During the campaign, radiosondes were launched over T3 at regular 6 h intervals (01:30, 07:30, 13:30 and 19:30 LT, Vaisala RS-92 radiosondes; ARM, 1993). For the IOPs, one additional radiosonde was launched at 10:30 LT to enhance diurnal coverage. Basic thermodynamic processing was performed following Jensen et al. (2015) to estimate convective forcing parameters such as the lifting condensation level (LCL), mixed-layer height (MLH), convective available potential energy (CAPE) and convective inhibition (CIN). For each of these parameters, surface parcels are defined by the level of the maximum virtual temperature in the lowest kilometer. This represents the most buoyant parcel in the boundary layer and maximizes the calculated CAPE (thus, our reported values are comparable to “most unstable” CAPE or MUCAPE). The MLH is calculated using the definition of Liu and Lang (2010) that determines the MLH from a combination of the gradient of potential temperature and the vertical wind shear using criteria based on the stability of the boundary layer and the presence of a low-level jet. Surface radiative flux estimates for this study follow the radiative flux analysis methods of Long and Ackerman (2000) and Long and Turner (2008). The clear-sky radiative flux estimates are produced by employing an empirical function fitting approach during observed clear-sky periods. The fitted coefficients from these clear-sky intervals are used to interpolate over cloudy periods, providing a continuous estimate of clear-sky irradiances and quality-controlled cloudy-sky fluxes. Detailed analyses of cloud radiative effects are located in Sect. 4.

Figure 2 presents the cumulative time series for the 2-year GoAmazon2014/15 dataset in terms of average daily profile values for basic cloud, precipitation, thermodynamical and dynamical observations from multi-sensor ground instruments at the T3 site. These efforts complement previous papers on seasonal variability for cloud conditions over the larger Amazon basin (e.g., Machado et al., 2004). We observe clear shifts in several quantities associated with the Amazon wet and dry seasons. Our ranges for wet and dry season months, as well as the IOPs, are shown in Fig. 2 as a reference to the appropriateness for those windows compared to the larger-scale conditions. The more pronounced shifts during the wet season include increased CF in the mid-to-upper troposphere (between 3 and 10 km; Fig. 2a), higher precipitation rates (over these daily integrations) and precipitable water (PW; Fig. 2b), as well as the buildup of relative humidity (RH) profiles through the middle levels (Fig. 2d). Previous studies suggest that CAPE, CIN and zonal/meridional winds (Fig. 2c, e and f) from radiosondes may also illustrate large-scale thermodynamical changes and moisture transport associated with wet, dry and transitional periods (e.g., Li and Fu, 2004; Fu and Li, 2004). Radiosonde daily maximum values indicate only small seasonal changes in CAPE and CIN, although we observe that the transitional periods between the dry and wet seasons promote maximum relative CAPE trends coupled with relatively lower CIN and heightened moisture. These are the primary ingredients that promote more frequent and strong convection, provided convection can be triggered (e.g., Machado et al., 2004). Although areal coverage of deeper convection is generally the largest during the wet season, recent profiler-based studies suggest the strongest storms (in terms of upward vertical air motion) were often observed towards the end of the dry season and into the transitional period (e.g., Giangrande et al., 2016; Nunes et al., 2016).

The diurnal variation of atmospheric state is illustrated in Fig. 3 and shows the evolutions for the mean and standard deviation of (a) CAPE, (b) CIN, (c) LCL and (d) MLH separated into dry (red bars) and wet (blue bars) components (e.g., Betts et al., 2002). CAPE increases after sunrise, reaching a maximum near midday, whereas CIN is maximum (largest negative value) overnight and decreases during the day. These behaviors are consistent with development of convection breaking the capping inversion and consuming CAPE. Both CAPE and CIN show a stronger diurnal cycle during the dry season compared to the wet season. The mean LCL increases by approximately 600–800 m from sunrise to the afternoon with larger magnitudes and range during the dry season. The mean MLH also increases by approximately 1 km from sunrise through the afternoon during the wet season, while during the dry season the increase is about 1.5 km. This increase in MLH is consistent with daytime solar heating. Separating the diurnal cycle into dry (red bars) and wet (blue bars) season components indicates slightly stronger diurnal cycle signatures in CAPE, increased CIN (e.g., larger negative values) and higher MLH for the dry season (similar to measurements obtained in the southwest Amazon by Fisch et al., 2004), with a suppressed diurnal cycle in LCL height.

To better inform the observed cloud system variations over the ARM T3 site from the large-scale environmental condition perspective, Fig. 4 plots the diurnal cycle of the large-scale vertical motion (omega), total advection of moisture and relative humidity. The total advection of moisture is the sum of the horizontal and the vertical advection: q_adv_t=-Vh⋅∇hq-ω∂q∂p. The total advection can be interpreted as 3-D moisture convergence. For the GoAmazon2014/5 period, the vertical component dominated the total moisture advection (not shown). These large-scale fields are derived from the ECMWF analysis outputs over the entire field campaign using a constrained variational analysis method of Zhang and Lin (1997). The upper-level state variables (wind, temperature, moisture) from ECMWF are adjusted to conserve column-integrated mass, moisture and energy. Surface rainfall rate from the SIPAM radar is used as a major constraint. Additional details on these large-scale fields for the GoAmazon2014/5 deployment can be found in Tang et al. (2016). This variational analysis is performed at 3-hourly intervals at 25 hPa vertical resolution over a domain of about 110 km in radius, with the center located at the T1 site (Fig. 1).

The omega field shows strong upward air motion in the middle and upper troposphere during the mid-to-late afternoon (Fig. 4a). The evening and early morning hours exhibit upward air motion confined below 3–4 km due to lower-level convergence and middle-level divergence (not shown), likely corresponding to the congestus clouds. Above that level, downward air motion is dominant. This downward motion is most pronounced between 06:00 and 09:00 LT. After sunrise, we observe low-level weak ascending motions and positive advection of moisture (Fig. 4d). Between 4 and 8 km in the RH field, we observe dry middle tropospheric conditions and relatively wetter conditions near the tropopause (Fig. 4g). Similar structures in all fields are found across wet and dry season breakdowns; however, middle- and upper-level descending motions during the evening and early morning hours are much stronger during the dry season, suppressing convection during those hours. In addition, the ascending motion between noon and late afternoon is much weaker in the dry season compared to the wet season. The dry season also exhibits reduced low-level positive moisture advection and a much dryer lower and middle atmosphere.

The DOE AAF G1 aircraft participated in two IOPs that coincided with the AMF deployment. Airborne measurements were conducted during 22 February–23 March 2014 and 6 September–4 October 2014, representative of the wet and dry seasons, respectively. The G1 flight patterns were designed to sample shallow and growing cumulus convective clouds that formed downwind from Manaus to examine the evolution of urban pollution and its effect on cloud and precipitation properties (Martin et al., 2017). Typical flights consisted of a series of level legs flown just below cloud base, just above cloud base and higher in growing cumulus clouds, including legs over the T3 ground site. In total, 16 and 19 flights in warm cumulus clouds were included in the wet and dry seasons, respectively.

The G1 payload was designed to measure the full spectrum of aerosol size from 0.015 to 3 µm and cloud particle sizes from 2 µm to 1.92 cm. For this study, three cloud particle distribution probes are combined to create the full drop-size distribution (DSD) depictions presented in Sect. 4. The Droplet Measurements Technologies (DMT) cloud droplet probe (CDP; 2–50 µm) is combined with the Spec Inc. two-dimensional stereo probe (2-DS; 10 µm–3 mm) between 20 and 50 µm by averaging the overlapping bins. The Spec Inc. high-volume precipitation spectrometer (HVPS; 150 µm–1.92 cm) is used for droplets larger than 500 µm. Cloud droplet distributions are combined by averaging the DSD for each instrument separately over these flight periods. This was done for in-cloud conditions only. DSDs from the CDP are used for drops smaller than 20 µm. The DSDs from the CDP and 2-DS are averaged between 20 and 50 µm, 2-DS DSDs are used between 50 and 500 µm, and HVPS DSDs are used for drops larger than 500 µm. The 2-DS probe occasionally contained artifacts known as “stuck bits”, i.e., when a photodiode becomes continuously occulted due to optical contamination or electronic noise (Lawson et al., 2006). Each flight was visually inspected for artifacts, which were manually removed from the combined DSDs. Cloud condensation nuclei (CCN) were measured with a dual-column system manufactured by DMT (operated with a constant pressure inlet at 600 mbar), and liquid water content (LWC) was measured using a multi-wire element probe (Science Engineering Associates (SEA) water content meter WCM-2000) with wire sizes the same as King and Johnson-Williams probes.

The 95 GHz W-band ARM cloud radar (WACR) (e.g., ARM, 2005; Giangrande et al., 2012) is the primary profiling instrument to characterize the cloud conditions during GoAmazon2014/5. Cloud masking and designation products are performed using the multi-sensor WACR preprocessing approach following active remote sensing of clouds methodologies (ARSCL; Clothiaux et al., 2000; Kollias et al., 2005, 2009) and additional quality-control refinements following Kollias et al. (2014). These retrievals merge observations from the WACR and a collocated laser ceilometer, micropulse lidar (MPL) and microwave radiometer (MWR) to better identify cloud boundaries in the vertical at high temporal (∼ 10 s) and vertical (∼ 24 m) resolution.

There are several limitations when designating cloud boundaries and hourly CF observations from vertically pointing cloud radars beyond the capabilities of single radar platforms or ARSCL methods (e.g., Lamer and Kollias, 2015; Oue et al., 2016). The primary limitation among these is that the WACR experiences attenuation in rain that manifests as erroneously low or missing cloud-top boundaries (e.g., Feng et al., 2009, 2014). To lessen these impacts within this Amazonian deployment that favors frequent precipitating cumulus, a collocated and well-calibrated 1290 MHz ultra-high frequency (UHF) radar wind profiler (RWP; 8∘ beamwidth, 200 m gate spacing, 6 s temporal resolution) was co-gridded to improve cloud coverage through deeper precipitating clouds (e.g., ARM, 2009; Giangrande et al., 2013, 2016). For this study, a modification to the ARSCL cloud boundary designation is produced by merging RWP profiles (operating in “precipitation” modes, as also described in Tridon et al. (2013) during precipitation intervals following similar ARSCL-type cloud profile processing (Feng et al., 2014). The substitution is accomplished using collocated surface rain gauge datasets to help define appropriate “precipitation periods”. These are defined as continuous time periods when the surface rain rate from the gauge exceeds 1 mm h-1. During these intervals, if more than 10 % of the derived WACR first echo-top heights associated with these precipitating clouds are found to be 500 m or more below the echo-top height as recorded by the RWP, a WACR attenuation flag is assigned and the RWP profiles and boundaries are inserted.

Figure 5 illustrates an example of the composite cloud designation for the 1 April 2014 event. Earlier during this event, both the WACR (Fig. 5a) and RWP (Fig. 5b) struggle to sample the thin and/or high cloud regions observed by ARSCL (Fig. 5c). CF estimates in these regions benefit from the additional ceilometer and MPL observations (not shown in Fig. 5) to detect clouds. Congestus clouds, including those that have cloud tops at ∼ 6 km, are observed reasonably well by both radars. In these times, surface precipitation is limited (Fig. 5d). A deep convective cloud system passes over T3 between 15:00 and 19:00 UTC. Heavy precipitation (surface measured rain rate > 60 mm h-1) is associated with extinction of the WACR signal, whereas the RWP is able to reconstruct cloud boundaries up to 13 km. Since the RWP is sensitive only to precipitation-sized particles, these methods will still underestimate the true cloud top. Additional precipitation periods are also identified by red bars on top of Fig. 5c, highlighting locations where the cloud boundary designation within precipitation is improved over traditional ARSCL methods.

A simple cloud-type classification is performed on the cloud boundary and masking dataset from Sect. 2.3. This approach follows McFarlane et al. (2013) and Burleyson et al. (2015). These methods classify clouds into seven categories according to the height of the cloud boundaries and cloud thickness. The cloud categories include shallow, congestus, deep convection, altocumulus, altostratus, cirrostratus/anvil and cirrus (definitions summarized in Table 1). Figure 5c provides an example of the cloud classifications for 1 April 2014. Cloud classification is used to separate surface radiative properties among the different cloud types. To accomplish this, the nearest cloud profile is matched to the 1 min surface radiative flux data. As with Burleyson et al. (2015), the lowest cloud type present in the column during that time is used to designate the shortwave and longwave radiative flux measurements (Fig. 5e and f) for that cloud type.

Figure 6 shows diurnal CF profile breakdowns for each cloud category. The “cirrus” and “shallow” cloud categories are combined into a single panel since these cloud definitions do not overlap in altitude. Seasonal variations in the diurnal CF by cloud category are described in the next section. Figure 6a indicates that cirrus clouds are the most commonly observed clouds during the afternoon and overnight hours, whereas shallow cloud observations dominate the early morning hours after sunrise into the mid-afternoon. Combining Fig. 6a with summary cloud occurrence values in Table 2, shallow cumulus clouds in the Amazon are observed with relatively high frequency throughout most of the day (∼ 22 %). Shallow clouds in the early morning align with low-level weak ascending air motions (Fig. 4a) and the positive advection of moisture (Fig. 4d). The most common cirrus cloud locations correspond to relatively high RH regions in the upper atmosphere seen in Fig. 4g, where the air is close to saturation with respect to ice (not shown).

Congestus (Fig. 6c) and deep convective (Fig. 6d) clouds are prominent starting at noon into the late afternoon, with peak CF coverage 1–2 h after local noon. Deeper clouds that include MCS passages (identifiable using SIPAM observations) appear to maintain higher CFs into the overnight hours (associated with trailing stratiform regions). Integrated column behaviors are similar to those found from satellite over Manaus from Machado et al. (2004); specifically, cloud coverage is high throughout the day, peaking after local noon and associated with increased cirrus (Fig. 6a), cirrostratus (Fig. 6b) and deeper convective clouds (Fig. 6d). The T3 location exhibits a pronounced diurnal cycle associated with deeper convection (as also in Saraiva et al., 2016). This pronounced behavior is further representative of the fortuitous placement for the T3 AMF site, wherein daily cloud life cycles also phase well with propagating sea breeze intrusions over this portion of the Amazon basin (Burleyson et al., 2016). Primary diurnal peak behaviors of these clouds (enhanced afternoon convection and reduced overnight convective development) are associated with mid- and upper-level upward motion in the afternoon and downward air motion from night to the early morning (Fig. 4a).

Congestus and altocumulus exhibit weak peaks in the pre-dawn hours (around 05:00 LT). This is observed primarily as a wet season congestus behavior, possibly comparable to suggestions in previous Manaus diurnal rainfall efforts (e.g., Machado et al., 2004). However, this contribution would typically be dwarfed when combining the rainfall contributions from other cloud types. We note that the non-precipitating categories of altocumulus, altostratus and cirrostratus (Fig. 6b, e and f) share similar diurnal phasing with cirrus clouds. Cirrus and cirrostratus are more commonly observed than alto-cloud designations. However, we have not differentiated the contributions to cirrus CF estimates that reflect deep convective or anvil cloud components from other cirrus clouds. Overall, inspection of large-scale forcing fields supports the notion that the diurnal cycle of high-level clouds is not well associated with the diurnal cycle of the large-scale dynamics and thermodynamics. This may be indicative of the importance of clouds that originate from anvil remnants from deeper convective clouds or advect from elsewhere, in addition to the clouds that are forced and developed locally.

Figure 7 shows the diurnal cycle of precipitation properties at the T3 site as observed by the surface gauges collocated with RWP observations. These plots include wet, dry and transitional season contributions (although we do not isolate these transitional months). Average precipitation rates reflect the average across precipitating and non-precipitating days, peaking around 12:00 to 16:00 LT (Fig. 7a), consistent with the deep convective CF in Fig. 6d. Note that the mean rainfall rates including only days when precipitation is present (Fig. 7b) are more comparable between wet and dry season events, suggesting T3 results in Fig. 7a primarily reflect the additional frequency of convection during the wet season and not its relative intensity. The total rainfall accumulations for the various regimes are presented in Fig. 7c. Rainfall accumulations from Fig. 7c have also been separated into convective and stratiform types as designated by the RWP when available (Giangrande et al., 2016). For the composite campaign (dashed line in Fig. 7d), convective precipitation is dominant at ∼ 76 % of the fractional accumulation (∼ 2300 mm) with a relatively flat contribution diurnally (especially during the dry season). Since this fractional accumulation is based on RWP estimates for convective fraction, it may tend to maximize convective precipitation fraction over traditional scanning radar-based retrievals (e.g., Steiner et al., 1995). This is because unlike basing these designations on radar reflectivity factor properties and buffering (proximity to convective cores based on intensity), profiler methods also distinguish columns with convective vertical air motions (including those from elevated, sloping updrafts that extend back into the transitional or trailing stratiform regions for MCSs), as well as congestus cloud precipitation (typically associated with clouds having echo tops exceeding 4 km) as “convective” rainfall. Stratiform precipitation (approximately 700 mm for this dataset) is more frequent (in terms of accumulation) during the overnight hours (30–60 %) and is associated with the trailing precipitation regions from the convective systems. Again, based on our cloud classifications, deep convective clouds would contain convective and widespread stratiform precipitation components.

Figure 7 also plots the diurnal breakdowns for average rainfall rate and fractional convective accumulations during the two wet and dry seasons over T3. Our dataset contains 103 wet season days responsible for approximately 1600 mm of precipitation and 52 dry season days responsible for approximately 600 mm of precipitation. The wet season months are associated with a factor of 2 increase in mean rainfall rates, but even larger increases occur during daytime hours with much smaller changes during the late evening and early morning (Fig. 7a). However, relative to those days having precipitation (Fig. 7b), the differences in the mean rainfall rate are less pronounced. This may support the notion that the dry season convection is stronger (instantaneously), since the overall convective cell coverage is also reduced during the dry season (e.g., Giangrande et al., 2016; Schiro, 2017). Wet and dry season convective rain fractions have similar values (∼ 80 %) throughout most of the day (Fig. 7d) and only diverge during the early morning hours when wet season convective rain fractions drop to as low as 20 %. These diurnal patterns suggest that organized MCSs pass over T3 primarily in the morning hours during the wet season but are infrequent and only have a small impact on the multi-month mean statistics.

Figure 8 plots seasonal breakdowns for the CF diurnal cycle from Fig. 6 for the wet and dry seasons. Pronounced CF profile increases are associated with deep convective and congestus clouds during the wet season, with the dry season having less organized cloud contributions (Ghate and Kollias, 2016). Overnight and/or pre-dawn deep convection and additional local congestus development are more common in the wet season. The distinct nighttime enhancement in stratiform precipitation (once again, often categorized under “deep convection” in Figs. 6d, 8g and h) during the wet season is consistent with previous findings that propagating convective cloud systems contribute to the observed diurnal cycle of deep convection (e.g., Burleyson et al., 2016; Tang et al., 2016). Early morning shallow cumulus CF profiles (Fig. 8a and b) indicate frequent low clouds during wet and dry seasons, consistent with a response to increased surface heating and an increase in the surface latent heat flux, with the wet season reporting additional shallow cloud development throughout the diurnal window (and the dry season consistent with elevated LCL heights). Two separate vertical peaks of shallow cumulus CF were observed between pre-dawn and early morning hours (03:00–09:00 LT) during the wet season: one right above the surface and the other at 2 km height. The surface peak is possibly associated with overnight fog being lifted with surface heating associated with the rising sun (e.g., Anber et al., 2015), while the elevated peak may be associated with radiative cooling of the residual boundary layer overnight. Cirrus CF stays elevated during the wet and dry seasons; however, cirrostratus/anvil CFs are substantially reduced during the dry season. This pattern suggests mostly a local deep convective contribution to cirrostratus/anvil during the dry season, with local convection and potentially some additional remnant anvil or decaying MCS cloud components advected over T3 during overnight hours under wet season conditions. Quantitative interpretation for these behaviors is challenging due to coupled cirrus–shallow cloud sampling factors during the overnight hours. For example, it is likely cirrus sampling is shielded (results stemming from an MPL detection) during the wet season due to the added presence of lower-level clouds and higher relative humidity/attenuation limiting the usefulness of the cloud radar. In contrast, clear low-level conditions during the overnight hours of the dry season would likely promote improved designation of cirrus. In this regard, wet and dry season cirrus cloud contrasts may be more pronounced than reported by this study.

As highlighted in Figs. 2 and 3, wet season thermodynamical conditions typically favor weaker CAPE, weaker CIN and higher RH in the lower to mid-atmospheric levels, while dry seasons feature stronger CAPE, stronger CIN and lower RH at the same levels. As inferred from the large-scale forcing fields in Fig. 4, wet season conditions favor higher column relative humidity, as well as heightened moisture convergence throughout the profile. The wet season also features more favorable omega fields at middle levels for shallow to deeper convective cloud transitions. This behavior is not surprising and also consistent with forcing datasets being constrained using mean domain precipitation estimates. However, as with first-year GoAmazon2014/5 studies (e.g., Collow et al., 2016), only weak correlations are found between cloud state and thermodynamic parameters (not shown). Nevertheless, coupled thermodynamical and environmental forcing conditions from Sect. 2 support these observations of more frequent cloudiness during the wet season, visible across almost all cloud categories (Fig. 8).

An important consideration when interpreting ARM T3 observations is the spatiotemporal representativeness of the cloud fractions over T3 within this central Amazon region. Satellite observations provide one avenue to better understand the representativeness of the location, while also providing insights into the advantages afforded by ground-based ARM cloud observations. Several recent GoAmazon2014/5 studies including Burleyson et al. (2016) and Giangrande et al. (2016) have investigated the representativeness of the T3 site as compared to satellite and/or radar perspectives. In these studies, they found evidence suggesting that T3-observed deep convection is well correlated with the regional cloud and precipitation to within a few hundred kilometers of the site. As a complementary reference, we approach the topic from a shallow cumulus perspective, as these clouds provide some of the better ground- and aircraft-based dataset opportunities collected during this campaign.

Figure 9 provides a diurnal comparison plot of shallow cumulus cloud fraction between T3 ground-based estimates, the Geostationary Operational Environmental Satellite (GOES) SatCORPS (Minnis et al., 2011) and the Moderate Resolution Imaging Spectroradiometer (MODIS; Platnick et al., 2003) products for wet and dry seasons (comparisons performed between February and December of 2014 when both satellite datasets are available). Both satellite cloud products are based on multi-channel passive sensors on board, providing fundamentally different cloud detection techniques compared to active sensors at T3. We chose these satellite data products because they provide large area coverage and frequent updates, making the comparison with single-point surface observations more amenable. The cloud property retrievals are available at 4 km spatial and 30 min temporal resolution for GOES, and at 1 km spatial and two overpasses during ∼ 10:30 and ∼ 13:30 LT for MODIS. Shallow cumulus clouds are defined at pixel-level cloud-top height below 3 km, consistent with T3 ground-based definitions. Shallow cumulus cloud fractions are then calculated at several spatial coverage domains (from 25 to 150 km) for GOES and at 25 km domain for MODIS. We find that the GOES mean CF estimates are similar in their diurnal cycle and magnitude as a function of the domain size, with increasing variability in CF estimates (shadings) to the smaller domain sizes. The MODIS mean CF estimates are higher than those obtained from GOES but share similar diurnal patterns and ranges of variability to the GOES smaller domain estimates.

In comparison with the satellite observations, single-point T3 ground estimates of shallow cumulus are the largest and carry the largest observational spread but demonstrate diurnal and peak behaviors that are roughly comparable to the satellite counterparts. Discrepancies in CF estimate (mean) magnitude are not surprising, with the largest discrepancies found in the wet season. These differences may be related to several factors including the obscured view of shallow clouds from passive satellite retrievals, for example, due to the blockage from higher clouds (i.e., multi-layer clouds). Moreover, over the entire field campaign period, single-layer clouds occur ∼ 48 % of the time, while multi-layer clouds occur ∼ 20 % of the time. During the wet seasons, multi-layer clouds occur twice as often as during the dry seasons (∼ 28 % vs. ∼ 11 %). The coarser resolution and/or sensitivity of the satellite platforms to shallower cumulus may also contribute to the lower mean CF values. The higher-resolution MODIS product shows higher shallow cumulus CF estimates than GOES during MODIS overpass times, demonstrating additional benefits of increased resolution in detecting shallow cumulus clouds, especially compared to ground-based active remote sensing observations. Finally, unavoidable discrepancies and variability still may trace to spatial domain versus column-temporal CF definition differences (e.g., Berg and Stull, 2002) or factors including localized circulations generated by river breezes (e.g., Burleyson et al., 2016). Overall, this shallow cumulus comparison between T3 ground-based and satellite observations suggests that mean cloud fractions near the T3 location should be representative of larger domain cloud properties to within a few hundred kilometers.

The average aggregated shortwave (SW) and longwave (LW) fluxes and CRE measured at the T3 site are given in Table 3, along with long-term results from the three TWP sites (Darwin, Manus, Nauru) as reported in Burleyson et al. (2015). SW CRE dominates (magnitude) as compared to LW CRE. The mean SW CRE (-94.4 Wm-2) and LW CRE (14.5 Wm-2) averaged across the diurnal cycle (nighttime included) over the entire GoAmazon2014/5 are most similar to those found at Manus, which is the cloudiest of the three TWP sites and most influenced by convection in the western Pacific warm pool. The Darwin, Australia, site has a strong monsoonal cycle (i.e., wet/dry season) and the Nauru site is strongly impacted by the El Niño–Southern Oscillation (ENSO) variability (Burleyson et al., 2015). Manus would be the one most qualitatively consistent with the GoAmazon2014/5 “green ocean” moniker. During the wet season, SW CRE (-121.5 Wm-2) is twice the dry season value (-60.4 Wm-2), although CREs for this region of the Amazon basin are substantially larger than for Darwin during all seasons. However, although mean behaviors initially appear similar, the properties from individual cloud types are not necessarily consistent between, for example, Manus and Manaus (e.g., see Burleyson et al., 2015; Table 4). A more thorough breakdown of the factors driving these specific differences (e.g., the time of day when a given cloud type is more or less prevalent, the frequency of multi-layer clouds or variance in the clear-sky downwelling SW or LW flux) is recommended as a future activity from these GoAmazon2014/5 datasets.

Table 2 gives bulk cloud frequency and their radiative characteristics separated by cloud types and by season. The results reveal the averaged reduction of downwelling SW flux when a particular cloud type is present. Consistent with the SW transmissivity results (Table 2) and those found by Burleyson et al. (2015), congestus and deep convective clouds dominate the conditional (e.g., not a mean property) SW CRE, while cirrus clouds have the smallest effect on downwelling SW flux. Note that the conditional CRE presented in Table 2 includes both single-layer clouds, as well as when additional cloud layers are above the lowest detected cloud layer. This is done deliberately to be consistent with the method used by Burleyson et al. (2015) such that the GoAmazon2014/5 results can be directly compared with their long-term results from the ARM TWP sites. Examination of the averaged conditional SW CRE calculated using only single-layer clouds reveals a relative reduction of ∼ 26 % for altocumulus and ∼ 20 % for shallow cumulus clouds and negligible difference in other cloud types. The reduction in SW CRE when single-layer clouds are considered is likely caused by frequent multi-layer cloud occurrence of cirrus/cirrostratus clouds over shallow cumulus or altocumulus (i.e., artificially inflating the surface SW CRE of cumulus clouds due to additional SW flux reflection by the upper-level clouds). The difference in conditional SW CRE between single- and multi-layer clouds does not change their contribution to the average CRE as discussed below.

Comparisons between wet and dry season diurnal behaviors for the frequency of the lowest clouds in the column and the associated mean SW CRE are shown in Fig. 12. Shallow cumulus clouds dominate the SW CRE in both seasons, although their frequency peaks 2 h earlier during wet season (10:00–11:00 LT) than during dry season (12:00–13:00 LT). While the dry season features reduced frequency and SW CRE of all cloud types, the contrast is most visible for the three convective cloud types. Shallow, congestus and deep convective cloud mean SW CRE in the wet season are 50, 69 and 72 % larger than those in the dry season, respectively (their mean SW CRE values across the diurnal cycle are shown in Fig. 12c, d).

From the previous section, shallow cumulus (those most frequently observed during the campaign) are associated with large discrepancies in cloud radiative effects between the wet and dry seasons (Table 2 and Fig. 13). Further investigation into these clouds and their radiative differences is enabled using aircraft observations available during the GoAmazon2014/5 campaign IOPs. As discussed in Sect. 2.2, three cloud particle size distribution probes are combined to create the full DSD (Fig. 13). Combining the cloud microphysical properties in shallow cumulus measured by aircraft observations and the cloud macrophysical properties measured by ground-based instrumentations allows us to explain the cloud radiative effect differences from wet and dry seasons reported in the previous section.

Cloud particle size distributions (Fig. 13c) in the wet season are characterized by a lesser occurrence of small droplets and a more frequent occurrence of large droplets when compared with cumulus clouds in the dry season. Total number concentration of cloud drops is more than a factor of 2 larger in the dry season than in the wet season (Fig. 13b). However, the corresponding LWC is roughly the same between the seasons (Fig. 13d). In situ cloud condensation nuclei (CCN) concentration is also larger in the dry season than in the wet season (Fig. 13a). Aircraft cloud and CCN measurements are consistent with studies that show clouds influenced by aerosol tend to have larger concentrations of smaller droplets and fewer precipitation-sized drops for clouds with similar LWC (e.g., Twomey, 1974; Cecchini et al., 2016). Ground-based radar measurements of single-layer shallow cumulus clouds at the T3 site show thicker clouds occurring more frequently in the wet season (Fig. 13e). Likewise, more frequent occurrence of large liquid water path (LWP) from the T3 ground-based MWR in the wet season is consistent with the presence of more robust (i.e., vertically developed) shallow cumulus clouds (Fig. 13f). Therefore, shallow cumulus clouds in the wet season are characterized by fewer but more frequent larger cloud droplets, while those in the dry season are characterized by more frequent smaller cloud droplets (Fig. 13b, c). Interestingly, these differences in DSDs result in comparable LWC between the wet/dry seasons (Fig. 13d). As a result, the stronger shallow cumulus conditional SW CRE in the wet season, which reflects the difference in microphysical properties, mainly arises from higher values of vertically integrated properties such as LWP and cloud thickness (Fig. 13e, f).