1. Introduction
Convection is an important component of tropical meteorology and the general circulation (Riehl and Malkis 1958; Schumacher et al. 2004; Trenberth and Stepaniak 2004; Richter et al. 2012). Yet general circulation models (GCMs) struggle to capture the intensity, timing, location, and transitions of tropical precipitating systems (Betts and Jakob 2002; Li et al. 2006; Vera et al. 2006; Yang and Smith 2006; Stratton and Stirling 2012; Bechtold et al. 2014; Folkins et al. 2014; Rieck et al. 2014; Anber et al. 2015; Adams et al. 2017; Fiedler et al. 2020; Hagos et al. 2021). These inaccuracies are manifested in deficiencies in the simulations of numerous global features such as the Madden–Julian oscillation (Kim et al. 2011; Miyakawa et al. 2012; Holloway et al. 2013), the spatial patterns of intertropical convergence zone (ITCZ) precipitation (Möbis and Stevens 2012; Voigt et al. 2014; Nolan et al. 2016; Wodzicki and Rapp 2020), the Walker circulation (Espinoza et al. 2016; Barichivich et al. 2018), and Kelvin waves (Serra et al. 2020). The Amazon basin covers 7 million km2 of rain forest and has one of the most complex tropical rainfall patterns on the planet (Saraiva et al. 2016). Rainfall over the Amazon is associated with several meteorological features including the ITCZ and South Atlantic convergence zone, Atlantic easterly waves, sea breeze fronts that migrate hundreds of kilometers inland, and mesoscale convective systems (MCSs). Numerical prediction systems (i.e., global climate models) must accurately represent the cloud population, precipitation, and subsequent latent heat release produced via all these mechanisms in order to correctly depict the global circulation (Williams 2002; Trenberth and Stepaniak 2004; Richter and Xie 2008; Nobre et al. 2009; Yin et al. 2013).
In the Amazon, shallow clouds (i.e., clouds with widths on average less than 1 km and cloud-top heights less than 3 km; Zhang and Klein 2013; Burleyson et al. 2015, 2016; Giangrande et al. 2017; Giangrande et al. 2020) are a major contributor to the total cloud population (Wright et al. 2017; Giangrande et al. 2020). These clouds occur nearly 28% of the time during the wet season (December–April) and more than 30% of the time from 1100 to 1600 local time (LT; local time is roughly UTC − 4 h in the Amazon) (Giangrande et al. 2017). Shallow clouds dominate the surface shortwave cloud radiative effect with a maximum between 1100 and 1500 LT (Giangrande et al. 2017). Because shallow cumulus clouds generally occur on spatial and temporal scales that are smaller than the model grid spacing and time step, shallow cumulus clouds and their impact on the thermodynamic profile of the environment must be parameterized. Deep convective clouds (i.e., clouds with bases less than 3 km and tops greater than 8 km, Giangrande et al. 2017) over the Amazon span a wide range of spatial scales (from tens to hundreds of kilometers) and temporal scales (from minutes to many hours), this variety of scales contributes to the challenges of prediction (Adams et al. 2017; Senf et al. 2018). Deep clouds during the wet season occur on average 9% of the time each day and more than 26% of the time from 1600 to 0100 LT (Giangrande et al. 2017).
The Green Ocean Amazon 2014/15 (GoAmazon2014/5, hereinafter GoAmazon) experiment took place from 1 January 2014 through 31 December 2015 in the region of Manaus, Brazil (Martin et al. 2016). GoAmazon involved a coordinated effort of nine surface-based observation sites [including Atmospheric Radiation Measurement (ARM) instruments] and aircraft flights. GoAmazon found that shallow cumulus clouds have a large influence on the meteorological conditions in the Amazon and the global radiative budget through the transition of such clouds to deep convection (Collow and Miller 2016). The most important factor for the transition of shallow-to-deep convection found in numerous GoAmazon studies was humidity in the boundary layer and middle troposphere (Itterly et al. 2016; Schiro et al. 2016; Giangrande et al. 2017; Zhuang et al. 2017; Chakraborty et al. 2018; Zhuang et al. 2018; Giangrande et al. 2020; Tian et al. 2021). However, difficulties exist in simulating the transition of shallow-to-deep convection (Li et al. 2006; Vera et al. 2006; Rochetin et al. 2014; Kang and Ryu 2016; Lee et al. 2019), contributing to underestimated rainfall rates in GCMs (Yin et al. 2013; Peters et al. 2017; Hagos et al. 2021), highlighting the challenges of implementing observational findings into a parameterized modeling framework.
Shallow-to-deep convective transitions are parameterized in GCMs as an interplay between buoyancy and dilution by entrainment (Wu et al. 2009; Schiro et al. 2016). Therefore, many variables can influence the parameterized transition of clouds including temperature and humidity (Wu et al. 2009). While the importance of humidity is well documented, transitions in models tend to be unresponsive to variations in humidity (Derbyshire et al. 2004; Biasutti et al. 2006; Dai 2006; Oueslati and Bellon 2013) ultimately causing misrepresentation of the diurnal cycle of convection (Folkins et al. 2014). Numerous studies have provided sufficient evidence that improved representation of entrainment rates and cold pools within the parameterizations of GCMs should lead to improved representation of the diurnal cycle (Hannah and Maloney 2011; Stratton and Stirling 2012; Ahn et al. 2019), yet improvement has been limited. Randall et al. (2016) attributes the limited success of GCMs to the independent development of the cumulus and boundary layer parameterizations, rudimentary coupling of said parameterizations, and processes being incorrectly represented (e.g., boundary layer air experiencing vertical ascent that can form a cumulus cloud in the free troposphere and precipitating downdrafts that penetrate into the boundary layer). Furthermore, GCMs commonly make use of trigger functions to initiate deep convection based on thresholds of atmospheric properties, for example relative humidity, which have known uncertainties. Some success in simulating convection at various scales has been had in the ECMWF model by incorporating humidity into the convection scheme (Bechtold et al. 2008).
Recently, to better represent the boundary layer and the statistics of shallow cumulus clouds over the southern Great Plains, an eddy-diffusivity mass-flux (EDMF) scheme was implemented in the MYNN (Mellor–Yamada–Nakanishi–Niino) planetary boundary layer parameterization (Olson et al. 2019) within the Weather Research and Forecasting (WRF) Model. This scheme is a promising addition to WRF but has only been tested in limited applications outside of the continental United States. Because the boundary layer plays such a crucial role in the transition of convection in the Amazon, the inclusion of EDMF in numerical simulations may offer new insight and improved prediction for convective transitions. We aim to better understand various factors that are important for the transition using convection-permitting modeling with two contrasting representations of shallow cumulus and directly examining the coevolution of ambient environments and shallow clouds leading up to transition to deep convection. By understanding these factors at high resolution, we can gain insight into important variables that need to be considered for GCM parameterizations. Specifically, the objectives of this study are as follows:
-
Determine the influence of the EDMF scheme on the population of tropical clouds that transition from shallow-to-deep convection during the wet season in the Amazonian rain forest. Specifically, to what degree does the frequency, areal coverage, and fraction of shallow clouds influence the likelihood of transition to deep convection?
-
Determine how various shallow cloud populations affect the ambient environment and the likelihood of convective transitions.
The paper is organized as follows. In section 2, we provide a brief description of the modeling setup and analysis techniques used to address the research objectives, followed by a discussion of the results in section 3, and concluding with a discussion of our results in the context of existing literature in section 4.
2. Method
a. Modeling setup
In this study, two individual month-long simulations are performed with the WRF Model using version 3.9.1 (Skamarock et al. 2008) in the geographical region provided in Fig. 1. The simulations begin on 1 March 2015 and end on 31 March 2015. A single domain (649 × 1349 grid points; 500 × 1200 grid points for analyses) is used that has a horizontal grid spacing of 2 km and 59 vertical levels that are stretched with height (a similar number of levels used in the High-Resolution Rapid Refresh model; Benjamin et al. 2016). The model top for both sets of simulations is 50 hPa with a 5000-m damping layer. Both simulations are initialized at 0000 UTC on 1 March using ECMWF reanalysis data (ERA-Interim; Dee et al. 2011) and lateral boundary conditions are updated every 6 h. No nudging is used in either set of simulations. Model physics parameterizations for these simulations are listed in (Table 1). The difference between the two simulations is the activation of the EDMF scheme (MYNN; Olson et al. 2019) within the MYNN planetary boundary layer (PBL) parameterization. The EDMF scheme is not activated for the control (hereinafter CTRL) simulation. More details pertaining to the EDMF scheme are discussed below. Note that a cumulus parameterization is not used for either simulation as deep convection is explicitly simulated. Additionally, the initial 48 h of the simulations are not analyzed so as to allow sufficient spinup time within the model. A comprehensive set of variables (provided in Table 2) is output every 10 min to provide higher temporal representation of the growth and decay of individual clouds and the environments in which the clouds occur.
Model physics used for simulations.
Model variables output in 10-min increments.
b. MYNN-EDMF scheme
The MYNN-EDMF scheme improves on the representation of nonlocal mixing in the convective boundary layer, specifically the boundary layer thermals (Olson et al. 2019), a current limitation in eddy-diffusivity-only parameterizations. In the MYNN-EDMF scheme, a variety of convective plumes ranging in diameters between 100 and 1000 m can be activated within the model grid column. No more than 10 plumes within a model grid column can be activated. The entrainment rate of each individual plume within a model grid column varies as a function of the plume diameter. In other words, each plume evolves independently and the plumes dynamically evolve with the life cycle of the boundary layer. This scheme is scale-adaptive with respect to the relevant scales of the meteorological conditions because the maximum plume width is a function of the boundary layer height. MYNN-EDMF is also scale-adaptive with respect to the model grid spacing because the plume width must be less than the horizontal model grid spacing. The EDMF scheme only activates in a model grid column if 1) at least one plume is produced, 2) there is a positive surface buoyancy flux, and 3) the model surface layer is superadiabatic in the lowest 50 m. If any of these conditions is not met, the mass-flux component of the scheme is not utilized in that model grid column. If all the conditions are met, the mass-flux component of the scheme is activated and the number of plumes, total updraft area, and individual plume areas are determined, followed by the integration of the plumes. The integration procedure for each of the plumes includes determining the entrainment rates, calculating variables (e.g., u, υ, thermal, moisture, and turbulent kinetic energy variables), and solving the buoyancy term and vertical velocity equation. Shallow cumulus clouds occur with the MYNN-EDMF parameterization when the plume extends beyond the lifted condensation level (LCL) and condenses. The advantage of this scheme is the better representation of shallow cumulus in gray zone model resolution (Δx = 3–10 km) where shallow cumulus is neither resolved nor parameterized explicitly, hence improving the discontinuity in the model from shallow-to-deep convection (Olson et al. 2019).
c. Cloud-type definitions
Brightness temperature and radar reflectivity are commonly used to define cloud types; however, the temporal and spatial resolution of the observing platforms influence the classification of clouds. Model simulations allow for more meteorological variables to be included in cloud classifications leading to stronger correlations between the cloud life stages to the environmental conditions. In this study, simulated clouds are defined by total cloud ice and cloud water content in the vertical column. Cloud is defined as a continual region of adjacent vertical grid points that exceed 1 × 10−5 kg kg−1 of cloud ice and/or cloud water. Cloud types in this study are divided into five categories based on the minimum and maximum heights that the total amount of cloud ice and cloud water exceeds 1 × 10−5 kg kg−1 in the vertical column (Table 3), following similar height thresholds used in Giangrande et al. (2017). A shallow cloud is defined as a cloud base and cloud top of less than 3 km. A congestus cloud has a cloud base of less than 3 km and a cloud top of less than 8 km. Congestus clouds are divided into low and high congestus on the basis of cloud-top height, where a cloud-top height of less than 5 km is low congestus and greater than 5 km but less than 8 km is high congestus (Wall et al. 2013). Last, a deep cloud has a top that is greater than 8 km and a base of less than 3 km. A cloud that has a cloud base of greater than 3 km and is precipitating is defined as stratiform rain. Nonprecipitating anvil and cirrus clouds are not considered in this method.
Cloud-type classifications.
d. Tracking convective clouds
Cloud types are determined from the cloud water and ice contents every 10 min and used to classify convective components. In PYFLEXTRKR, (the Python version of “FLEXTRKR”; Feng et al. 2018, 2019), a deep convective cloud (DCC) feature is defined as a region with high congestus and/or deep clouds for four or more adjoining grid points (16+ km2). A low cloud simply consists of low congestus or shallow clouds. After the cloud field has been classified into categories of deep core, low cloud, or clear, individual DCC objects are labeled, and the algorithm tracks each feature in time looking for deep convective clouds that overlap in area by more than 50% each time step. If a DCC feature has two consecutive time periods (20 min) of increased deep core area, the DCC is identified as a deep convection track (DCT) of interest. DCTs are further divided into subsets by lifetime, and for this study only DCTs that exist for at least 1 h are analyzed. The environment in which a DCT is first identified in is analyzed within 3 h prior to its identification (180–10 min prior), similar to the decay time scale of 3.5 h found in Adams et al. (2017), within 100 km of the identification location. This identification location will be hereinafter named the DCTI and represents the location and time of initiation. Various analysis box regions (100, 50, 25, and 12 km from DCTI) were examined to investigate the sensitivity of the background environment to the distance from the DCTI. However, because the goal of this work is to determine bulk mesoscale metrics of convective favorability, a distance of 100 km from DCTI is used for analysis (i.e., a 100 km × 100 km box region). In addition, the size of the region was deemed adequate after examining the areal coverage distributions of all DCTs that occurred over the simulation period. It is also important to note that with this method, many DCTs can occur in similar locations. However, the exact 100 km × 100 km environment will vary based on DCTI.
e. Observations
The following datasets are used in this study to compare the simulation results between 3 and 31 March 2015. The level-3 Integrated Multisatellite Retrievals (IMERG) from the Global Precipitation Measurement (GPM) mission (Hou et al. 2014; Huffman et al. 2015a,b) provides observed rainfall amounts over the Amazon basin. The microwave-estimated precipitation and microwave-calibrated infrared precipitation estimates from GPM and local rain gauges are intercalibrated, merged, and interpolated together providing a rainfall estimation (i.e., IMERG). This product has a spatial resolution of 0.1° × 0.1° (∼11 km) with a 30-min temporal resolution. The Tropical Rainfall Measurement Mission (TRMM) data product 3B42-Daily (Huffman et al. 2016) is also used to determine rainfall amounts in the study domain as a complementary dataset to GPM. This product has a temporal resolution of 3 h and a spatial resolution of 0.25° × 0.25° (∼28 km).
Cloud properties including cloud fraction are determined from the MODIS-Aqua level-3 data product (Platnick et al. 2019). This dataset is the average of the day and night retrievals of the spectral channels that are common on VIIRS and MODIS. The product has a spatial resolution of 1° × 1° spatial resolution. Observations collected from the Mobile Aerosol Observing System MAOS at the ARM Mobile Facility (AMF) (see Fig. 1) include radiosonde data [SONDEWNPN; compiled by Keeler et al. (2014)] and derived planetary boundary layer height [PBLHTSONDE1MCFARL; bulk Richardson method with 25 points variable; compiled by Riihimaki and Sivaraman (2014)], with approximately four radiosondes per day. It is important to note that there is large variability among different planetary boundary layer estimates in observations and that the Bulk Richardson method compared the best to the simulations.
The System for the Protection of Amazonia (SIPAM) S-band radar (hereinafter SIPAM) is located near 3.14°S, 59.99°W (located at MAOS in subsequent figures and text). SIPAM data are gridded to 2 km × 2 km horizontal resolution and 0.5-km vertical resolution, with a temporal resolution of 12 min (Schumacher and Funk 2018). Rainfall is estimated from SIPAM using the radar corrected reflectivity at the height level of 2.5 km within 160 km of the radar (Oliveira et al. 2016). Oliveira et al. (2016) summarize the biases of GPM-IMERG relative to SIPAM, concluding that during the wet season of 2015, GPM-IMERG drastically underestimated precipitation because isolated convective cells were not being captured by the satellite platform.
3. Results
a. Rainfall characteristics
The daily rain amounts over the subset domain (blue box in Fig. 1) for the period of 3–31 March 2015 from the simulations and GPM and TRMM are shown in Fig. 2. The domain average daily rain rate for GPM is 10.84, 11.41 mm day−1 for TRMM, 11.42 mm day−1 for EDMF, and 12.85 mm day−1 for CTRL. Both satellite-borne platforms indicate a region of increased precipitation in the northeast (0.5°N–5°S, 47.5°–51°W), near the coastline. The simulations produce much less precipitation in this region, likely due to the challenge of simulating the propagation of oceanic convection onto land associated with sea breeze (Burleyson et al. 2016) and biases in the forcing dataset over the ocean. Interestingly, an area in the southeast of the simulation domain has precipitation rates similar to the observations in the northeast, suggesting that the oceanic properties in the forcing dataset that influence convective formation may also have a location bias. The largest discrepancy in precipitation between the simulations is within 4°–8°S, 55°–57.5°W. Both simulations produce excess rainfall in this region relative to the observations. In this region, MCSs are common in March (Romatschke and Houze 2013; Rehbein et al. 2018, 2019) and it is possible that the simulations overestimate the lifetime and intensity of such systems, a common issue of the convective/stratiform rainfall ratio in WRF.
The diurnal cycle of tropical precipitation has been known to have significant errors in numerous modeling frameworks. The mean simulated rain rate by time of day and standard error are computed for the model domain and are compared with GPM (Fig. 3, along with Fig. 1 and Table 1 in the online supplemental material). Here the simulations and observations both have two separate maxima of precipitation, one overnight (0000 LT for CTRL, 0200 LT for EDMF, and 0300 LT for GPM) and a stronger maximum in the afternoon (1200 LT for CTRL, 1300 LT for EDMF, and 1500 LT for GPM). The diurnal cycle of maximum precipitation in the simulations is approximately 2 h too early relative to GPM (1300 LT vs 1500 LT). The daytime lag of 2 h between the simulations and GPM is consistent spatially throughout the domain. The difference in time of maximum precipitation during the day suggests that precipitating clouds are transitioning earlier than observations indicate. The largest difference in rain rate between the simulations and GPM for nearly all times is in the western region. While EDMF seems to show improvements in the timing and intensity of simulated precipitation overnight in the whole domain and during the day in the central domain, daytime convection in general produce excessive precipitation in both simulations.
Near MAOS (central region), observations of rainfall from GPM and SIPAM are compared with EDMF and CTRL (Fig. 2 in the online supplemental material). Here, the simulations underestimate the frequency at which low rainfall rates (0.1–10 mm h−1) occur relative to observations. The frequencies of rainfall rates between 10 and 30 mm h−1 agree well between the simulations and radar (SIPAM), while the satellite drastically underestimates the frequency at which high rainfall rates occur. In comparison with CTRL, EDMF reduces the excessive frequency of intense precipitation (10–100 mm h−1) and compares better to the SIPAM radar observations, (consistent with the reduced daily rainfall rates shown in Figs. 2 and 3. While there are clear differences qualitatively and statistically) between the simulations and observations (and the observations themselves), the objective of this study is to quantify the influence of shallow clouds on the probability of deep convection in a modeling framework, therefore the differences in precipitation amounts should not affect the overall results. Future simulations could implement more sophisticated techniques (such as Tai et al. 2021) to improve the verification statistics but is beyond the scope of this work.
b. Cloud-type frequency
Simulated cloud types need to be constrained by observations to quantify prediction biases that can be utilized to further explain biases in precipitation. Brightness temperatures Tb can be used as a proxy for cloud-top height, with the caveat that upper-level clouds may shield lower clouds in the simulations and observations. While brightness temperatures may not be the best proxy for shallow cloud identification, for a domain of this size (∼20° longitude and ∼10° latitude), homogeneous and simultaneous cloud-type classifications can really only be done using satellite-derived proxies—in this case brightness temperatures. The distribution of simulated and observed infrared brightness temperature from the NASA Global Merged IR V1 data (Janowiak and Xie 2017) at the same analysis periods are shown in Fig. 4a. Note this dataset (hereinafter MergedIR) was regridded from the native grid of 4 km to 10 km, which may reduce the frequency of the coldest cloud tops. CTRL has a greater frequency of higher brightness temperatures (290–300 K), likely due to a greater frequency of clear sky and fewer shallow clouds [cloud-top temperatures > 280 K are used to classify shallow clouds, as in Adams et al. (2017)]. EDMF and MergedIR have similar frequencies of warmer brightness temperatures (290–300 K), indicating that the frequencies of clear skies are more comparable. A comparison of cloud fraction between the simulations and MODIS-Aqua (daytime and nighttime overpasses at the same time as the simulations) shows that both simulations overestimate very low cloud fractions (clear skies) when compared with the satellite, with CTRL overestimating low cloud fractions most often (Fig. 4b). EDMF brightness temperatures between 270 and 295 K, the temperature range of simulated shallow clouds, occur at a higher frequency than observations. This implies that EDMF simulates more shallow clouds than that captured by the satellite. The simulations both slightly underestimate the frequency of brightness temperatures between 220 and 260 K. While the coldest brightness temperatures (180–210 K) are a relatively rare occurrence in both the simulations and the satellite observations, the decrease in frequencies between the simulations and satellite observations are consistently offset by 5 K, with the simulations having a higher frequency of the coldest temperatures. This suggests that the deep and high cloud frequencies for the month of March 2015 are greater in the simulations than in the satellite observations.
The frequency of simulated cloud types as defined by cloud-base height and cloud-top height (Table 3) are shown in Fig. 5. EDMF produces a significantly higher frequency of shallow clouds in the entire domain relative to CTRL (10%–30% more often in EDMF than CTRL). Along the river network, EDMF shallow clouds occur at a minimum frequency of 20%–30% of the time due to cooler surface temperature during the day and likely surface divergence due to river-breeze circulations. In the same area along the river network, CTRL shallow clouds occur at a lower frequency than EDMF, 10%–20% of time. The areal coverage of shallow clouds within the whole domain increases from 1800 to 0700 LT for EDMF and CTRL, reaching an initial maximum at 0700 LT, which is attributed to fog (Fig. 6). However, following the initial maximum, the areal coverage of shallow clouds in EDMF decreases over the next hour, and then increases for a second maximum (of similar magnitude as the first maximum) by 1100 LT, while CTRL continues to decrease from 0700 to 1800 LT. The diurnal cycle of the simulated shallow clouds in EDMF compares better to the observations during the GoAmazon field campaign (see Fig. 9 in Giangrande et al. 2017) than CTRL, particularly the peak timing at ∼1100 LT and the following sharp decrease in the afternoon as shallow clouds transition to deep convection.
While the production of shallow clouds was significantly greater in EDMF at all times of the day, the frequency of low and high congestus clouds is more comparable between the simulations (Fig. 5). In the regions of the domain where the greatest frequencies of shallow clouds are located, the highest frequencies of congestus are also found (5%–8%). The difference in temporal frequency between the simulations is nearly randomly distributed and at most ±1%–3%. Interestingly, along the eastern portion of the river network congestus clouds are more abundant in CTRL than EDMF, while EDMF congestus are more prevalent along the river network north and south of MAOS. The areal coverage of congestus clouds (labeled as Transitioning clouds in Fig. 6b) reaches a maximum in the early afternoon for both EDMF and CTRL (1200 and 1300 LT, respectively). The timing of the greatest areal coverage of congestus clouds may be comparable between the simulations, but the areal coverage is significantly different at those times (p value < 0.05), whereas overnight the simulations are in more agreement with one another.
Of interest to this study is the response of deep convective clouds to increased shallow cloud populations. It has been shown that there are significantly more shallow clouds in EDMF than CTRL, but the frequencies of congestus clouds and deep convective clouds are nearly the same between the simulations (Figs. 5g–l). The difference in frequency between EDMF and CTRL is less than 1% and the regions of preferred deep convection are similar. These results indicate that while shallow clouds were produced at greater frequencies in EDMF than in CTRL, the average deep convection frequency did not significantly increase (p value = 0.34) in EDMF. When analyzing cloud-type frequency by time of day, the areal coverage of deep convection is slightly greater in EDMF for nearly every hour (Fig. 6), with the greatest difference being at times when shallow and low congestus also have greater areal coverage. However, the difference in areal coverage between EDMF and CTRL deep clouds is not statistically significant (p value = 0.56).
In relationship to clouds transitioning between the four cloud types, the following can be inferred for daytime convection. Shallow clouds (fog) form in the early morning, reaching a maximum at 0700 LT. At 0800 LT, shallow clouds begin to transition to congestus clouds, reaching a maximum at ∼1200 LT (the secondary peak of shallow clouds in EDMF at 1100 LT leading to the large difference in low congestus clouds between EDMF and CTRL). The congestus clouds transition to deep convective clouds after 1200 LT and continue to increase in the afternoon and eventually peak overnight (Fig. 6d. As deep convection strengthens and persists throughout the night, stratiform clouds increase in coverage likely associated with nocturnal MCSs (not shown).
In summary, EDMF produces shallow clouds at significantly higher frequencies than CTRL. Satellite observations indicate CTRL produces too few shallow clouds and EDMF reduces the bias. While shallow clouds in EDMF are more abundant, the frequencies of the other convective cloud types were very similar between the two simulations and between the simulations and observations. Drastic differences in total rainfall between the simulations are contributed to the significantly higher rain rates per cloud type in CTRL when compared with EDMF, even for the higher cloud types with comparable areal coverages. Additional analysis is needed to understand the complex coupling of the boundary layer, radiation, and microphysics schemes in WRF and how this coupling affects rain production but is beyond the scope of this work at this time.
c. Average environmental conditions
The results above describe the variability in cloud properties between EDMF and CTRL, specifically the increased frequency of shallow clouds produced in EDMF. The average environment simulated by both simulations in cloud-free and rain-free conditions are compared with sounding profiles averaged from over 100 radiosondes during the GoAmazon campaign of March 2015 (Fig. 3 in the online supplemental material). In this case, both simulations produce an average environment very similar to that which is observed. The surface temperature is approximately 0.25–0.5 K warmer in the simulations than observed. Interestingly, the observed dewpoint temperature has an abrupt decrease above the surface followed by a subtle increase that neither simulation replicates. The similarities in the simulated and observed environment gives confidence in both simulations and indicates that the simulations are robust.
As was discussed in the Introduction, relative humidity is an important deterministic variable for development of tropical convection. Relative humidity in both simulations reaches a surface maximum at 0600 LT and a surface minimum at 1400 LT (Fig. 7). Between 0800 and 1800 LT, the relative humidity rapidly increases between the surface and 1 km, during the time period of increased precipitation. Between 1 and 2 km, relative humidity has local maxima for both simulations at all times, followed by a rapid decrease above 5.2 km. While the trends are similar between the simulations, generally EDMF has greater relative humidity near the surface during the later portion of the morning and into the afternoon, caused by cloud shading, cooling the surface, decreasing the surface temperatures. Parameterized shallow clouds act to increase the difference by approximately 3% at 14 and 1600 LT. These relatively small differences in the lower levels can be attributed to the variability in the evolution of the planetary boundary layer, evaporative cooling of precipitation of shallow clouds, subtle differences in surface temperatures (Fig. 4 in the online supplemental material), and cloud coverage.
Further analysis of convective metrics including most-unstable convective available potential energy (MUCAPE), most-unstable convective inhibition (MUCIN), LCL, and PBL height for the EDMF simulation within the MAOS domain and radiosondes (hereinafter “sondes”) (over 300 sondes from February to April 2015) is provided in Fig. 8. Mean simulated MUCAPE varies between 1000 and 1500 J kg−1, with the minimum occurring at 0700 LT and the maximum 2200 LT. The sonde calculated MUCAPE has a similar trend as the simulation, with the variability of the simulated MUCAPE within the range of the observations. The greatest difference between mean observed and simulated MUCAPE, approximately 500 J kg−1, occurs at 0700 LT, likely because of the higher frequency of shallow clouds (fog) produced in the simulation relative to observed clouds. Simulated mean MUCIN compares well to the sonde data as well, within 30 J kg−1 for all time periods. Simulated PBL height is underestimated when compared with the sonde at 1300 LT when shallow clouds are still more prevalent in EDMF but compares remarkably well overnight. The differences could be caused by the two methods used in estimating PBL height between the sonde and simulation. Simulated LCLs are approximately 300 m higher than the sonde in the evening but are within 100 m for the remaining time periods. In general, the simulation does reproduce a similar convective environment as observed in the MAOS region.
Overall, the increased presence of shallow clouds in EDMF did not significantly alter the average environmental characteristics when compared with CTRL. The similarity in large-scale environmental characteristics between the simulations likely explains why the frequencies of deeper cloud types are nearly identical. The simulated environments were also similar to the observations, giving confidence in the results of the simulations.
d. Deep convective environment
From the 29-day simulation period, over 2000 deep convection tracks (hereinafter DCTs) were identified in the domains of both simulations. As the frequency of high congestus and deep clouds were nearly identical between EDMF and CTRL, it was not entirely surprising that the number of DCTs that formed were nearly the same, 2492 tracks versus 2441 tracks, respectively. The regions that favored DCTs were also very similar between the simulations and on average a DCT is within 100 km of 1.81 other DCTs. The greatest number of DCTs for both simulations formed between midnight and early morning (0500 LT), while the fewest DCTs formed between 1200 and 1600 LT (Fig. 6d). For the EDMF domain divided into six regions, 231 tracks form in the northeast, 448 tracks form in the north central, 386 tracks form in the northwest, 405 tracks form in the southeast, 416 form in the south central, and 606 tracks form in the southwest. An analysis of the lifetimes and areal size of CTRL and EDMF DCTs indicate that the median lifetime is 1 h and 20 min representing the lifetime of isolated convective cells and the median area of DCTs is 400–500 km2 for both simulations (not shown). The average rainfall rates of EDMF DCTs during their lifespans are slightly greater than CTRL DCTs (11.91 vs 8.96 mm h−1; not shown).
The mesoscale environmental conditions of regions (100 km × 100 km box) centered at the location of DCT initiation (hereinafter DCTI, representing the location and time of initiation) are analyzed within 3 h (18 time samples at 10-min increments) prior to the initiation of DCTs for both simulations. The goal of this analysis is to determine how the variability of cloud frequency alters environments that favor DCT initiation. Shallow clouds occupy the largest fraction of a region that favors DCT initiation, 40% for EDMF and 32% for CTRL prior to a DCT forming. The remaining cloud types occupy less than 10% of the region that forms DCTs for both simulations. A linear polynomial function is computed to determine the rate of change that each cloud-type population experiences leading up to a DCTI. Shallow clouds in a region that forms a DCT increase at a rate of 1.8% h−1 for both simulations, low congestus clouds increase at a rate of 1.8%–2.4% h−1, and high congestus clouds increase at a rate of 1.2%–1.8% h−1 (Fig. 5 in the online supplemental material). These results indicate that the increased presence of shallow clouds in an EDMF region that will form a DCT does not influence the rate at which the cloud population transitions to deeper clouds. Instead, it may be likely that a minimum fraction of shallow clouds within the region is needed to modulate the environment and promote shallow-to-deep convective transitions, and both simulations surpass that fraction. This hypothesis will be further examined next. Analysis of the atmospheric conditions for regions that favor DCTI indicate that the initial higher frequency of shallow clouds in EDMF leads to minimal differences between the simulations, as was the case for the average environment (Figs. 3 and 4 in the online supplemental material). Temperature and dewpoint temperatures differences within the 3 h leading up to DCTI are less than 0.2 K in the boundary layer and free troposphere and relative humidity is within 1.1%. As a result of the consistent environmental conditions between the simulations, the environment that will form a DCT in EDMF will be used as a comparison with EDMF environments that do not lead to the formation of DCTs (see section 3e). EDMF is used for this analysis instead of CTRL because the cloud properties and precipitation features were relatively more comparable to observational datasets than CTRL.
DCTs are further divided into classes on the basis of the cloud coverage prior to initiation to determine the interaction between clouds and the environment (Fig. 9, Tables 4–7). The classes are binned by cloud fraction coverage (every 10%) in order to capture the rapidly evolving cloud field within a 10-min period. Note that the 10 cloud fraction classes were combined to 5 classes (every 20%, Table 4) to reduce redundancy in discussion and increase figure readability. This analysis (Fig. 9) shows that, 3 h prior to DCTI, environments that the largest proportion of DCTs form in have cloud fractions of 40%–60% (class 3). Ten minutes prior to DCTI, environments that the most DCTs will form in have cloud fraction coverages of 60%–80% (class 4; Fig. 10). Classes 2 and 4 are very common as well, with class 1 (cloud fractions < 20%) being the least common. Most environments that favor DCTI experience a significant increase in cloud coverage (Fig. 10; indicated as an increase in class from start to end) with the cloud coverage changing the least approximately 40 min of initiation. On average a DCT will develop in a region in which the cloud coverage class will increase by 1 class (+0.97), a 20%–40% increase in clouds. For the most common starting cloud coverage class (class 3), more than 30% of environments that favor DCTs will experience an increase of 40%–60% cloud coverage (Fig. 9). Shallow clouds are the majority cloud type for each cloud fraction class leading up to the initiation of convection (Fig. 11). The cloudier classes (3–5) see a reduction in shallow cloud coverage by 10%–15% with the largest reductions occurring 120–60 min prior to the initiation of convection. During the night periods, the reduction in shallow clouds is lower for the cloudier classes than during the daytime periods. This result suggests that in the Amazon, the cloud coverage tendencies play a very important role in convective initiation in addition to the initial cloud coverage.
Description of the cloud coverage used for deep convection track (DCT) classification and the number of environments sampled in each class prior to deep convection track initiation (DCTI).
Number of deep convection environments sampled in each cloud fraction class 180–10 min prior to deep convection track initiation (DCTI).
The influence of cloud coverage is apparent when the average atmospheric conditions prior to DCTI are divided into subsets by class (Fig. 12). Surface temperatures 3 h prior to DCTI have a difference of 2.5 K between class 1 and class 5 that decreases to a difference of 1.5 K 10 min prior to DCTI (Fig. 12). Boundary layer relative humidity 3 h prior to DCTI varies among the classes by more than 10% and less than 6% within 10 min of DCTI (Fig. 12). At the top of the boundary layer (∼0.8 km), the difference in temperature between the classes is less than 1 K and relative humidity converges. Specific humidity 3 h prior to DCTI is near 17.4 g kg−1 for all classes at the surface, with the specific humidity profiles converging for all classes leading up to DCTI, as the specific humidity for the cloudier classes slightly decreases (Fig. 6 in the online supplemental material). These results suggest that if the boundary layer humidity exceeds 80% and the shallow cloud population detrains enough moisture to the lower free troposphere to increase the midlevel humidity to more than 90%, convection will be favored.
e. Contrasts between non-DCT environments and DCT environments
Non-DCTs are defined as regions in which five or less DCTs occurred over the month period and have less than 0.007% deep cloud coverage (Fig. 7 in the online supplemental material). The non-DCT regions include areas with high and low frequencies of shallow clouds and lower precipitation rates. A separate analysis that explored robust DCTs, defined as deep convective objects with lifetimes greater than 3 h and maximum areas greater than 2000 km2, which can be considered MCSs (Liu and Zipser 2013), is shown in the online supplemental figures. Subsequent figures of non-DCTs showing a temporal component are computed as a running mean of 3-h periods with 10-min intervals over the full 29-day analysis period. The large-scale environments were found to be significantly different in regions that did not favor DCTs than in those that did (Fig. 13). Specifically, midlevel vertical velocities (Fig. 13e) in regions where convection will form have greater upward motion within 3 h of initiation. The greatest positive difference being for cloud coverage class 5. Frequency distributions demonstrate that DCT regions have a significantly greater probability of nonzero vertical velocities at 500 hPa when compared with non-DCT regions (Fig. 13e). Later it will be shown that moisture at all height levels is significantly greater prior to the initiation of deep convection.
The cloud population prior to DCT initiation is dominated by shallow clouds, nearly 40% (∼49% for robust DCTs; Fig. 8 in the online supplemental material) of the region, whereas in non-DCT regions shallow clouds account for less than 32% (Fig. 14). As compared with non-DCT environments, as shallow clouds persist in an environment that will initiate a DCT, rain production from clouds that are not yet deep convective objects increases leading up to initiation. In addition, dewpoint temperatures increase above the boundary layer, ambient temperatures decrease throughout much of the troposphere, particularly within the boundary layer as evaporative cooling and cold pools occur, and relative humidity has a pronounced increase in the midtroposphere (Figs. 15a,e,i). Within the boundary layer, specific humidity decreases but rapidly increases above the boundary layer (nearly up to 1 g kg−1) as shallow clouds transition to low congestus clouds. Above 3 km, temperatures remain colder on average in environments that will initiate a DCT than in non-DCT environments, likely because of downdraft strengthening in the congestus clouds. These results suggest detrainment from increasing amount of shallow and low congestus clouds gradually precondition the regions by transporting boundary layer moisture upward, moistening the lower to midtroposphere to favor subsequent DCTI. The same analysis was performed on regional subsets to examine the variation of DCTs and non-DCTs environments based on geographical area (i.e., west, central, and east), with the largest differences being found in the western region. Specifically, surface temperatures prior to a DCTI were more than 3 K colder than non-DCTs and surface relative humidity was 15% greater prior to a DCTI (not shown).
When convective environments are divided into subsets by cloud fraction class, class 5 has the largest differences in boundary layer temperatures (colder) when compared with the non-DCT regions with the differences increasing as time approaches the initiation of deep convection. Surface temperatures in cloud fraction class 1 are also colder than regions in which no deep convection forms, however, directly above the surface, the temperatures are warmer. As initiation approaches and clouds increase and deepen, the differences in surface and midlevel humidity for DCT regions of class 1, 3, and 5 and non-DCT regions increase to more than 5% at the surface and 10% in the midlevels. The temperature and moisture tendencies are greater for the lower cloud fraction classes as the increase in cloud fraction is more significant (Fig. 10b).
There is a significant variability in the cloud populations when the environments that will form DCTs are divided into subsets by time of day (middle and bottom panels of Fig. 14). The shallow cloud population during daytime hours in regions that will form DCTs and non-DCT regions are within 5% of one another (35% and 30%) and differ by more than 9% at night (43% and 34%). Although the shallow cloud population in regions that will form DCT and non-DCT regions are within 5% of one another during the day, the environments are significantly different (Fig. 9 in the online supplemental material). The cause of the significant differences in the low-level temperatures is likely the increase in precipitation in regions that will form DCTs. As more shallow and congestus clouds precipitate within 90 min of DCTI, the difference between the surface temperature in regions that form DCTs and the non-DCT environment increases to more than 1.5 K. It is likely that during this time period cold pools created by precipitating shallow and congestus clouds have a higher probability of intersecting, promoting aggregation and widening of newly triggered convective updrafts (Feng et al. 2015). Within 10 min of initiation, the differences in surface temperatures are greater than 2.5 K. The differences in relative humidity at the surface increase from 2% to over 11% 10 min prior to the initiation of DCTs. At the height of the boundary layer, the differences in relative and specific humidity are the smallest, but the difference quickly increases above it, as shallow clouds deepen and extend beyond the top of the boundary layer, transporting moisture. These results suggest that during the day, the fraction of shallow clouds within a region must exceed 35% and at least 40% of all clouds in the environment must be precipitating (middle panel of Fig. 14, black line) to favor the development of deep convection.
Overnight, the shallow cloud population nears 49% coverage in regions that will form DCTs and less than 35% in regions that do not form DCTs (bottom panel of Fig. 14). Relative to the daytime environments, night environments have a higher proportion of shallow and precipitating clouds. The environment in which DCTs form in is again colder than the non-DCT environment, but the difference in the surface temperature is less than during the day because of the lack of insolation and lower rain rates (Fig. 10 in the online supplemental material). Relative humidity at the surface is greater than relative humidity during daytime by more than 12% in environments that favor and do not favor convection. The important feature that is apparent in the nighttime environment is the effect of shallow clouds. Shallow clouds increase the relative humidity in a region that will form a DCT by more than 10% in the midlevels and the specific humidity by more than 1 g kg−1, comparable to the increases seen during the daytime.
f. Convective metrics of non-DCT environments and DCT environments
Averaged convective metrics between regions that will form a DCT (all DCTs and binned by cloud fraction classes) and non-DCT regions show very subtle (near negligible) favorability. The average and median MUCAPE have a difference of less than 100–200 J kg−1 for the two regions (DCT and non-DCT) for all periods, and a difference of less than 10 J kg−1 of MUCIN. Nelson et al. (2021) found that a subtle erosion of CIN could favor DCTI in the RELAMPAGO-CACTI field campaign region, however, in this study mean MUCIN (and MUCAPE) tendencies remain almost neutral over the 3-h period prior to DCTI. The inconsistency in the results here as compared with Nelson et al. (2021) is likely due to the distinctly different regions. More analysis would need to be performed to identify days in which the large-scale environments are similar between GoAmazon and RELAMPAGO-CACTI. That being said, the mean and median of MUCAPE and MUCIN provide little distinction in convective favorability between DCT and non-DCT regions, variability in the maximums of the metrics distributions are quite contrasting and may be more useful in determining favorability. MUCAPE values often exceed 2000 J kg−1 in DCT regions (especially during the day), which is not the case for non-DCT regions. MUCIN in DCT regions is often greater than −10 J kg−1, especially during the day, and at night is commonly greater than MUCIN in non-DCT regions (<−15 J kg−1). Robust DCTs had similar findings (Fig. 11 in the online supplemental material), with the exception of lower maximum MUCAPE values when compared with MUCAPE values from all DCTs. These results indicate that analyzing the full distribution of convective metrics may be more applicable in the Amazon to identify regions of convective favorability versus mean and median values. The mean and median height of the LCL is lower for DCT regions for both night and day periods by more than 100 m during the day (Fig. 16). Freezing level heights are also slightly lower in DCT regions, which will act to increasing latent heating and buoyancy, strengthening updraft development. These results were consistent for both the regional subsets of DCTs and robust DCTs (supplemental Fig. 11). Itterly et al. (2016) found that traditional convective metrics in the Amazonian basin cannot be used to differentiate between convective and nonconvective days, and the same could be inferred from these simulations if only the average and median are computed. In addition to traditional convective metrics discussed above, integrated buoyancy from the top of the boundary layer to the freezing level, following Ahmed and Neelin (2018), was computed for non-DCT regions and DCT regions prior to initiation (not shown). This metric also did not vary significantly for regions that favored convection to those that did not.
4. Discussion and conclusions
In this study, two convection-permitting simulations were performed to examine how different treatments of shallow cumulus clouds influence the simulated cloud populations over the Amazon. The simulations provide a unique perspective on the coevolution of cloud populations and pre-deep-convective initiation environments over the Amazonian rain forest at both high spatial and temporal resolution. Numerous observational studies have examined such environments, though over limited areas, limiting the application of the results to predictive metrics. The significant number of deep convection tracks over the month period also provides an opportunity to characterize a broad range of convective types and their environments. Shallow clouds in the simulations act to decrease boundary layer temperatures, increase relative and specific humidity, slightly increase MUCAPE, thereby promoting the growth of congestus clouds that act to precondition the midlevels. While an areal coverage of 32% shallow clouds was found to be important precursor to deep convective initiation, it was also found that the rate at which cloud coverage increases may also play an important role in deep convective favorability. In general, an increase in cloud coverage of at least 20% during a prior 3-h period favors deep convection initiation. As clouds increase, surface temperatures decrease to a range of 23.0°–24.7°C, relative humidity increases to a range of 85%–97%, and MUCAPE increases to above 1000 J kg−1. The importance in increased relative and specific humidity for convective favorability was found in past studies as well (Zhuang et al. 2017; Chakraborty et al. 2018; Giangrande et al. 2020; Tian et al. 2021). Consistent with previous results (Itterly et al. 2016), the average and median values of traditional convective metrics were found to be less robust in identifying regions of convective favorability from regions where convection is less likely to develop, thus a more rigorous suite of metrics (including thorough statistics of MUCAPE and MUCIN distributions) is necessary for the Amazon.
The eddy-diffusivity mass-flux scheme (EDMF) significantly increases the frequency and areal coverage of shallow clouds when compared with the control (CTRL) simulation, however, the frequency of higher clouds and deep convection was not significantly different, nor were the average environmental conditions. One possibility as to why the increased frequency of shallow clouds did not significantly impact the probability of deep convection is that a proportion of shallow clouds produced in EDMF are short-lived and dissipate quickly (i.e., detrainment is greater than in CTRL, updrafts are narrower than CTRL), limiting the impact on the background environment. A second possibility is that both simulations exceed a shallow cloud population threshold that increases the probability of deep convection through the modulation of the background environment. However, the spatial organization of shallow and congestus clouds prior to deep convection initiation could be different. For example, more aggregated shallow clouds could favor transition to deep convection via reduced entrainment of free tropospheric air. Future studies should focus on more detailed analysis of shallow cloud life cycles and organizations leading up to transition to deep convection.
Biases present in the spatial coverage and diurnal patterns of precipitation indicate that both simulations still misrepresent some characteristics of the cloud population in comparison with observations. All cloud types precipitate less in EDMF than their counterpart clouds in CTRL (up to 2 mm h−1 less for deep clouds), reducing the diurnal rainfall amounts in EDMF. The consistencies in the large-scale environmental conditions and the divergence of precipitation between observations and the simulations, especially at night, demonstrates the challenges that come with numerical predictions in the tropics where subtle discrepancies can cause immense variability in precipitation and vice versa (i.e., precipitation feedbacks on the environment). The discrepancies could also be related to how the planetary boundary layer, microphysics, and radiation parameterizations are coupled in the model. At this time, more investigation into the use of EDMF in multiple regions (and seasons) at various spatial resolutions is needed to gain a deeper insight to whether the representation of shallow clouds is systematically improved.
Acknowledgments.
A thank you is extended to the three anonymous reviewers whose comments and suggestions helped to elevate this paper. This study is supported in part by the U.S. Department of Energy Office of Science Biological and Environmental Research as part of the Atmospheric System Research (ASR) Program. Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of Energy under Contract DE-AC05-76RLO1830. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility located at Lawrence Berkeley National Laboratory, operated under Contract DE-AC02-05CH11231. Data were obtained from the Atmospheric Radiation Measurement (ARM) Program sponsored by the U.S. Department of Energy Office of Science, Office of Biological and Environmental Research, Climate and Environmental Sciences Division. A special thank you is given to Zhixiao Zhang for providing source code to modify the Air Force Weather Package most-unstable CAPE and CIN calculation in WRF.
Data availability statement.
Datasets, model configuration files, and scripts for this research are available online (https://doi.org/10.5281/zenodo.4695118). Note that because of the total model output and derived output file sizes exceeding 20 terrabytes, only samples of model output and derived fields are archived. Access to complete model output can be found on the NERSC HPSS Archive System at /home/b/barb672/ upon establishment of an account.
REFERENCES
Adams, D. K., H. M. J. Barbosa, and K. P. Gaitán De Los Ríos, 2017: A spatiotemporal water vapor-deep convection correlation metric derived from the Amazon Dense GNSS Meteorological Network. Mon. Wea. Rev., 145, 279–288, https://doi.org/10.1175/MWR-D-16-0140.1.
Ahmed, F., and J. D. Neelin, 2018: Reverse engineering the tropical precipitation–buoyancy relationship. J. Atmos. Sci., 75, 1587–1608, https://doi.org/10.1175/JAS-D-17-0333.1.
Ahn, M.-S., D. Kim, S. Park, and Y.-G. Ham, 2019: Do we need to parameterize mesoscale convective organization to mitigate mean-state trade off? Geophys. Res. Lett., 46, 2293–2301, https://doi.org/10.1029/2018GL080314.
Anber, U., P. Gentine, S. Wang, and A. H. Sobel, 2015: Fog and rain in the Amazon. Proc. Natl. Acad. Sci. USA, 112, 11 473–11 477, https://doi.org/10.1073/pnas.1505077112.
Barichivich, J., E. Gloor, P. Peylin, R. J. W. Brienen, J. Schöngart, J. C. Espinoza, and K. C. Pattnayak, 2018: Recent intensification of Amazon flooding extremes driven by strengthened Walker circulation. Sci. Adv., 4, eaat8785, https://doi.org/10.1126/sciadv.aat8785.
Bechtold, P., M. Köhler, T. Jung, F. Doblas-Reyes, M. Leutbecher, M. J. Rodwell, F. Vitart, and G. Balsamo, 2008: Advances in simulating atmospheric variability with the ECMWF model: From synoptic to decadal time-scales. Quart. J. Roy. Meteor. Soc., 134, 1337–1351, https://doi.org/10.1002/qj.289.
Bechtold, P., N. Semane, P. Lopez, J.-P. Chaboureau, A. Beljaars, and N. Bormann, 2014: Representing equilibrium and nonequilibrium convection in large-scale models. J. Atmos. Sci., 71, 734–753, https://doi.org/10.1175/JAS-D-13-0163.1.
Benjamin, S. G., and Coauthors, 2016: A North American hourly assimilation and model forecast cycle: The Rapid Refresh. Mon. Wea. Rev., 144, 1669–1694, https://doi.org/10.1175/MWR-D-15-0242.1.
Betts, A. K., and C. Jakob, 2002: Study of diurnal cycle of convective precipitation over Amazonia using a single column model. J. Geophys. Res., 107, 4732, https://doi.org/10.1029/2002JD002264.
Biasutti, M., A. H. Sobel, and Y. Kushnir, 2006: AGCM precipitation biases in the tropical Atlantic. J. Climate, 19, 935–958, https://doi.org/10.1175/JCLI3673.1.
Burleyson, C. D., C. N. Long, and J. M. Comstock, 2015: Quantifying diurnal cloud radiative effects by cloud type in the tropical western Pacific. J. Appl. Meteor. Climatol., 54, 1297–1312, https://doi.org/10.1175/JAMC-D-14-0288.1.
Burleyson, C. D., Z. Feng, S. M. Hagos, J. Fast, L. A. T. Machado, and S. T. Martin, 2016: Spatial variability of the background diurnal cycle of deep convection around the GoAmazon2014/5 field campaign sites. J. Appl. Meteor. Climatol., 55, 1579–1598, https://doi.org/10.1175/JAMC-D-15-0229.1.
Chakraborty, S., K. A. Schiro, R. Fu, and J. D. Neelin, 2018: On the role of aerosols, humidity, and vertical wind shear in the transition of shallow-to-deep convection at the Green Ocean Amazon 2014/5 site. Atmos. Chem. Phys., 18, 11 135–11 148, https://doi.org/10.5194/acp-18-11135-2018.
Collow, A. B. M., and M. A. Miller, 2016: The seasonal cycle of the radiation budget and cloud radiative effect in the Amazon Rain Forest of Brazil. J. Climate, 29, 7703–7722, https://doi.org/10.1175/JCLI-D-16-0089.1.
Dai, A. G., 2006: Precipitation characteristics in eighteen coupled climate models. J. Climate, 9, 4605–4630, https://doi.org/10.1175/JCLI3884.1.
Dee, D., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Derbyshire, S. H., P. Beau, P. Bechtold, J.-Y. Grandpeix, J.-M. Piriou, J.-L. Redelsperger, and P. M. M. Soares 2004: Sensitivity of moist convection to environmental humidity. Quart. J. Roy. Meteor. Soc., 130, 3055–3079, https://doi.org/10.1256/qj.03.130.
Espinoza, J. C., H. Segura, J. Ronchail, G. Drapeau, and O. Gutierrez-Cori, 2016: Evolution of wet-day and dry-day frequency in the western Amazon basin: Relationship with atmospheric circulation and impacts on vegetation. Water Resour. Res., 52, 8546–8560, https://doi.org/10.1002/2016WR019305.
Feng, Z., S. Hagos, A. K. Rowe, C. D. Burleyson, M. N. Martini, and S. P. de Szoeke, 2015: Mechanisms of convective cloud organization by cold pools over tropical warm ocean during the AMIE/DYNAMO field campaign. J. Adv. Model. Earth Syst., 7, 357–381, https://doi.org/10.1002/2014MS000384.
Feng, Z., L. R. Leung, R. A. Houze Jr., S. Hagos, J. Hardin, Q. Yang, B. Han, and J. Fan, 2018: Structure and evolution of mesoscale convective systems: Sensitivity to cloud microphysics in convection-permitting simulations over the United States. J. Adv. Model. Earth Syst., 10, 1470–1494, https://doi.org/10.1029/2018MS001305.
Feng, Z., R. A. Houze, L. R. Leung, F. Song, J. C. Hardin, J. Wang, W. I. Gustafson, and C. R. Homeyer, 2019: Spatiotemporal characteristics and large-scale environments of mesoscale convective systems east of the Rocky Mountains. J. Climate, 32, 7303–7328, https://doi.org/10.1175/JCLI-D-19-0137.1.
Fiedler, S., and Coauthors, 2020: Simulated tropical precipitation assessed across three major phases of the Coupled Model Intercomparison Project (CMIP). Mon. Wea. Rev., 148, 3653–3680, https://doi.org/10.1175/MWR-D-19-0404.1.
Folkins, I., T. Mitovski, and J. R. Pierce, 2014: A simple way to improve the diurnal cycle in convective rainfall over land in climate models. J. Geophys. Res. Atmos., 119, 2113–2130, https://doi.org/10.1002/2013JD020149.
Giangrande, S. E., and Coauthors, 2017: Cloud characteristics, thermodynamic controls and radiative impacts during the Observations and Modeling of the Green Ocean Amazon (GoAmazon2014/5) experiment. Atmos. Chem. Phys., 17, 14 519–14 541, https://doi.org/10.5194/acp-17-14519-2017.
Giangrande, S. E., D. Wang, and D. B. Mechem, 2020: Cloud regimes over the amazon basin: Perspectives from the GoAmazon2014/5 campaign. Atmos. Chem. Phys., 20, 7489–7507, https://doi.org/10.5194/acp-20-7489-2020.
Hagos, S. M., L. R. Leung, O. A. Garuba, C. Demott, B. Harrop, J. Lu, and M. Ahn, 2021: The relationship between precipitation and precipitable water in CMIP6 simulations and implications for tropical climatology and change. J. Climate, 34, 1587–1600, https://doi.org/10.1175/JCLI-D-20-0211.1.
Hannah, W. M., and E. D. Maloney, 2011: The role of moisture–convection feedbacks in simulating the Madden–Julian oscillation. J. Climate, 24, 2754–2770, https://doi.org/10.1175/2011JCLI3803.1.
Holloway, C. E., S. J. Woolnough, and G. M. S. Lister, 2013: The effects of explicit versus parameterized convection on the MJO in a large-domain high-resolution tropical case study. Part I: Characterization of large-scale organization and propagation. J. Atmos. Sci., 70, 1342–1369, https://doi.org/10.1175/JAS-D-12-0227.1.
Hou, A. Y., and Coauthors, 2014: The Global Precipitation Measurement mission. Bull. Amer. Meteor. Soc., 95, 701–722, https://doi.org/10.1175/BAMS-D-13-00164.1.
Huffman, G. J., D. T. Bolvin, D. Braithwaite, K. Hsu, R. Joyce, C. Kidd, E. J. Nelkin, and P. Xie, 2015a: NASA Global Precipitation Measurement Integrated Multi-satellitE Retrievals for GPM (IMERG). Algorithm Theoretical Basis Doc., version 4.5, 30 pp., http://pmm.nasa.gov/sites/default/files/document_files/IMERG_ATBD_V4.5.pdf.
Huffman, G. J., D. T. Bolvin, and E. J. Nelkin, 2015b: Integrated Multi-satellitE Retrievals for GPM (IMERG) technical documentation. NASA/GSFC Code 612 Tech. Doc., 48 pp., http://pmm.nasa.gov/sites/default/files/document_files/IMERG_doc.pdf.
Huffman, G. J., D. T. Bolvin, E. J. Nelkin, and R. F. Adler, 2016: TRMM (TMPA) Precipitation L3 1 day 0.25 degree × 0.25 degree V7. Goddard Earth Sciences Data and Information Services Center (GES DISC), accessed 2 March 2021, https://doi.org/10.5067/TRMM/TMPA/DAY/7.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long–lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, https://doi.org/10.1029/2008JD009944.
Itterly, K. F., P. C. Taylor, J. B. Dodson, and A. B. Tawfik, 2016: On the sensitivity of the diurnal cycle in the Amazon to convective intensity. J. Geophys. Res. Atmos., 121, 8186–8208, https://doi.org/10.1002/2016JD025039.
Janowiak, J., and P. Xie, 2017: NCEP/CPC L3 Half Hourly 4km global (60S – 60N) Merged IR V1. Greenbelt, MD, Goddard Earth Sciences Data and Information Services Center (GES DISC), accessed 18 June 2020, https://doi.org/10.5067/P4HZB9N27EKU.
Kang, S.-L., and J.-H. Ryu, 2016: Response of moist convection to multi-scale surface flux heterogeneity. Quart. J. Roy. Meteor. Soc., 142, 2180–2193, https://doi.org/10.1002/qj.2811.
Keeler, E., R. Coulter, and J. Kyrouac, 2014: Balloon-Borne Sounding System (SONDEWNPN). 2015-04-01 to 2015-05-01, ARM Mobile Facility (MAO) Manacapuru, Amazonas, Brazil; AMF1 (m1). ARM Data Center, accessed 3 March 2021, https://doi.org/10.5439/1021460.
Kim, D., A. H. Sobel, E. D. Maloney, D. M. W. Frierson, and I.-S. Kang, 2011: A systematic relationship between intraseasonal variability and mean state bias in AGCM simulations. J. Climate, 24, 5506–5520, https://doi.org/10.1175/2011JCLI4177.1.
Lee, J. M., Y. Zhang, and S. A. Klein, 2019: The effect of land surface heterogeneity and background wind on shallow cumulus clouds and the transition to deeper convection. J. Atmos. Sci., 76, 401–419, https://doi.org/10.1175/JAS-D-18-0196.1.
Li, W., R. Fu, and R. E. Dickinson, 2006: Rainfall and its seasonality over the Amazon in the 21st century as assessed by the coupled models for the IPCC AR4. J. Geophys. Res., 111, D02111, https://doi.org/10.1029/2005JD006355.
Liu, C. T., and E. Zipser, 2013: Regional variation of morphology of organized convection in the tropics and subtropics. J. Geophys. Res. Atmos., 118, 453–466, https://doi.org/10.1029/2012JD018409.
Martin, S. T., and Coauthors, 2016: Introduction: Observations and Modeling of the Green Ocean Amazon (GoAmazon2014/5). Atmos. Chem. Phys., 16, 4785–4797, https://doi.org/10.5194/acp-16-4785-2016.
Miyakawa, T., Y. N. Takayabu, T. Nasuno, H. Miura, M. Satoh, and M. W. Moncrieff, 2012: Convective momentum transport by rainbands within a Madden–Julian oscillation in a global nonhydrostatic model with explicit deep convective processes. Part I: Methodology and general results. J. Atmos. Sci., 69, 1317–1338, https://doi.org/10.1175/JAS-D-11-024.1.
Möbis, B., and B. Stevens, 2012: Factors controlling the position of the intertropical convergence zone on an aquaplanet. J. Adv. Model. Earth Syst., 4, M00A04, https://doi.org/10.1029/2012MS000199.
Nakanishi, H., and M. Niino, 2006: An improved Mellor–Yamada level-3 model: Its numerical stability and application to a regional prediction of advection fog. Bound.-Layer Meteor., 119, 397–407, https://doi.org/10.1007/s10546-005-9030-8.
Nakanishi, H., and M. Niino, 2009: Development of an improved turbulence closure model for the atmospheric boundary layer. J. Meteor. Soc. Japan, 87, 895–912, https://doi.org/10.2151/jmsj.87.895.
Nelson, T. C., J. Marquis, A. Varble, and K. Friedrich, 2021: Radiosonde observations of environments supporting deep moist convection initiation during RELAMPAGO-CACTI. Mon. Wea. Rev., 149, 289–309, https://doi.org/10.1175/MWR-D-20-0148.1.
Nobre, P., M. Malagutti, D. F. Urbano, and E. Giarolla, 2009: Amazon deforestation and climate change in a coupled model simulation. J. Climate, 22, 5686–5697, https://doi.org/10.1175/2009JCLI2757.1.
Nolan, D. S., S. N. Tulich, and J. E. Blanco, 2016: ITCZ structure as determined by parameterized versus explicit convection in aquachannel and aquapatch simulations. J. Adv. Model. Earth Syst., 8, 425–452, https://doi.org/10.1002/2015MS000560.
Oliveira, R., V. Maggioni, D. Vila, and C. Morales, 2016: Characteristics and diurnal cycle of GPM rainfall estimates over the Central Amazon region. Remote Sens., 8, 544, https://doi.org/10.3390/rs8070544.
Olson, J. B., J. S. Kenyon, W. A. Angevine, J. M. Brown, M. Pagowski, and K. Suŝelj, 2019: A description of the MYNN–EDMF scheme and the coupling to other component in WRF–ARW. NOAA Tech. Memo. OAR GSD-61, 42 pp., https://doi.org/10.25923/n9wm-be49.
Oueslati, B., and G. Bellon, 2013: Convective entrainment and large-scale organization of tropical precipitation: Sensitivity of the CNRM-CM5 hierarchy of models. J. Climate, 26, 2931–2946, https://doi.org/10.1175/JCLI-D-12-00314.1.
Peters, K., T. Crueger, C. Jakob, and B. Möbis, 2017: Improved MJO-simulation in ECHAM6.3 by coupling a Stochastic Multicloud Model to the convection scheme. J. Adv. Model. Earth Syst., 9, 193–219, https://doi.org/10.1002/2016MS000809.
Platnick, S., K. G. Meyer, P. Hubanks, R. Holz, S. A. Ackerman, and A. K. Heidinger, 2019: VIIRS Atmosphere L3 Cloud Properties Product. Version-1.1. NASA Level-1 and Atmosphere Archive and Distribution System (LAADS) Distributed Active Archive Center (DAAC), accessed 18 June 2020, https://doi.org/10.5067/VIIRS/CLDPROP_D3_VIIRS_SNPP.011.
Randall, D., C. DeMott, C. Stan, M. Khairoutdinov, J. Benedict, R. McCrary, K. Thayer-Calder, and M. Branson, 2016: Simulations of the tropical general circulation with a multiscale global model. Multiscale Convection-Coupled Systems in the Tropics: A Tribute to Dr. Michio Yanai, Meteor. Monogr., No. 56, https://doi.org/10.1175/AMSMONOGRAPHS-D-15-0016.1.
Rehbein, A., T. Ambrizzi, and C. R. Mechoso, 2018: Mesoscale convective systems over the Amazon basin. Part I: Climatological aspects. Int. J. Climatol., 38, 215–229, https://doi.org/10.1002/joc.5171.
Rehbein, A., T. Ambrizzi, C. R. Mechoso, S. A. I. Espinosa, and T. A. Myers, 2019: Mesoscale convective systems over the Amazon basin: The GoAmazon2014/5 program. Int. J. Climatol., 39, 5599–5618, https://doi.org/10.1002/joc.6173.
Richter, I., and S.-P. Xie, 2008: On the origin of equatorial Atlantic biases in coupled general circulation models. Climate Dyn., 31, 587–598, https://doi.org/10.1007/s00382-008-0364-z.
Richter, I., S.-P. Xie, A. T. Wittenberg, and Y. Masumoto, 2012: Tropical Atlantic biases and their relation to surface wind stress and terrestrial precipitation. Climate Dyn., 38, 985–1001, https://doi.org/10.1007/s00382-011-1038-9.
Rieck, M., C. Hohenegger, and C. C. van Heerwaarden