The ability of global climate models (GCMs) to simulate climatological precipitation and other features of the hydrological cycle accurately is acceptable by some metrics, especially at large scales. Regionally, however, there can be substantial discrepancy in a multi-model ensemble, both in the annual or seasonal historical precipitation climatology as well as in end-of-century changes. Characterizing this intermodel spread and identifying leading uncertainty patterns and underlying physical pathways is important in constraining climatological biases and projections of future change. This dissertation looks at three aspects of precipitation uncertainty in ensembles.

First, El Nino-Southern Oscillation (ENSO) teleconnections are analyzed in an atmosphere-only ensemble to gauge the ability of atmospheric components of GCMs to reproduce ENSO precipitation teleconnections. This serves as a test for how well models simulate the atmospheric response to sea surface temperature forcing in the immediate ENSO vicinity, as well as how accurately they reproduce the large-scale tropical-to-midlatitude dynamics leading to teleconnected precipitation. While individual models have difficulty in simulating the exact spatial pattern of teleconnections, they demonstrate skill in regional amplitude measures and sign agreement of the precipitation teleconnections at the grid point level, which lends value to the use of such measures in global warming projections.

Next, objective spatial analysis techniques are applied to a fully-coupled GCM ensemble in order to visualize patterns of uncertainty in end-of-century precipitation changes and in the historical climatology. Global patterns are considered first, with the tropics exerting a clear dominance in intermodel spread, mainly within zones of deep convection or along convective margins. Regional domains are considered second, with a focus on the wintertime midlatitude Pacific storm track. A key region of end-of-century precipitation change uncertainty is identified at the terminus of the storm track, and large-scale circulation processes related to model differences in upper-level jet increases are found to play a role. These results help pinpoint a source of intermodel spread in projected precipitation changes along the North American west coast, especially for the Southern California region.

Last, an existing perturbed physics ensemble is examined in order to understand the parameter sensitivity of climatological precipitation and other fields. This ensemble consists of integrations in which four parameters in the deep convection scheme were systematically varied. Models of parameter dependence are constructed for precipitation, and this process--termed metamodeling--is a computationally cheap alternative to brute-force sampling of parameter space in the GCM. A quadratic metamodel performs generally well but fails to capture sensitive regions of high nonlinearity for certain parameter ranges. A second metamodel is constructed by combining an approach from the engineering literature with the spatial uncertainty patterns used above, and it proves adept at capturing sensitive regions where its quadratic counterpart fails. Finally, when more than one field is optimized simultaneously, it is often the case that a set of parameter values that optimizes one field can degrade performance in another. Concepts from multiobjective optimization are used to quantify these tradeoffs.

Teleconnections are remote impacts from one location to another, which are transmitted through planetary-scale wave motions, e.g., Rossby waves. Teleconnections can be forced by tropical heat sources, for example, El Nino/Southern Oscillation (ENSO), and result in many societally important impacts in remote ocean basins and continents. It is thus important to understand the complex pathway and wave dynamics in teleconnections. This dissertation discusses the role of baroclinic-barotropic interaction in three different aspects of atmospheric teleconnections.

The first major question addressed is what mechanisms control the interhemispheric teleconnections from tropical heat sources. These are investigated using an intermediate complexity model [a Quasi-equilibrium Tropical Circulation Model (QTCM)] and a simple linear two-level model with dry dynamics. Illustrating the interhemispheric teleconnection process with an Atlantic Warm Pool principal case, the heat source directly excites a baroclinic response that spreads across the equator. Three processes involving baroclinic-barotropic interactions---shear advection, surface drag, and vertical advection---then force a cross-equatorial barotropic Rossby wave response. An analysis of these processes in QTCM simulations indicates that: (1) shear advection has a pattern that roughly coincides with the baroclinic signal in the tropics and subtropics; (2) surface drag has large amplitude and spatial extent, and can be very effective in forcing barotropic motions around the globe; (3) vertical advection has a significant contribution locally and remotely where large vertical motions and vertical shear occur. A simple model is modified to perform experiments in which each of these three mechanisms may be included or omitted. By adding surface drag and vertical advection, and comparing each to shear advection, the effects of the three mechanisms on the generation and propagation of the barotropic Rossby waves are shown to be qualitatively similar to the results in QTCM. It is also found that the moist processes included in the QTCM can feed back on the teleconnection process and alter the teleconnection pattern by enlarging the prescribed tropical heating in both intensity and geographical extent, and by inducing remote precipitation anomalies by interaction with the basic state.

The second major question addressed concerns the dynamics that maintain the sea level pressure (SLP) anomalies in the western Pacific that are an integral part of the classic ENSO pattern. Traditional studies of the ENSO response in the tropics assume a single deep baroclinic mode associated with the tropospheric temperature anomalies. However, the SLP anomalies in the western Pacific are spatially separated from the baroclinic signal in the NCEP-NCAR reanalysis, CMIP5 models, and QTCM. Separation of ENSO SLP anomalies in the tropical Pacific into baroclinic and barotropic components indicates that the barotropic component contributes throughout the tropics and constitutes the primary contribution in the western Pacific. To demonstrate the roles of baroclinic and barotropic modes in ENSO teleconnections within the tropics, a series of QTCM experiments is performed, where anomalies in the interactions between baroclinic and barotropic modes are suppressed over increasingly wider latitudinal bands in the tropical Pacific. If this suppression is done in the 15N-15S band, the pressure signals in the western Pacific are only partly removed, whereas if it is done in the 30N-30S band, the anomalies in the western Pacific are almost entirely removed. This suggests the following pathway: interactions with sea surface temperature (SST) anomalies create the baroclinic response in the central and eastern Pacific, but baroclinic-barotropic interactions, arising substantially in the subtropical Pacific, generate a barotropic response that yields the SLP anomalies in the western Pacific.

The third major question addressed is by what pathways the baroclinic response forced by El Nino SST anomalies gives rise to the barotropic component of ENSO teleconnections, focusing especially on the subtropical Pacific. This is examined by diagnosing from NCEP-NCAR reanalysis the interaction terms in a vertical mode decomposition that yields an effective Rossby wave source, and by analyzing the response of QTCM to the NCEP-NCAR effective Rossby wave source. Among the three barotropic Rossby wave source contributions (shear advection, vertical advection and surface drag), the leading contribution is from shear advection, and more specifically, the mean baroclinic zonal wind advecting the anomalous baroclinic zonal wind. Vertical advection is the smallest term, while surface drag tends to cancel and reinforce the shear advection in different regions through damping on baroclinic mode, which spins up a barotropic response. There are also non-trivial impacts of transients in the barotropic wind response to ENSO. Both tropical and subtropical baroclinic vorticity advection contribute to the barotropic component of the Pacific subtropical jet near coast of North America, where the resulting barotropic wind contribution approximately doubles the zonal jet anomaly at upper levels, relative to the baroclinic anomalies alone. In this view, the effective Rossby wave source in the subtropics simply arises from the basic-state baroclinic flow acting on the well-known baroclinic ENSO flow pattern that spreads from the deep tropics into the subtropics over a scale of equatorial radius of deformation. This is inseparably connected to the leading deep tropical Rossby wave source that arises from eastern Pacific climatological baroclinic winds advecting the tropical portion of the same ENSO flow pattern.

Conditional instability and the buoyancy of convective plumes drive moist convection in the tropics but have a variety of representations in model convective schemes. Generally, instability—generated on the large-scale—is removed by local convection and precipitation. This process is reflected in statistical relationships between precipitation and bulk column measures of relevant variables, such as humidity and CAPE. A first-order relationship that can be leveraged for climate model diagnostics is the precipitation ‘pickup’ in which precipitation conditionally averaged on humidity undergoes a sharp, sudden increase. The strength and location of the pickup—referred to as the critical point—is used to assess model sensitivity to lower-free tropospheric moisture. Observational data from the Atmospheric Radiation Measurement (ARM) permanent field observational sites are augmented with satellite observations of precipitation and temperature as an observational baseline. Coupled Model Intercomparison Project Phase 6 (CMIP6) models are assessed. Models perform relatively well in regards to the critical point of the pickup when model temperature biases are accounted for. The conditional average precipitation is decomposed into the product of the probability of raining and mean precipitation during raining times (conditional intensity), showing that models which do perform well do so through compensating biases—model conditional intensity that is too low at a given humidity is compensated in part by excessive probability of precipitation.

Biases uncovered are further explored with the use of an entraining plume model. The estimation of a bulk plume entrainment rate which we term the “pseudo-entrainment,” captures the trade-off between lager temperature lapse rates and smaller subsaturation across the models. This pseudo-entrainment diagnostic is also a reasonable indicator of the critical value of integrated buoyancy for precipitation onset.

Using a combination of reanalysis and observational products (ARMBE, COSMIC2, and ERA5), the precipitation onset is expressed as a weighting between CAPE-like and subsaturation-like factors, here including the condensate loading and direct water vapor effect on buoyancy. The evolution of convection and its feedback to the environment are shown to have a considerable impact in making this relationship more precise. Virtual and condensate effects have a significant effect on plume buoyancy; however, the weighting between instability and subsaturation proves robust when these effects on plume buoyancy are introduced.

Moist convection drives much of the circulation and precipitation in Earth's atmosphere, especially in the tropics, and thus has far-reaching consequences for human activities. The parameterization of moist convection in general circulation models (GCMs) remains a major source of error in these models, highlighting the importance of understanding the physical processes by which deep convection interacts with larger scales. A physical pathway contributing to the uncertainty of parameterizations concerns how environment air enters convective updrafts, i.e., entrainment. This dissertation addresses several aspects of the physics of convection, including causal pathways by which the environmental water vapor impacts convection and the implications of this for model parameterizations, related process-oriented diagnostics for comparing convective processes in GCMs to observations, and theoretical foundations for a different view of the entrainment process.

First, it has been noted that as the atmospheric column water vapor (CWV) exceeds a critical threshold, precipitation sharply picks up. This ``pickup'' has been demonstrated to be consistent with the lower-free-tropospheric humidity impacting updraft buoyancy---thus convection---through entrainment. By performing a set of parameter perturbation experiments in the NCAR Community Earth System Model (CESM), we establish the causal relationship of the observed transition to deep convection---only when the convective scheme includes substantial entrainment can the model reproduce the pickup. The change in the precipitation--CWV relation also leads to a moist troposphere compared with when there is no entrainment. The reevaporation of convective precipitation is found to have only minor effects on the pickup.

Second, from observations and reanalyses, we compile and expand the set of statistics characterizing the transition to deep convection at fast timescales---termed ``convective transition statistics.'' Given that the spatial autocorrelation scales of tropospheric temperature and humidity are greater than that of precipitation, the precipitation--CWV relationship is robust to spatial resolution up to ~1� and time-averaging up to ~6 hours. The critical CWV at which precipitation starts to pick up increases with bulk tropospheric temperature, while the corresponding critical column relative humidity (CRH) decreases. This is consistent with prior entraining plume calculations. The CWV value relative to critical appears to be an effective predictor of conditional instability (hence precipitation) with only minor geographic variations in the tropics. The distribution of precipitation intensity drops rapidly for low CWV, and develops into a robust long-tail distribution for CWV around and above critical. The robustness of convective transition statistics---especially to the spatial-temporal resolution---suggests that these are suitable for model diagnostic purposes.

Third, we compared the convective transition statistics calculated using high-frequency (1-6 hourly) output from a set of GCMs to the observed. Comparing statistics among models that primarily differ in representations of moist convection suggests that convective transition statistics can substantially distinguish differences in convective representation and its interaction with the large-scale flow, while models that differ only in spatial–temporal resolution, microphysics, or ocean–atmosphere coupling result in similar statistics. Most of the models simulate some version of the observed sharp pickup of precipitation as well as that convective onset occurs at higher CWV but at lower column RH as temperature increases. However, departures from observations in various aspects of the precipitation--CWV relationship are also noted in many models.

Lastly, a puzzle regarding entrainment profiles in the vertical is addressed. Observations and large-eddy simulations has pointed to a ``deep-inflow'' updraft mass flux structure, in which mass enters the updraft through a deep layer in the lower troposphere. Looking for a simple explanation for the observed deep inflow, we investigate the nonlocal response of vertical velocity field to buoyancy under the anelastic framework. We find that the vertical structure of response is determined by the horizontal length scales contributing to the buoyancy structure. For a wide range of conditions relevant for isolated cumulonimbus and organized systems, the nonlocal dynamics entailing interaction between the buoyant layer and the surface results in the deep inflow. Furthermore, the largest, most heavily-precipitating contributions to convection are suggested here to be a simplifying factor for their representations in convective parameterizations.

The total amount of precipitation integrated across a precipitation cluster (contiguous precipitating grid cells exceeding a minimum rain rate) is a useful measure of the aggregate size of the disturbance, expressed as the rate of water mass lost or latent heat released, i.e. the power of the disturbance. Probability distributions of cluster power are examined during boreal summer (May-September) and winter (January-March) using satellite-retrieved rain rates from the Tropical Rainfall Measuring Mission (TRMM) 3B42 and Special Sensor Microwave Imager and Sounder (SSM/I and SSMIS) programs, model output from the High Resolution Atmospheric Model (HIRAM, roughly 0.25- 0.50 resolution), seven 1- 2� resolution members of the Coupled Model Intercomparison Project Phase 5 (CMIP5) experiment, and National Center for Atmospheric Research Large Ensemble (NCAR LENS). Spatial distributions of precipitation-weighted centroids are also investigated in observations (TRMM-3B42) and climate models during winter as a metric for changes in mid-latitude storm tracks. Observed probability distributions for both seasons are scale-free from the smallest clusters up to a cutoff scale at high cluster power, after which the probability density drops rapidly. When low rain rates are excluded by choosing a minimum rain rate threshold in defining clusters, the models accurately reproduce observed cluster power statistics and winter storm tracks. Changes in behavior in the tail of the distribution, above the cutoff, are important for impacts since these quantify the frequency of the most powerful storms. End-of-century cluster power distributions and storm track locations are investigated in these models under a “business as usual” global warming scenario. The probability of high cluster power events increases by end-of-century across all models, by up to an order of magnitude for the highest-power events for which statistics can be computed. For the three models in the suite with continuous time series of high resolution output, there is substantial variability on when these probability increases for the most powerful precipitation clusters become detectable, ranging from detectable within the observational period to statistically significant trends emerging only after 2050. A similar analysis of National Centers for Environmental Prediction (NCEP) Reanalysis 2 and SSM/I-SSMIS rain rate retrievals in the recent observational record does not yield reliable evidence of trends in high-power cluster probabilities at this time. Large impacts to mid-latitude storm tracks are projected over the West Coast and eastern North America, with no less than 8 of the 9 models examined showing large increases by end-of-century in the probability density of the most powerful storms, ranging up to a factor of 6.5 in the highest range bin for which historical statistics are computed. However, within these regional domains, there is considerable variation among models in pinpointing exactly where the largest increases will occur.

Constraining precipitation processes in climate models with observations is crucial to accurately simulating current climate and reducing uncertainties in future projections. This work presents robust relationships between tropical deep convection, column-integrated water vapor (CWV), and other thermodynamic quantities analyzed with data from the DOE Atmospheric Radiation Measurement (ARM) Mobile Facility in Manacapuru, Brazil as part of the GOAmazon campaign and are directly compared to such relationships at DOE ARM sites in the tropical western Pacific. A robust relationship between CWV and precipitation, as explained by variability in lower tropospheric humidity, exists just as strongly in a tropical continental region as it does in a tropical oceanic region. Given sufficient mixing in the lower troposphere, higher CWV generally results in greater plume buoyancies through a deep convective layer. Although sensitivity of convection to other controls is suggested, such as microphysical processes and dynamical lifting mechanisms, the increase in buoyancy with CWV is consistent with the sharp increase in precipitation observed.

Entraining plume buoyancy calculations confirm that CWV is a good proxy for the conditional instability of the environment, yet differences in convective onset as a function of CWV exist over land and ocean, as well as seasonally and diurnally over land. This is largely due to variability in the contribution of lower tropospheric humidity to the total column moisture. Over land, the relationship between deep convection and lower free tropospheric moisture is robust across all seasons and times of day, whereas the relation to boundary layer moisture is robust for the daytime only. Using S-Band radar, these transition statistics are examined separately for unorganized and mesoscale-organized convection, which exhibit sharp increases in probability of occurrence with increasing moisture throughout the column, particularly in the lower free troposphere. An observational basis for an integrated buoyancy measure from a single plume buoyancy formulation that provides a strong relation to precipitation can be useful for constraining convective parameterizations. A mixing scheme corresponding to deep inflow of environmental air into a plume that grows with height provides a weighting of boundary layer and free tropospheric air that yields buoyancies consistent with the observed onset of deep convection across seasons and times of day, across land and ocean sites, and for all convection types. This provides a substantial improvement relative to more traditional constant mixing assumptions, and a dramatic improvement relative to no mixing. Furthermore, it provides relationships that are as strong or stronger for mesoscale-organized convection as for unorganized convection.

Downdrafts and their associated parameters are poorly constrained in models, as physical and microphysical processes of leading order importance are difficult to observe with sufficient frequency for development of robust statistics. Downdrafts and cold pool characteristics for mesoscale convective systems (MCSs) and isolated, unorganized deep precipitating convection in the Amazon are composited and both exhibit similar signatures in wind speed, surface fluxes, surface equivalent potential temperature () and precipitation. For both MCSs and unorganized convection, downdrafts associated with the strongest modifications to surface thermodynamics have increasing probability of occurrence with decreasing height through the lowest 4 km and show similar mean downdraft magnitudes with height. If is approximately conserved following descent, a large fraction of the air reaching the surface likely originates at altitudes in the lowest 2 km. Mixing computations suggest that, on average, air originating at heights greater than 3 km would require substantial mixing, particularly in the case of isolated cells, to match the observed cold pool . Statistics from two years of surface meteorological data at the GOAmazon site and 15 years of data at the DOE ARM site on Manus Island in the tropical western Pacific show that conditioned on precipitation levels off with increasing precipitation rate, bounded by the maximum difference between surface and its minimum in the profile aloft. Robustness of these statistics observed across scales and regions suggests their potential use as model diagnostic tools for the improvement of downdraft parameterizations in climate models.

The hydrologic cycle is significantly impacted by climate change, affecting both human societies and ecosystem health. Climate change-driven changes in the hydrological cycle can be subtle and difficult to identify, due to noise arising from natural climate variability and uncertainties in the tools used to understand this forced change, such as climate models. This dissertation investigates these changes across different scales, to discern the effects of climate change amidst inherent variability and modeling challenges. It explores the complex interplay between climate change, climate variability, and their impacts on the hydrological cycle, with a particular focus on precipitation patterns, wildfires, and droughts. Central to this research is the application of machine learning (ML) techniques to dissect and understand these phenomena on various spatial and temporal scales.

The first main research theme focuses on enhancing the understanding of the climate change signal within historical climate model simulations and future projections of precipitation on a global scale. In Chapter 2, utilizing ML methods and state-of-the-art climate model simulations, we identify the anthropogenic fingerprints in historical records of annual maximum daily precipitation across four observational and seven reanalysis datasets. Chapter 3 is dedicated to understanding future changes in precipitation patterns and reducing the uncertainty of these projections. We introduce an emergent constraint on the tropical atmospheric overturning circulation, a major contributor to model uncertainty in future precipitation patterns. Using ML, we estimate the observational estimates of this emergent constraint, yielding a constrained distribution of the future change in circulation (from -1.41±1.06 %/K to -2.20±0.93 %/K). Through this approach, we provide constrained spatial patterns of future precipitation changes, important for impact assessments.

The second main theme focuses on the regional and ecosystem-scale impacts of climate change and climate variability, through changes in the hydrological cycle. In Chapter 4, we focus on the historical fire season onset timing in 13 of California’s ecoregions, showing that onset is primarily controlled by climate variability and change via altering fuel moisture. Through this mechanistic knowledge, we quantify the contribution of climate change to the advancing trend in onset. We use an ML-based dynamical adjustment technique to separate the effects of climate change and variability on climatic drivers. We show that climate change has contributed to an advancement of onset by 5-55 days during the 1992-2020 period, across 11 out of 13 ecosystems. Chapter 5 focuses on the impact of droughts on California’s forests, particularly the severe 2012-2015 drought, which led to massive tree die-offs in the Sierra Nevada forests. We investigate the vulnerability of forests to drought, with a special interest in why certain areas were more affected and in the drought resistance of southern versus northern Sierra forests. Utilizing remote sensing and climate data, the research identifies a drought sensitivity timescale and examines the interaction between this timescale and drought severity to understand the spatial and temporal patterns of tree mortality. ML is employed to analyze factors contributing to tree mortality, revealing that forests in the Northern Sierras would be susceptible if the drought severity were spatially uniform. The study also explores potential future impacts of climate change on drought severity and forest vulnerability, using global climate simulations to predict changes in drought patterns and their effects on forest die-offs.

There has been a vast uptake of machine learning methods in climate science research over the last few years. The variety of data-driven approaches used throughout this thesis highlights the wide range of ML applications for understanding hydroclimate change and its impacts. These applications pave the way for future implementation of data-driven methods in the climate sciences, especially in separating the impacts of climate change from internal variability.