The Planck feedback (e.g. Hansen et al. 1984) is the strongest negative climate feedback and results from the Planck function (Eq. A1-1), where the blackbody radiation of an object is a function of its temperature. This is the most fundamental climate feedback, simply stating that the warmer the Earth, the more it radiates.
Following the Clausius-Clapeyron equation (EA-11), specific humidity increases approximately exponentially with temperature (at a rate of about 6 to 7 % K-1), under the assumption of constant relative humidity (e.g. Raval and Ramanathan 1989, Allan 2012, Dewey and Goldblatt6 2018). The higher concentrations of water vapor increase the greenhouse effect and accelerates the warming. This water vapor feedback (Held and Soden 2000) is the strongest warming climate feedback (e.g. Soden and Held 2006, Dessler 2013, Ingram 2013).
The lapse rate feedback is due to vertically non-uniform atmospheric heating. If the upper troposphere warms more than the lower troposphere, the lapse rate decreases and the emission to space moves closer to the emission from Earth’s surface. Thus, the greenhouse effect is less efficient, and the sign of this feedback is negative. The lapse rate feedback is robustly negative in the tropics where the temperature profile follows the moist adiabatic lapse rate, but can be 6 Dewey and Goldblatt 2018 observed decreasing OLR for surface temperatures in excess of ~298 K, whilst a continued nonlinear increase in column integrated water vapor.
positive in high latitudes in the presence of stable stratification that suppresses vertical mixing, thereby confining the warming to a thin near-surface layer (Soden and Held 2006, Pithan and Mauritsen 2013, Goosse et al. 2018).
But the lapse rate is a function of RH as well, where the non-uniform tropospheric heating can change as a result of non-uniform change in RH with warming. Because the radiative effect (IR absorption) of water vapor is roughly proportional to the logarithm of its relative humidity (Allan et al. 1999, Roca et al. 2000), the radiative impact of a change in water vapor concentration is greatest (smallest) at initially low (high) relative humidities as illustrated in Fig. 1-5 below. A change in the lapse rate can thus result from only a small increase in initial upper tropospheric relative humidity (which is significantly drier than the lower troposphere).
FIGURE 1-5: Sensitivity of the change in clear-sky OLR with tropospheric RH (dOLR/dTRH) to the mean tropospheric RH from the European Centre for Medium-Range Weather Forecasts ReAnalysis project over the time period 1979 to 1993. The figure is extracted from Allan et al. (1999).
Because these clear-sky feedbacks (Planck, lapse rate, water vapor) are not independent of each other, the effects of them are often combined assuming constant relative humidity, as advocated by Held and Shell (2012). An example of this is seen in Fig. 1-4 above, where the ensemble mean values of these feedbacks are reduced from about –3.2 W m-2 K-1, –0.5 W m-2 K-1 and +1.8 W m-2 K-1 respectively, to about –1.8, –0.1 and +0.05 W m-2 K-1 (Caldwell et al. 2016, Zelinka et al. 2020). This repartitioning limits the spread in general circulation model (GCM) projections around the ensemble mean (Ingram 2013), reduces the covariance between the lapse rate and water vapor feedbacks (Caldwell et al. 2016), and avoids dealing with feedback computations relative to unrealistic supersaturated base states (Held and Shell 2012).
Cloud Feedbacks and Radiative Effects
In cloudy scenes, two competing radiative effects stand against each other; the SW and the LW. At TOA, the cloud radiative effect (CRE) is defined as the difference between the downwelling and upwelling radiative fluxes in all-sky situations minus the difference in clear-sky situations (Matus and L’Ecuyer 2017):
The shortwave cloud radiative effect (SWCRE) is due to the high albedo of clouds (Stephens et al. 2015). The SWCRE therefore has a cooling effect on the Earth system as it prevents solar radiation from being absorbed. The SWCRE is sensitive to the size of the reflective droplets (liquid cloud droplets are smaller and closer than ice crystals to SW wavelengths), and the condensed water amount (optically thick clouds reflect more than optically thin clouds) (Stephens and Tsay 1990, Hogan et al. 2003, Henderson et al. 2012).
Clouds also absorb LW radiation. Effectively so in the 8 and 11 μm range, thereby closing the atmospheric window (Wallace and Hobbs 2006, Ahrens 2013). The longwave cloud radiative effect (LWCRE) is due to clouds’ ability to decrease the OLR, by absorbing surface emitted IR radiation at their bases and emitting OLR from their tops. Because temperature decreases with altitude in the troposphere, the temperature at the cloud top is colder than the surface temperature, so OLR emitted from clouds is less than OLR emitted from Earth’s surface. Hence, the LWCRE has a warming effect and the higher the cloud altitude, the colder the emission temperature, and the greater the difference between OLR emitted at cloud top and Earth’s surface.
The SWCRE and LWCRE always compete where clouds are present. Because warm low liquid clouds have higher albedo than cold high ice clouds, and because their cloud tops are not much colder than the surface (~10 degrees), the net CRE is dominated by the stronger SWCRE. However, for high ice clouds, the warming LWCRE is considerably more important for the net CRE.
The SW and LW cloud feedbacks are then the changes in SWCRE and LWCRE at TOA with surface warming (Vaillant de Guélis et al. 2018). In the tropics, the principal regions of interest for the SW cloud feedback are tropical subsidence regions off the west coast of continents characterized by low clouds (often referred to as the stratocumulus regions), whilst the LW cloud feedback is interesting for convective regions. Figure 1-6 below shows global mean cloud feedbacks in CFMIP1 and CFMIP2 models. It shows that the LW cloud feedback (red) is dominated by changes in cloud cover (“Amount” in Fig. 1-6), altitude and optical depth, whilst the SW cloud feedback (blue) is only sensitive to cloud cover and optical depth.
The LW cloud feedback is positive in models (e.g. Zelinka et al. 2013, 2016, 2020, Ceppi and Gregory 2017) primarily due to rising cloud altitudes of high clouds with surface warming (Fig. 1-6a,b). The positive sign is only partly offset by decreasing high cloud cover (Fig. 1-6a,b). Rising cloud altitudes and dropping cloud temperatures have also been reported in process-oriented observational studies. For example, Igel et al. (2014), observed rising anvil bases and decreasing cloud-top temperatures measured by CloudSat with warmer surface temperature. However Vaillant de Guélis et al.7 (2018) showed that the LW cloud feedback is negative in observations by CALIPSO and CERES, due to decreasing opaque cloud cover with surface temperature. The decreasing opaque cloud cover induces a negative LW cloud feedback whose magnitude is more than twice the positive LW cloud feedback induced by rising opaque cloud altitudes. This is an important result as it shifts (i) the sign of the total observed LW cloud feedback from positive to negative and (ii) the principal determining variable from opaque altitude to cover. How the anvil width changes with surface warming is diverging in previous observational studies (e.g. Lindzen et al. 2001, Rapp et al. 2005, Su et al. 2008, Igel et al. 2014), likely due to different study regions, evaluation methods and observational instruments (Hartmann and Michelsen 2002).
FIGURE 1-6: Global mean (red) LW, (blue) SW, and (black) net cloud feedbacks decomposed into amount, altitude, optical depth, and residual components for (a) all clouds, (b) non-low clouds only, and (c) low clouds only. Open symbols are for CFMIP1 models and filled symbols are for CFMIP2 models. Multimodel mean feedbacks are shown as bars. The figure is extracted from Zelinka et al. (2016).
Like the LW cloud feedback, the SW cloud feedback is also positive in models (Fig. 1-4, 1-6a). The positive sign is explained by decreasing tropical marine boundary layer cloud cover with SST, which induces a warming due to more SW absorption (1-6c) (Ceppi and Gregory 2017, Zelinka et al. 2012, 2020). The total SW cloud feedback is only partly off-set by increasing cloud optical depth with surface warming, which poses a range of negative SW feedbacks (Fig. 1-6, Zelinka et al. 2016). There seems to be a general consensus on the sign of the rate of change of marine boundary layer cloud cover and SST between model and observational studies (as observed in e.g. Zhai et al. 2015), but the exact rate of change of this decrease is not known and is the principal reason for intermodel spreads (Figs. 1-4, 1-6). The sensitivity of the marine boundary layer cloud cover to SST is identified as an emergent constraint (Focus Box 1) and a key uncertainty for climate models (Bony and Dufresne 2005, Ceppi et al. 2017).
Is there a way to decide which quantities of the current climate are relevant for climate change? Emergent constraints (Klein and Hall 2015) answer this question by examining the collective behavior that emerges unexpectedly in climate model ensembles. They are physically explainable empirical relationships between characteristics of the current climate and long-term climate prediction that emerge in collections of climate model simulations (Klein and Hall 2015).
Klein and Hall (2015) identifies the following three potential emergent constraints for cloud feedbacks: (1) low-level cloud optical depth, (2) subtropical marine low-level cloud cover, and (3) lower tropospheric mixing. These are listed in the inserted table below: A key feature for the overall decrease in marine boundary layer cloud cover with SST seems to be a greater moisture contrast between the boundary layer (BL) and the free troposphere (FT) with warmer surface temperatures that can arise either due to enhanced surface latent heat flux, or through greater surface evaporation (Kamae et al. 2016). Regardless, the moisture contrast between the BL and FT means a greater vertical gradient of moist static energy that causes enhanced vertical mixing and drying at the capping inversion, and an effectively deeper BL with horizontally smaller clouds (Brient and Bony 2013, Myers and Norris 2013, Wood and Bretherton 2016).
The study of low cloud cover with SST is motivated by recent work in e.g. Klein et al. (2017) and Zelinka et al.8 (2020). They found that the sensitivity of low cloud cover to SST was small compared to other cloud controlling factors (estimated inversion strength, RH, pressure velocity), but that the change in SST with unit global warming was about 10 times greater than the change in any of the other cloud controlling factors, making the predicted change in low cloud cover per unit global warming most sensitive to SST.
On the global scale, the Earth system can be thought of as a closed system, but on shorter timescales, local and regional states are dominated by e.g. seasonal or diurnal scale processes. Knowing that different components of the Earth system adjust to a climate forcing on different timescales, and that different atmospheric processes and mechanisms are important on different timescales, how can we assume that the individual feedback terms are timescale-invariant? Klein et al. (2017) assumed sensitivities of the low cloud cover to various cloud controlling factors to be constant in time, but ended their work with a thorough discussion on the validity of the assumption of timescale invariant feedback terms.
Principal Hypotheses About the Tropical Atmospheric Water Cycle’s Response to Surface Warming
The previous section discussed changes in clear-sky and cloudy-sky feedbacks separately. This section discusses the collective response of the atmospheric water cycle to SST over the full tropical belt, including both clear-sky and cloud feedbacks. Here I also explain the most discussed hypotheses concerning the tropical atmospheric water cycle’s response to surface warming, with special emphasis on the iris and Fixed-Anvil Temperature (FAT) hypotheses that are investigated in this work.
The hydrological cycle is expected to intensify, with increasing surface evaporation following the warming, making more atmospheric water available for precipitation. Globally, already wet regions (precipitation rate > evaporation rate) grow wetter, whilst dry regions (evaporation > precipitation) become even drier in a nonuniform response of the global water cycle referred to as “wet-get-wetter” and “dry-get-drier” trend (IPCC AR5, 2013). It is illustrated in Fig. 1-7 by observed trends in surface specific humidity over oceans (a) and predicted precipitation changes in the RCP4.5 scenario (c-f).
Hypotheses regarding the responses of the tropical region to surface warming discuss an intensified hydrological cycle, characterized by a narrowing of the ascending branch of the Hadley circulation with stronger updrafts in moist convective regions and horizontally broader clear-sky subsidence regions, as illustrated in the conceptual schematic from Su et al. (2017) based on ensemble means of climate model simulations under global warming (Fig. 1-8).
8 Zelinka et al. (2020) compared results of climate sensitivity and radiative feedbacks in CMIP5 and CMIP6 models.
FIGURE 1-7: a) Trends in column-integrated water vapor observed by the Special Sensor Imager over ocean surfaces over the period 1988 to 2012. b) Global annual average column-integrated water vapor over ocean surfaces relative to the 1988 to 2007 average. c-f) Seasonal multi-model ensemble mean of 42 CMIP5 models for projected changes in precipitation (%) over the period 2016 to 2035 relative to 1986 to 2005 under RCP4.5. The figures are extracted from IPCC AR5 (2013).
FIGURE 1-8: Schematic of the tightening of the Hadley circulation in a warmer climate. Light colors represent the current climate and dark colors a warmer climate. The figure is extracted from Su et al. (2017).
The intensified tropical hydrological cycle is fueled by greater moisture supply from enhanced surface evaporation and horizontal moisture convergence that enable greater latent heat release in the updraft. The stronger updrafts cause convective anvils to rise, but also greater precipitation efficiency, leaving fewer hydrometeors to build up the anvil cloud, effectively shrinking the horizontal extent of the anvil and allowing more OLR to be emitted to space in the clear skies outside of the convective clouds (Su et al. 2017).
This reasoning is the basic theory for the iris hypothesis (first postulated by Lindzen9 et al. 2001, and subsequently revised in e.g. Fu et al. 2002, Hartmann and Michelsen 2002, Lin et al. 2002, 2004, 2006, Su et al. 2008, Mauritsen and Stevens10 2015.) in which precipitation efficiency increases with SST, leading to smaller convective anvils and effectively more OLR as fewer hydrometeors are left to build these (Fig. 1-8). Radiatively, the increase in OLR is balanced by an increase in latent heat release due to greater convective precipitation efficiency.
To a first approximation, the iris hypothesis compares the extent of the tropical moist and dry regions. It received its name from the adaptive behavior of the system to open (close) to allow for more (less) OLR to be emitted to space with warmer (colder) surface temperature, in much the same way as the eye’s iris opens/closes to light changes (Lindzen et al. 2001). This thinking is in line with Pierrehumbert (1995) who suggested a regulation by the area of large-scale subsidence regions (radiator fins), in which the clear-sky GHE is determined by the water vapor feedback. Such radiator fins (illustrated in Fig. 1-8) are consistent with broader clear-sky subsidence regions with surface warming.
Within a stronger updraft, it is possible that SW reflection increases, which could explain the thermostat hypothesis, proposed by Ramanathan and Collins11 (1991), where thicker cirrus anvils act like a thermostat regulating the amount of SW radiation that can be absorbed by the surface. Observational support for this hypothesis was found in e.g. Lebsock et al.12 (2010) and Igel et al.13 (2014).
In the Fixed-Anvil Temperature hypothesis (FAT), first presented by Hartmann and Larson (2002) (and subsequently assessed in a number of papers, e.g. Zelinka and Hartmann 2010, 2011, Seeley et al. 2019), cloud altitudes rise in response to surface temperature increase. They rise in such a way that they stay at the same temperatures as before the warming, which means that the temperature at the anvil detrainment level remains constant, making the anvil emission temperatures and the OLR independent of surface temperature. The altitude of anvil detrainment (convective outflow) occurs at the altitude where radiative cooling decreases most rapidly with height, which is where the saturation vapor pressure of water vapor becomes so low that water vapor no longer has a significant effect on the emitted radiation.
9 Lindzen et al. (2001) used cloud observations from the Japanese Geostationary Meteorological Satellite-5 over the western Pacific (30°S to 30°N and 130°E to 170°W).
10 Mauritsen and Stevens (2015) found that inclusion of an iris effect in the ECHAM6 GCM brought simulated equilibrium climate sensitivity closer to observed values.
11 Ramanathan and Collins (1991) used radiation data from the Earth Radiation Budget Experiment and sea surface temperature from weather satellites and ships over the time period 1985 to 1989, including the El Niño event in 1987. They found that cirrus anvils were thicker and their SW reflectivity anomaly higher over the warmest sea surface temperatures during the 1987 El Niño event.
12 Lebsock et al. (2010) found positive correlations between precipitation rate (AMSR-E) and high-cloud reflectivity (CERES-EBAF) in the tropics.
13 Igel et al. (2014) observed physically thicker cloud anvils and increasing ice water path in convective systems derived from CloudSat observations.
The super-greenhouse effect (SGE, e.g. Hallberg and Inamdar 1993, Raval and Ramanathan 1989) refers to those tropical locations where the Planck function fails to stabilize the climate (Stephens et al. 2016), which is effectively where OLR decreases with SST. Following the equations outlined in Valero et al. (1997), the greenhouse effect G can be expressed as:
Where ε is the emissivity of the Earth’s surface and σ the Stefan-Boltzmann constant. The SGE is thus present when:
Dewey and Goldblatt (2018) find that the decrease in OLR with SST occurs for SSTs > 298 K (Fig. Ba). Column moistening by deep convection in cloudy regions and advection of moisture in clear-sky regions, increases the proportion of OLR that originates from the cold high troposphere rather than the warm surface.
FIGURE B: (a) Observational OLR dependence on SST, and OLR output for various humidity values. The red dashed line is the mean, and the black circles correspond to the panels in (b). (b) Spectra of thermal emission altitude. The bottom four panels are the surface temperatures and RH where the observational OLR intersects model curves, and the top two panels are at higher surface temperatures, both with 100 % RH. The background color indicates the atmospheric temperature, and the white line is the altitude at which optical depth is unity. The figure is extracted from Dewey and Goldblatt (2018).
Bony et al.14 (2016) revisited both the iris and the FAT hypotheses using three GCMs and proposed the idea of a “stability iris mechanism” that links the two, where the detrainment is controlled by stability (the ratio of the radiative cooling to vertical velocity). Tropical stability (σ) is balanced by radiative cooling (Q) and radiatively driven subsidence rate (ω): σ = Q / ω (Allan 2012). Greater atmospheric moisture associated with warmer temperatures moves the environmental lapse rate closer to the moist adiabatic lapse rate, thus increasing the atmospheric stability and diminishing the radiative cooling and the subsidence rate needed to balance it, resulting in less convective outflow.
Present Day Limitations with Observations and Models
Issues that deal with circulation, cloud and climate relations require analyses of observations to test for theories and hypotheses. These must be able to accurately sample the key water cycle variables with high temporal and spatial resolution over the full tropical belt, and must cover a long enough time period to account for natural and interannual variability – with the dominant tropical interannual variability being the El Niño-Southern Oscillation.
That said, observations are snapshots of reality that require analysis and theories to explain what they show. The reason is that all variables change together in observations, and they can therefore not be used to study causality. Nor can they be used to study isolated physical processes or mechanisms that require all other aspects of the atmosphere to be held fixed.
For a long time, the observational coverage was limited to land stations or ship tracks at sea. For that reason the observational coverage is much denser and continuous records extend much further back in time over land than over oceans. Fig. 1-9 shows the number of observations made in oceanic 1° × 1° grid boxes following ship tracks over the first decade in the 19th, 20th and 21st centuries. The top figure shows that the tropical Pacific Ocean was completely unobserved during the 19th and early 20th century. It took until the dawn of the satellite era and the launch of the first weather satellites Vanguard 2, Explorer-6, Explorer-7 and TIROS 1 (Television InfraRed Observational Satellite) in the late 1950s and early 1960s (Capderou 2014) for atmospheric observations to be possible over ocean surfaces outside of ship tracks and unpopulated land surfaces. These satellite observations provided the first global views of the atmospheric water cycle and TOA radiative fluxes. Fig. 1-10 below shows the first image of the Earth taken by a weather satellite. This image was taken by the TIROS 1 satellite, whose objective was to test experimental television techniques designed to develop a worldwide meteorological satellite information system (https://airandspace.si.edu/collection-objects/tiros-meteorological-satellite/nasm_A19650289000, accessed September 8, 2020). The satellite carried two television cameras, one of low and a second of high resolution.
14 Bony et al. (2016) conducted their simulations with the GCMs from MPI, IPSL and NCAR.
FIGURE 1-9: Data counts per 1° × 1° latitude, longitude of surface pressure observations by ships in the International Comprehensive Ocean-Atmosphere Data Set, used by the National Oceanic and Atmospheric Administration and European Centre for Medium-Range Weather Forecasts 20th century reanalyses, for three selected time periods: (A) 1800–1810, (B) 1900–1910, and (C) 2000–2010. The figure is extracted from Smith et al. (2019).
Back then, satellites were equipped with passive sensors that could not observe the detailed vertical structure of the atmosphere, which is perhaps why previous work only studied one vertical range (e.g. Ross et al. 2002, Gettleman et al. 2006, Läderch and Raible 2013). Detailed observations could be made in field campaigns with radiosondes or aircraft measurements, but as these only observed a local restricted region, they under sampled the tropics and missed the large-scale context. It was not until the turn of the millennium that active sensors (e.g. lidars and radars) were mounted onboard satellites, providing means of detailed observations of global coverage. By now, these instruments have been in use long enough to account for natural variability and detect significant trends.
In contrast to observations, climate models are very useful for tracking processes as well as for projections of historical and future climate scenarios. They are rooted to their best ability in physical laws, but all equations cannot be solved analytically, which is why the models must rely on numerical estimations and parameterizations.
However, not all processes are described equally in all models, nor do they all make the same assumptions. This leads to significant intermodel spreads of key results with a wide range of projections (as illustrated in Figs. 1-4 and 1-6) when the models do not converge towards a single behavior (e.g. Andrews et al. 2012, Vial et al. 2013, Po-Chedley et al. 2018, Zelinka et al. 2016, 2020). This is particularly concerning for feedback mechanisms and climate sensitivity that must be computed by numerical models.
Still, as computer power increases and more data storage is made available, climate model performance is constantly advancing, with more accurate predictions of higher resolutions and longer records.
Principal Science Question
Given that the primary feedback uncertainties are derived to uncertainties in how moisture and clouds vary with surface warming, this work aims to improve on our understanding of the relationships between the RH profile and cloud characteristics with SST.
Focus on the Instantaneous Timescale and Large-Scale Regimes
Short and local scales are where we come the closest to observing covariations of variables of interest. Instantaneous observations of one or two of the key water cycle variables (moisture, clouds, precipitation) have been made in the past, but the study in Chapter 3 observes instantaneous covariations of the RH profile and cloud cover with SST over the full tropical belt and under the influence of the large-scale circulation. The principal questions asked are:
• Do RH and cloud cover vary differently with SST under large-scale ascent and descent?
• Do RH and cloud cover vary differently with SST in the presence and absence of precipitation?
• Do low liquid and high ice cloud covers vary differently with SST?
• Does the RH profile vary differently with SST in the vicinity of low liquid clouds and high ice clouds?
Focus on Different Temporal and Spatial Scales
Chapter 3 studies covariations of some of the key components of the tropical atmospheric water cycle on the instantaneous grid box scale. It is thus restricted to one temporal scale and one spatial scale. In contrast, the study in Chapter 4 assesses the dependence of the water cycle variables’ covariations and responses to SST warming on the choice of temporal and spatial scale by observing them on four different timescales (daily, monthly, seasonal, annual), as well as in both spatially global values and more process-oriented grid box values.
Spatial and temporal scales are not independent of each other, as the larger the atmospheric system/phenomenon, the longer its expected lifespan or the adjustment time to changes for the components and variables it interacts with (e.g. Orlanski 1975, Steyn et al. 1981). Process-oriented studies are best executed on small and short spatial and temporal scales, whilst climatological long-term trends are most relevant on the global scale, as climate change is constrained by the global energy budget and the net radiative flux at TOA. Time-varying climate predictions have previously been made with models (e.g. Armour et al. 2013, Gregory and Andrews 2016), but I am unaware of observational studies that deal with the sensitivity of the atmospheric water cycle to surface temperature across timescales.
The principal questions asked in this study are:
• How do RH and cloud cover, altitude, temperature vary with SST on different timescales? This question targets the principal uncertain terms in the feedback equation.
• How does the low cloud cover vary with SST on different timescales? This question is central to the SW cloud feedback.
• How do high ice cloud characteristics (cover, altitude, temperature) vary with SST over the warmest waters? This question is central to the LW cloud feedback and relates to both the iris and FAT hypotheses.
• How do cloud characteristics covary and how do they vary with the RH profile? These are the second order uncertain terms in the feedback equation.
Secondary questions ask if the signs and magnitudes of these rates of changes are robust across timescales? These are novel questions from an observational perspective.
Answers to the science questions listed above require a comprehensive observational dataset. In Chapter 2, I build a dataset of observational diagnostics on a regular grid of 1° × 1° spatial resolution that can be directly relatable to climate studies. This resolution is a common standard today, which roughly translates into 100 km × 100 km in the tropics.
To answer these questions, I have built a synergistic dataset of collocated instantaneous observations of the tropical atmospheric water cycle. Chapter 2 presents these observational products and the instrument payloads from which they are retrieved. In Chapter 3, I assess how the RH profile and cloud cover varies with SST on the instantaneous timescale under the influence of the tropical large-scale circulation (i.e. the questions asked in Sect. 1.5.1). In Chapter 4, I assess the dependence of the water cycle variables’ covariations and responses to surface warming on the choice of temporal and spatial scale by observing them on four different timescales (daily, monthly, seasonal, annual), as well as in both spatially global values and more process-oriented grid box values (the questions asked in Sect. 1.5.2). I end this manuscript by a concluding chapter, where I have summarized the answers to the principal science questions and put the results in perspective.
Table of contents :
1.1 Complexity Facets of the Atmospheric Water Cycle
1.2 Radiation and the Water Cycle
1.3 Principal Hypotheses About the Tropical Atmospheric Water Cycle’s Response to Surface Warming
1.4 Present Day Limitations with Observations and Models
1.5 Principal Science Questions
2 The Synergistic Dataset
2.1 Introduction to the Dataset
2.2 Relative Humidity from SAPHIR
2.3 Cloud Characteristics from GOCCP
2.4 Near-Surface Precipitation from the CloudSat 2C-PRECIP-COLUMN Product
2.5 Surface Temperature and Vertical Pressure Velocity from ERA5
2.6 Satellite Collocations
3 Variations on the Instantaneous Timescale Under Large-Scale Circulation Constraints Summary of the Paper and Its Main Outcomes
3.4 Analyses of the Tropical Atmospheric Water Cycle’s Variation with SST in Different Regimes
4 Covariations on Different Temporal and Spatial Scales
4.3 Analysis of Observed Water Cycle Variable Covariations on Different Timescales
4.4 Discussion of These Results
4.5 Summary and Conclusions