward-propagating inertio-gravity waves [Matsuno, 1966], and Chen and Houze [1997] suggested that this mode is especially common because its period resonates with the cycle of deep convection followed by boundary-layer recovery. Note that at a given instant in time, a large-scale disturbance typically contains several separated clusters of enhanced convection (e.g. see Figure 2). The composites in Figure 7–9 smooth out this granularity, leaving just the envelope of the large-scale disturbance.

A similar plot showing the evolution of SSTs around MCS is presented in Figure 10. In this figure the composites have the same spatial scale as Figures 7–9, but the temporal resolution is three hours instead of one. Coarser temporal resolution is needed because these measurements come from a polar-orbiting satellite with more limited sampling. The measurements used to generate this figure were all made within several hours of 1330 local time. Because a SST retrieval cannot be made in moderate or heavy rain, points are marked if data are available for less than 50 MCS. These points occupy a small region near the storm center and close to the time of the cold-core peak – places where precipitation is expected. Outside of this region, statistically significant cold anomalies on the order of −0.1 °C are widespread, and they propagate westward following the large-scale disturbance. The coldest SST anomalies are around −0.6 °C and are found near the storm center following the time of the cold-core peak.

Figure 10.

The evolution of SST anomalies over a large domain surrounding MCS. Only daytime measurements are shown. The composites are centered on the location of the storm center at the time when the cold-core reaches its peak coverage. The lag relative to the time of the cold-core peak is shown in top left of each panel. Each panel is the average of three hours of data (e.g. the upper left panel is the average of data from 13.5 to 10.5 hours before the cold-core peak). Black dots show anomalies that are significantly different from zero at the 95% confidence level. Gray dots show sparse data: regions where a retrieval was made for less than 50 MCS. These measurements represent the temperature in the top 1 mm of the ocean.

The composites in Figure 10 show that convection can reduce SSTs over scales of hundreds of kilometers around the storm, but the mechanisms that cause the cooling are not immediately clear. Several possibilities exist: (1) shading by anvils cools the sea surface, (2) convection strengthens the large-scale near-surface winds, which stir the upper ocean and redistribute the solar heating over a deeper layer of water, or (3) convective downdrafts cool and dry the boundary layer, which enhances the air-sea heat flux [Zipser, 1969]. To investigate the mechanism that cools the SSTs we study similar composites but for the nighttime measurements, which were all made within several hours of 0130 local time. This is presented in Figure 11. Widespread cool SST anomalies beneath MCS are not seen at night. This has two implications: First, note that anvils can contain large hydrometeors, which have the potential to scatter microwaves and bias the SST retrieval. If the SST retrieval was impacted by such a bias, then cool SST anomalies beneath MCS would also be seen at night. Since this is not the case, the cool SST anomalies beneath MCS during the day are a real feature. Second, while low-level winds and convective downdrafts may be important cooling mechanisms on smaller scales, shading of the ocean surface from anvils causes the widespread cooling of SSTs following convection.

Figure 11.

Similar to Figure 10, but for nighttime measurements.

Given that the ocean mixed layer in the warm pool is around 30 meters deep [Lukas and Lindstrom, 1991], it may seem surprising that SSTs cool by several tenths of a degree over the lifetime of a MCS. How can shading from anvils produce this much cooling? Over the warm pool, the properties of the top few meters of the ocean can fluctuate rapidly and are often distinct from the rest of the mixed layer. This distinct upper layer, which is called a diurnal warm layer, is common for two reasons. First, solar heating of the ocean is concentrated near the surface. Half of solar energy that penetrates the surface is absorbed in the top 0.5 meters of the ocean [Soloviev and Lukas, 2006]. Second, precipitation exceeds evaporation by 1–2 meters per year over the warm pool, so the near-surface water is relatively fresh and stratified [Schmitt, 2008]. In the absence of strong wind-driven mixing, this stratification limits the effects of surface heat fluxes to the top few meters of the ocean [Soloviev and Vershinsky, 1982; Soloviev and Lukas, 1996]. The cooling of the sea surface during the hours following convection is likely due to a modification of the diurnal warm layer, not the entire mixed layer.

4c. The influence of SST on the triggering of deep convection

We have described the evolution of SSTs surrounding MCS, and now we will turn the analysis around and determine whether or not the SSTs influence the triggering of convection. To address this question we identify the instances in which no convective clouds are observed, and then compute the likelihood that convection will fire during the following several days. We then determine if the likelihood of triggering convection depends on the initial SST. The methods are described in Section 3e.

Figure 12 shows the evolution of cold-cloud fraction, which is a measure of the coverage of deep convective clouds, over a five-day period following instances when no cold clouds are observed. When interpreting this figure, note that the cold-cloud fraction typically reaches a maximum around 6–12 hours after convection initiates [Chen and Houze, 1997]. In this figure, the data are split into a composite in which SST anomalies are initially warm and initially cold. For warm SSTs, the cold-cloud fraction rebounds to its average value by day three and exceeds the average value on days four and five. Meanwhile, for cold SSTs, the cold-cloud fraction does not return to the average value over the five-day period. This means that if a region starts from a state with no convection, then convection is more likely to fire during the following five days if the SSTs are initially warm than if the SSTs are initially cold.

Figure 12.

The relationship between SST and triggering of deep convection. Cold-cloud fraction is a measure of the coverage of deep convective clouds. Instances of zero cold-cloud fraction are identified, and the evolution of the cold-cloud fraction over the following five days is shown. The “warm SST” line shows the cases when a warm SST anomaly was observed initially (i.e. at time 0), and the “cold SST” line is defined similarly. Error bars show the 95% confidence interval of the mean. The dashed line shows the climatological cold-cloud fraction. See Section 3e for details about the method.

While the triggering of convection does depend on the local SSTs, it also depends on stability of the atmosphere. Graham and Barnett [1987] showed that on monthly timescales, deep convection over the warmest tropical oceans is related to the surface wind divergence more closely than to local SST. Thus, while warm SSTs encourage convection, this effect can be temporarily overpowered by changes in atmospheric stability resulting from large-scale disturbances.

5. Testing the Hypotheses on the Balanced Cloud Radiative Effect

Having described the MCS lifecycle and evolution of the surrounding environment, we now return to the question of why the average SW and LW CRE cancel over the warm pool. First, consider the hypothesis that CRE is balanced due to a fortuitous coincidence [Kiehl, 1994]. MCS produce an ensemble of clouds with net CRE ranging from −104 Wm⁻² to +42 Wm⁻², which are the 5^th- and 95^th-percentiles, respectively (Figure 4). The upper limit in this range may be less than the true value since the MCS tracking algorithm does not capture thin cirrus. Furthermore, MCS have a substantial negative net CRE when averaged out to 10 hours after the cold-core peak (Figure 9). The CRE balance results from a cancelation between an immediate cooling effect from thick anvil cloud and a delayed warming effect from thin cirrus. A wide range of CRE values is seen, and this contradicts the strong coincidence hypothesis. However, it is possible that MCS form an ensemble of clouds with a variety of radiative effects, but the ensemble-mean net CRE is small due to a fortuitous coincidence. The weak version of the coincidence hypothesis cannot be ruled out.

Next, consider the hypothesis that the MCS cloud population is driven to a net neutral radiative effect by the radiative heating of anvils [Hartmann and Berry, 2017]. This hypothesis states that radiative heating lengthens the lifetime of anvils by lifting the entire cloud layer and by preferentially heating the cloud base, and therefore encouraging turbulent mixing within the cloud. These effects are strongest in thin and medium anvils, so radiative heating causes these clouds to persist longer than thick anvils. Because the MCS tracking results do not include thin cirrus, the radiative heating hypothesis cannot be tested here. However, since optically intermediate (3.6 < τ ≤ 23) and optically thick (τ > 23) clouds can be tracked, they can be compared to gain some insight into the effect of radiative heating on cloud lifetime. Of the tracked MCS clouds, the fraction of optically intermediate cloud is comparable to that of optically thick cloud (Figure 4). Furthermore, the lifecycle stages I and II, in which the MCS is growing and optically thick clouds dominate, do not last longer than lifecycle stages IV and V, in which the MCS is dissipating and optically intermediate cloud is more common (Table 2). These findings are consistent with Hartmann and Berry’s [2017] results. They found that thick and intermediate clouds are observed with about the same frequency, while the frequency of occurrence of thin cirrus is an order of magnitude larger (see their Fig. 2). These results suggest that radiative heating is especially important for the persistence of thin cirrus. Thus, future tests of the radiative heating hypothesis should focus on thin cirrus.

Next, consider the hypothesis that feedbacks between clouds, circulation, and SST cause the balanced CRE over the west Pacific [Hartmann et al. 2001]. We find that convection produces bright anvil clouds that start with SW CRE values around −230 Wm⁻² (Figure 7). In the hours following convection, cool SST anomalies as large as −0.6 °C are observed beneath anvils. Since MCS often occur in clusters, the cool SST anomalies can span hundreds of kilometers. The cooling of the sea surface following convection is seen during the day but not at night, so it must be caused by shading from anvils (Figure 10–11). Furthermore, convection is more likely to develop if SST anomalies are initially warm than if they are initially cold (Figure 12).

These results support the cloud-circulation-SST feedback hypothesis. Most notably, the main criticism of the cloud-circulation-SST hypothesis, namely that the SSTs evolve too slowly to explain the cancelation in CRE, is inconsistent with our results. We find that shading from anvils can significantly cool the underlying SSTs on timescales shorter than one day. Furthermore, the cooling of SSTs following convection is typically expressed over hundreds of kilometers, which is consistent with the length scales assumed by Hartmann et al. [2001].

One shortcoming of the cloud-circulation-SST feedback hypothesis is that it does not account for the relationship between atmospheric stability and convection. Waliser and Graham [1993] suggested that convection is related to both SST and the stability of the free troposphere, and therefore the feedbacks between SST and convection can be temporarily disrupted by forcing from large-scale disturbances. Our results support their claim. We find that convection often occurs within westward-propagating inertio-gravity waves (Figure 7–9). This result demonstrates that atmospheric stability and large-scale meteorology influence convection. However, it does not suggest that SSTs are unimportant. For instance, Chen and Houze [1997] argued that westward-propagating inertio-gravity waves are especially common because they are able to synchronize with the cycle of deep convection followed by recovery of the boundary layer and SST. Our results confirm that clouds, convection, and SST are coupled through two-way interactions, but these interactions can be modulated by forcing from large-scale disturbances.

6. Summary and Conclusion

Geostationary satellite data are used to objectively track MCS over the tropical west Pacific. We show that cloud radiative, macrophysical, and microphysical properties evolve significantly over the MCS lifecycle. In the growing and mature stages of the MCS, most clouds are optically thick, have tops between 100 and 120 hPa, and have a net CRE of around −50 Wm⁻². In the dissipating stages the cloud top sinks towards 200 hPa, clouds of medium optical depths become more common, the population of clouds shifts toward smaller CRE and more positive net CRE. Ice crystals become slightly larger as the MCS ages. Following convection, shading from anvils causes cool SST anomalies of up to −0.6 °C. Large westward-propagating disturbances containing multiple neighboring MCS are common, so the cooling of SSTs is typically expressed over hundreds of kilometers. Triggering of convection is more likely to occur if the local SSTs are anomalously warm than if they are cold, but this relationship is modulated when large-scale disturbances alter the stability of the free atmosphere.

These findings are used to investigate the cause of the close balance between SW and LW CRE over the west Pacific warm pool. Three hypotheses for the CRE balance are considered:

1. The CRE balance results from a fortuitous coincidence.

2. Feedbacks between cloud albedo, large-scale circulation, and SST require that net CRE be similar in neighboring regions of active and suppressed convection.

3. Radiative heating of the cloud causes medium and thin anvils to persist longer than thick ones, which causes net neutrality of the cloud population.

Our results support the cloud-circulation-SST feedback hypothesis (2) and show that earlier criticism of this hypothesis is not supported by observations. The SST responds on short time scales to the presence of MCS that shade the surface from insolation. These SST variations feed back on the likelihood of deep convection, which produces clouds with negative net CRE.

Future work should focus on further testing these hypotheses – especially the radiative heating hypothesis (3). This will require better understanding of the effects of radiative heating on the lifecycle of thin cirrus clouds. It might be possible to accomplish this by combining measurements from geostationary satellites with measurements from polar-orbiting sensors that are better suited for detecting thin cirrus (e.g. lidar or infrared sounders). Alternatively, the hypotheses on the balance in CRE could be evaluated using model simulations. Ideally this would involve a model that resolves both in-cloud turbulence and large-scale circulations that span the warm pool. With such a model it may be possible to perform targeted experiments that perturb the different processes one at a time. This is challenging at present because of the large computational cost, and the likely strong sensitivity of the results to microphysical parameterizations. In any case, determining why the SW and LW CRE are delicately balanced over the warm pool, and whether or not they will remain balanced in the future, is a problem that remains a high priority in climate research.

Figure 8.

The evolution of LW CRE over a large domain surrounding MCS. The composites are centered on the location of the storm center at the time when the cold-core reaches its peak coverage. The composites show LW CRE from 10 hours before the cold-core peak to 10 hours after. The lag relative to the time of the cold-core peak is shown in top left of each panel. Contours are in increments of 15 Wm⁻². Note that individual MCS are typically around 100–200 km in diameter, which is much smaller than this domain.