Theses and Dissertations
Permanent URI for this collectionhttps://hdl.handle.net/10217/100335
Browse
Recent Submissions
Item Open Access Improving predictions and generating actionable forecast insights for downslope windstorms with machine learning(Colorado State University. Libraries, 2025) Zoellick, Casey L., author; Schumacher, Russ, advisor; Rasmussen, Kristen, committee member; Barnes, Elizabeth, committee member; Nelson, Peter, committee memberDownslope windstorms are an extreme weather phenomenon characterized by accelerating winds down the lee slope of a mountain with gusts often exceeding 45 m s-1. These impact society through damage directly related to the high winds, ground transportation concerns in the vicinity of the windstorm, aviation impacts through the accompanying mountain wave turbulence, and fueling the rapid intensification and spread of wildfires such as the 2018 Camp Fire, the 2021 Marshall Fire, and the 2023 Lahaina Fire. Despite improvements in numerical weather prediction and observational datasets, predictability of these windstorms still rarely exceeds 12 hours further exacerbating their impacts. Recent advances have made machine learning (ML) more accessible to researchers and have shown promise in improving forecasts of other extreme weather phenomena. We first present models driven by two different types of ML architectures that classify wind events as moderate or high at three locations along the Rocky Mountain Front Range: Cheyenne, Wyoming; Fort Collins, Colorado; and Boulder, Colorado. The first type of architecture is the random forest (RF), which is comprised of multiple decision trees, and the second type is the convolutional neural network (CNN), which is a deep learning method that excels at image recognition. These models make forecasts at the Day 1 and Day 2 lead times based on predictors derived from a 12-km version of the WRF operated at Colorado State University. The results show improvement over the direct weather model forecasts. CNNs show enhanced event detection capability compared to the RFs but with a higher false alarm rate limiting their utility in some cases. Next, explainable artificial intelligence (XAI) techniques are presented. Feature importances indicate that the ML models rely on predictors at geographic locations that align with known atmospheric variables important to downslope wind forecast along the terrain. Also, a framework for reducing the dimensions of the predictor data and clustering these data with a Gaussian mixture model yields insights to the forecast ML models' performance and the synoptic conditions in which downslope windstorms along the Front Range occur. The ML models perform better in regimes characterized by prominent synoptic features such as cold air advection or the presence of the jet stream aloft. Lastly, we investigate whether increasing the resolution of the traditional weather model creating the ML predictors results in performance improvements. We use NOAA's High Resolution Rapid Refresh (HRRR) model to derive input predictors for newly trained CNNs and observe a decrease in false alarms that results in an overall performance boost over the direct HRRR forecasts. A case study on the Marshall Fire is conducted and indicates that the HRRR-based CNN is able to correctly forecast the subsequent downslope wind event before the wind event is explicitly depicted in the HRRR output itself. This study is an example of how ML fused with current weather models closes the forecast gap in these impactful weather phenomena with incomplete physical understandings.Item Open Access Insights from machine learning-based forecasts of convective hazards and environments(Colorado State University. Libraries, 2025) Mazurek, Alexandra Callahan, author; Schumacher, Russ S., advisor; Rasmussen, Kristen L., committee member; van den Heever, Susan C., committee member; Chen, Haonan, committee member; Hill, Aaron J., committee memberSevere convective thunderstorms and their associated hazards are costly, damaging, and difficult to predict. Machine learning (ML) techniques are rapidly being developed and deployed in an effort to predict severe thunderstorms more quickly and with greater accuracy than traditional methods. With these developments, there is a need to understand how ML-based weather prediction systems rely on atmospheric data and generate their forecasts. This work probes a number of ML-based convective thunderstorm-related forecasts over the contiguous United States to 1) understand how they make their predictions, 2) diagnose where their strengths and deficiencies may lie, and 3) explore how well their predictions resemble physical characteristics of the atmosphere. The insights gleaned from this research aim to support operational use of ML-based forecast guidance. First, probabilistic ML-based forecasts of severe convective hazards (i.e., tornadoes, hail, and thunderstorm-driven winds) from the Colorado State University Machine Learning Probabilities (CSU-MLP) system are studied using an explainable machine learning technique known as Tree Interpreter (TI). TI provides context to the CSU-MLP forecasts by disaggregating its forecast probabilities into "contributions" by each of the environmental variables that are used to train the model. This technique allows one to see the extent to which each atmospheric "ingredient" contributes to the final predictions. Results of this work show that CSU-MLP uses environmental information to make its predictions in ways that resemble the climatology and environments of severe storms, and the values of the contributions generally scale with values of the environmental inputs, effectively enhancing the interpretability of the ML system. Second, CSU-MLP forecast performance is examined across different synoptic regimes in an effort to understand which types of environmental conditions tend to lead to skillful versus less-skillful forecast performance. Self organizing maps (SOMs), which are a type of ML, are employed to statistically diagnose regimes across two years of reanalysis data. The skill of day-2 CSU-MLP probabilistic tornado, wind, and hail forecasts are examined across the SOM-identified regimes. This work shows that SOMs are successful at identifying distinct atmospheric patterns using only surface-based convective available potential energy (SBCAPE) and vertical wind shear as inputs. At times, the best- and worst-performing CSU-MLP forecasts occur under highly similar atmospheric conditions, though the best-performing forecasts tend to be characterized by strong synoptic forcing and many storm reports. Third, forecast output from three deep learning weather prediction (DLWP) models, GraphCast, Pangu-Weather, and FourCastNetv2, is studied to investigate how well they model severe storm environments and capture convection-related parameters. This work explores both native and derived fields from 22 months of daily forecasts from these three models, all of which were initialized with input conditions from the Global Forecasting System (GFS). The output is compared to ERA-5 reanalysis and GFS forecasts, both broadly and for specific convective events. Overarching results from this study show that the DLWP model forecasts tend to be characterized by less moisture and greater instability compared to ERA-5. For specific events, the DLWP forecasts can reasonably capture convective environments at least a week in advance and are competitive with the GFS. However they tend to underforecast the vertical wind shear magnitude, and their limited vertical resolution can lead to overly smooth profiles that lack key details such as stable layers.Item Open Access Data-driven improvements to GPROF-based satellite snowfall retrievals with a focus on mountain snowfall(Colorado State University. Libraries, 2025) Gonzalez, Ryan L., author; Kummerow, Christian, advisor; Liston, Glen, committee member; Chiu, Christine, committee member; Rasmussen, Kristen, committee member; Notaros, Branislav, committee memberSnowfall is a critical component of Earth's hydrological and climate system despite only 5% of Earth's annual precipitation falling as snow. Satellite-based snowfall estimates, particularly those obtained from the Global Precipitation Measurement (GPM) Microwave Imager (GMI), struggle to accurately estimate the total annual snowfall accumulations, especially in mountainous regions of the world. Part of the challenge is due to the reference precipitation used in the GMI-based algorithms, while radiometers struggle to distinguish between the microwave signatures of surface snowpack and snowfall. The aim of this dissertation is to evaluate the impact machine learning-based GMI retrievals have on snowfall estimates, explore how temperature and climatological adjustments to the reference precipitation can provide additional information to the retrieval, and asses if these changes lead to improved snowfall accumulations required for modeling the lifecycle of snow. A key objective of this study is to improve snowfall accumulation estimates in mountainous areas, where snowpack is a critical component of water storage. First, snowfall rates estimated from the Goddard Profiling Algorithm (GPROF) for GMI are compared using three types of GPROF algorithms: one Bayesian (GPROF V7) and two neural network versions (GPROF-NN 1D and GPROF-NN 3D). The highest detection and quantitative statistics are observed using GPROF-NN 3D with both neural network retrieval algorithms outperforming the Bayesian version. It is shown that artificial biases in the retrieval statistics can result from the selected threshold for snow/no-snow classification. Coincident in-situ snowfall and radar data are also used to evaluate the temperature dependency of the reflectivity-snowfall (Z-S) relationship and how it impacts the GPROF retrievals. Second, an evaluation of the three GPROF algorithms is conducted in the mountains of the western United States. Using data from a snow reanalysis dataset, water year snowfall accumulations from the Multi-Radar Multi-Sensor (MRMS) are adjusted to produce more realistic snowfall magnitudes and spatial patterns. These adjustments were found to decrease errors in snowfall accumulation estimates for all three retrieval algorithms, resulting in significant improvements when compared to independent SNOTEL observations. These results provide a positive outlook for snowfall retrievals in mountainous regions by incorporating additional information to the retrieval algorithm. Finally, a framework for incorporating satellite precipitation estimates into a snow evolution model in the western United States is presented that offers a flexible design to account for different study domains. The objective of this framework is to present an approach for deriving snow water equivalent (SWE) from satellite precipitation estimates given the difficulties of directly measuring SWE from passive microwave sensors. A UNet-based retrieval model is used to estimate precipitation at 30 minute time resolution across the currently available passive microwave and infrared sensors. The initial precipitation estimates were found to have a systematic bias across the study period, which, after correction, produced realistic spatial patterns of snow depth and snow water equivalent, but underestimated the magnitudes compared to two reference snow model simulations.Item Open Access Cold pool propagation and cold pool-land surface interactions(Colorado State University. Libraries, 2025) Falk, Nicholas Michael, author; van den Heever, Susan C., advisor; Grant, Leah D., advisor; Bell, Michael M., committee member; Schumacher, Russ S., committee member; Venayagamoorthy, Subhas K., committee memberConvective cold pools are important components of the Earth system as they influence processes such as deep convective initiation, storm longevity and intensity, surface energy fluxes, and aerosol transport. The overarching goal of the research outlined in this dissertation is to investigate the propagation characteristics of cold pools, as well as the interactions between cold pools and the land surface. The three studies comprising this dissertation use field campaign observations and high-resolution numerical simulations to investigate these cold pool processes. The first study evaluates a popular density current propagation speed equation using a large, novel set of radiosonde and dropsonde observations. First, data from pairs of sondes launched inside and outside of cold pools, along with the theoretical density current propagation speed equation, are used to calculate sonde-based propagation speeds. Second, radar/satellite- based propagation speeds are calculated by manually tracking the propagation of cold pools and correcting for advection due to the background wind. Comparisons of the propagation speeds calculated in these different ways demonstrate that sonde-based and radar- based propagation speeds are strongly correlated for US High Plains cold pools, suggesting the density current propagation speed equation is appropriate for use in midlatitude continental environments. Sonde-based propagation speeds are largely insensitive to how cold pool depth is defined, since the preponderance of negative buoyancy is near the surface in cold pools. Sonde-based propagation speeds can vary by ~300% based on where and when the sondes were launched, suggesting sub-mesoscale variability could have a major influence on cold pool propagation. The impacts of topographic slope on daytime haboob propagation speeds and dust lofting are examined in the second study comprising this dissertation, along with how these impacts are modulated by surface roughness length. A suite of 40 idealized, large-eddy simulations are conducted with varied linear topographic slopes and surface roughness lengths. It is found that on flat ground, greater surface roughness increases drag on haboobs and causes haboobs to dissipate faster, thereby decreasing both haboob propagation speeds and associated dust lofting. As the topographic slope is increased, an upslope anabatic wind forms which causes downslope haboob propagation speeds to decrease and upslope haboob propagation speeds to be mostly unchanged. Anabatic winds act to loft dust as well, leading to increased masses of dust being lofted jointly by the haboob and anabatic wind as topographic slopes are increased. The third study investigates the individual and synergistic impacts of cold pools and land surface heterogeneity on convection initiation. Idealized large eddy simulations of deep convection over the Amazon rainforest are conducted testing realistic and homogenized vegetation, along with realistic and eliminated cold pools. Convection initiation is more frequent over forested than deforested areas due to more favorable thermodynamics. Heterogeneous vegetation aggregates storm initiation locations compared with homogeneous vegetation. Over heterogeneous vegetation, cold pools propagate into deforested regions, thereby initiating storms and disaggregating storm initiation locations. Convection initiation locations are randomly distributed over homogeneous vegetation, with or without cold pools, demonstrating that cold pools have minimal impacts on convection initiation locations over homogeneous vegetation. The findings of this dissertation research shed new light on fundamental cold pool processes and should be helpful for improving the representation of cold pools in forecast and climate models. Several avenues for future research are discussed.Item Embargo Extreme rainfall mechanisms in Hurricane Fiona (2022)(Colorado State University. Libraries, 2025) Nieves Jiménez, Angelie, author; Bell, Michael M., advisor; Schumacher, Russ, committee member; Anderson, Brooke, committee memberHurricane Fiona's (2022) historical heavy precipitation devastated the Caribbean Island of Puerto Rico after it made landfall as a category 1 hurricane. Rainfall accumulation totals in southern interior region areas surpassed 900 mm during 18 – 19 September 2022. To analyze the rainfall mechanisms, we use output from the Hurricane Analysis and Forecast System (HAFS) configuration "B" modeling system and observations from the Puerto Rico Next Generation Weather Radar (NEXRAD) Level 2 Doppler radar and rain gauges around the island. Quantitative precipitation estimates from radar and rainfall measurements suggest that HAFSB simulated reasonably well the precipitation amounts and location. HAFSB track differences from the real trajectory contributed to discrepancies between the simulated and observed rainfall. We investigate three stages of the Hurricane Fiona rain event, each focusing on different processes. The first stage is associated with the primary eyewall and rainfall produced through boundary layer convergence. The second stage focuses on the principal rainband affecting the island and is associated with rainfall enhancement from vertical wind shear interactions with Fiona's potential vorticity. The third and final stage analyzes the enhancement of a "tail rainband" both over open water and the southern portion of the island as Hurricane Fiona kept strengthening west of Puerto Rico. Our findings support the hypothesis that evaporative cooling within inner core rainfall from stage one and two sets up a favorable environment for isentropic uplift to enhance rainfall production in stage three. Additional enhancements of the rainfall occurred over Puerto Rico's high terrain by orographic effects.Item Open Access Eulerian and Lagrangian analyses of bioaerosol transport in three deep convective storm morphologies(Colorado State University. Libraries, 2025) Davis, Charles M., author; van den Heever, Susan C., advisor; Kreidenweis, Sonia M., committee member; Jathar, Shantanu, committee memberIn this thesis, we investigate the entrainment and transport of aerosol particles in a representative isolated deep convective storm, supercell, and squall line using idealized high-resolution mesoscale model simulations. We focus our investigation on the extent to which air from rainy surface regions, which have been noted in the literature to be sites of aerosolization of biological particles, is able to enter and subsequently be transported within these storm morphologies. We also investigate the residence time in supersaturated environments experienced by these parcels as they are entrained. The first part of this study quantifies the magnitude and timing of entrainment of air from the surface, and from rainy surface regions specifically, in all three storm morphologies. We use inert tracer quantities to constrain the timing with which rainy (referred to as rain-sourced tracers) and other surface air (referred to as fixed-source tracers) is entrained into the storms, and the fraction of each storm's updraft that is composed of air from these regions. At its peak, the isolated convective storm entrains the greatest proportion of surface-based air seen in any of the storms. However, it also attains the smallest concentrations of rain-sourced tracer and the smallest proportion of rain-sourced tracer in its updraft, indicating that significantly less of its entrained surface air originates in regions of potential rain-induced aerosolization of bioaerosols. The squall line and supercell attain greater values of both these quantities and sustain them for longer periods, indicating that more air in their updrafts originates in rainy regions. For light rain (>= 1 mm/hr), the squall line and supercell entrain comparable concentrations of air from rainy regions, but for heavy rain (>= 40 mm/hr) the squall line entrains significantly more. The second part of this work investigates the specific pathways by which surface air is entrained into these storms as well as the environments experienced by entrained surface-based air parcels. We do this by calculating parcel trajectories using the output of the aforementioned mesoscale simulations, initializing air parcels at various times within each storm's life cycle, and separately evaluating the trajectories of parcels originating in rainy and non-rainy regions. The isolated convective storm simply moves over and entrains the parcels not initialized in rainy regions into its updraft directly by its strong surface convergence. The squall line and supercell entrain non-rainy parcels by gust-front lofting, in which the circulation at the leading edge of the cold pool lofts the parcels to a level at which they can be entrained by the updraft behind the gust front. The isolated convective storm entrains parcels originating in rainy regions via the horizontal vortical circulation in the head of the cold pool, which lofts them and redirects them towards the updraft. The squall line also entrains rainy parcels by this same circulation in its cold pool. The supercell, on the other hand, entrains rainy parcels from a relatively narrow region within and just outside of the leading edge of the forward-flank downdraft's cold pool via a combination of gust front lofting and the known phenomenon of the "recycling" of some negatively buoyant air from the forward-flank downdraft's cold pool into its updraft. We find that the time these entrained parcels spend in supersaturated environments is a strong function of storm morphology. Parcels entrained into the squall line spend nearly twice as long on average in supersaturated regions as entrained parcels in the other two storm types. This arises because the squall line parcels take longer to reach mid-levels after first being lofted by the circulation in the head of the cold pool. This longer transit time is due to the upshear tilt of the updraft, as well as from more complex 3D variations in the structure of the gust front and updraft.Item Open Access The impact of stratospheric aerosol injection: a regional case study(Colorado State University. Libraries, 2025) Cohen, Sabrina L., author; Hurrell, James W., advisor; Keys, Patrick W., committee member; Lombardozzi, Danica L., committee memberThe detrimental effects of anthropogenic climate change have become ubiquitous as global greenhouse gas emissions concentrations continue to increase. As a result, research into proposed climate intervention (CI) techniques to offset some of the most damaging effects of climate change is increasing, with the idea that CI could provide more time for humanity to pursue decarbonization. One of the most researched CI techniques is stratospheric aerosol injection (SAI), which would reflect a small portion of sunlight away from Earth to reduce or lower temperature increases. While many studies have analyzed SAI's potential global impacts on climate variables, such as temperature and precipitation, relatively few have examined regional impacts on variables more intimately tied to human well-being, such as crop productivity. Thus, using climate model data, we analyze the projected impacts of one future climate change and three SAI scenarios on four Global South regions already socioeconomically vulnerable to climate change: South Asia (SAS), East Asia (EAS), South Central America (SCA), and West Africa (WAF). We find that, in the SAI scenarios, heat extremes are reduced and wet season precipitation, soil moisture and crop productivity increase relative to the climate change scenario in all four regions. Further, SAI clearly ameliorates crop productivity losses produced by climate change in WAF and SCA, with less clear benefits in SAS and EAS. Our study indicates the potential for SAI (in the scenarios examined here) to alleviate some of climate change's adverse impacts on human welfare.Item Open Access Regularized linear regression to estimate the spatial sensitivity governing the pattern effect, comparative analysis to contemporary methods, and observational applications(Colorado State University. Libraries, 2025) Fredericks, Leif, author; Rugenstein, Maria, advisor; Thompson, David W. J., advisor; Cooley, Daniel S., committee memberHow the spatially varying temperature field affects global radiation (i.e., the "pattern effect") is crucial to understanding how sensitive Earth's temperature is to anthropogenic forcing. We capture this phenomenon in a sensitivity map using regularized linear regression. When trained on 1,000 simulated years in a climate model, the resulting sensitivity maps are consistently able to explain over 75% of the variance in net top-of-atmosphere radiation in an out-of-sample internal variability test. However, when the training data are constricted to 24 years to mirror the length of available observations, that value ranges between 0% and 75% with a median of 50%. This implies that 24-year observational sensitivity maps produced by our method carry significant uncertainty. Tested against the forced climate response in an RCP 8.5 simulation, the ideal 1,000-year training case captures ~75% of the forced response magnitude, while sensitivity maps derived from 24-year periods are unreliable for projecting the warming scenario. Acknowledging the implication that our results depend highly on the particular behavior of the last two decades, we present the first physically interpretable radiative feedback sensitivity maps derived entirely from observations. We then unify several alternative methods under a common training and testing procedure. These methods all generate predictive frameworks from internal variability, except for an included Green's function. The latter approach was the primary method used to generate pattern effect sensitivity maps prior to the methods discussed in this thesis, so it grounds our comparative analysis to the current state-of-the-field. All methods match or improve upon the Green's function's ability to predict internal variability, but vary widely in their ability to predict a step forcing 4xCO2 warming simulation.Item Open Access Impacts of historic anthropogenic aerosol forcing on large climate ensembles through the lens of poleward energy transport(Colorado State University. Libraries, 2024) Needham, Michael Robert, author; Randall, David A., advisor; Rugenstein, Maria, committee member; van Leeuwen, Peter Jan, committee member; Rugenstein, Jeremy, committee memberIn discussions of the human impact on Earth's climate, aerosols receive much less attention than greenhouse gases. And yet, the change in the global mean effective radiative forcing from anthropogenic aerosols was roughly of the same magnitude (but of opposite sign) as the change in greenhouse gases throughout much of the twentieth century. Aerosols also represent the largest uncertainty in the effective radiative forcing, due to their complex interactions with clouds and solar radiation. Complicating this even further, aerosols are relatively short-lived within the atmosphere, and thus exhibit a large degree of variability in space and time. This dissertation presents a set of studies which investigate the ways in which historic anthropogenic aerosols may have impacted the Earth's weather and climate, through the analysis of a large number historic climate model simulations which comprise so-called large ensembles. Analysis of these ensembles allows for the isolation of some forced signal (e.g., the influence of aerosols) from the noise (i.e., the background variability of the model). This leads to conclusions through the analysis of summary statistics across members of the ensemble population which would be impossible to make based on only one or a few simulations. In particular, these studies show that the emission of aerosol precursors from Europe and North America increased the northward transport of heat from the southern into the northern hemisphere in an ensemble of simulations performed with version 2 of the Community Earth System Model (CESM2). The additional heat transport was in excess of 0.25 PW. This is an increase of at least 4-5% compared to the baseline maximum transport of between 5-6 PW which occurs in the mid-latitudes. At latitudes away from these maxima, the increase was a much larger percentage of the total. This anomalous northward energy transport was accomplished by changes in both atmospheric and oceanic processes. These include a southward shift of the Intertropical Convergence Zone (ITCZ) associated with changes in the Hadley cells; an increase in the frequency of extratropical cyclones in the north Atlantic; a strengthening of the Atlantic Meridional Overturning Circulation (AMOC); as well as changes to multiple ocean processes across the Indo-Pacific. Comparison of these results to the literature indicates that this modeled response to aerosols in CESM2 is likely too large. Furthermore, analysis of two additional large ensembles reveals that this over-sensitivity of CESM2 cannot be due to some deficiency in the model. Instead, it is demonstrated that the difference is the result of changes to the historical emission estimates between phase 5 and phase 6 of the Coupled Model Intercomparison Project (i.e., CMIP5 and CMIP6). This finding leads to the hypothesis that the higher interannual variability associated with a change from decadal-scale CMIP5 emissions to annual-scale CMIP6 emissions is the ultimate cause of the overzealous response of the model. Testing this hypothesis likely will provide the most fertile ground for future work.Item Open Access Radiative feedbacks in tropical organized convection and the Madden-Julian oscillation(Colorado State University. Libraries, 2024) Hsiao, Wei-Ting, author; Maloney, Eric D., advisor; Rugenstein, Maria A. A., committee member; Kummerow, Christian D., committee member; Randall, David A., committee member; Mueller, Nathaniel D., committee memberThe organization of tropical deep convection is supported by radiative feedbacks, in which high clouds and moisture anomalies associated with convection imposes anomalous longwave (LW) radiative heating in the atmosphere, further supporting convection. Despite an abundance of studies using numerical simulations, the interactions between tropical convective organization, radiative feedbacks, and the large-scale atmospheric environment have not been comprehensively examined in real-world observations. The present dissertation examines such interactions among tropical mesoscale organized convection, radiative feedbacks, and the Madden-Julian oscillation (MJO) using a set of observation-derived data products, including retrievals using spaceborne satellites and ground-based precipitation radar, along with combined products and reanalyses. The main findings in each chapter are summarized as follows: (1) higher sea surface temperature and stronger low-level wind shear strength enhance tropical mesoscale convective activity, increasing cirrus cloud cover and LW heating generated per unit precipitation. (2) the estimation of LW cloud-radiative feedback (LW CRF), defined as the LW cloud-radiative heating produced per unit precipitation, is sensitive to the precipitation data set used. (3) radiatively driven circulation and the associated moistening effects in the MJO can be derived in a weak-temperature-gradient framework and a linear baroclinic model. The result suggests that LW heating moistens the MJO more efficiently than the total apparent heat source, while shortwave (SW) radiative effects dry the MJO. (4) The LW CRF of the MJO is spatially inhomogeneous, with stronger feedbacks over the tropical Indian ocean and to the northwest of Australia, but weaker feedbacks over the tropical western and central Pacific. The spatial pattern may be determined by the spatial distribution of preferred convective types and precipitation efficiency.Item Open Access Bridging human and artificial intelligence for skillful, trustworthy, and insightful seasonal-to-decadal climate prediction(Colorado State University. Libraries, 2024) Rader, Jamin K., author; Barnes, Elizabeth A., advisor; Rasmussen, Kristen L., committee member; Hurrell, James W., committee member; Stevens-Rumann, Camille S., committee memberSeasonal-to-decadal climate variability is inherently difficult to predict and is intimately connected to human and natural systems worldwide. Skillful forecasts on two-month to ten-year timescales would enable proactive and informed decision-making for many industries, including fisheries, water management, and agriculture. Understanding the behavior of seasonal-to-decadal climate variability provides context for our changing environment. Neural networks, a class of artificial intelligence tools, are well-suited for exploring teleconnections, precursors, and patterns of variability, since they can identify complex relationships within immense quantities of data. Neural networks have traditionally been used as "black-box" models that produce predictions but are inherently difficult to explain. There has been a recent push to develop "interpretable" models that can be understood by human scientists. In this dissertation, I bridge human and artificial intelligence to leverage interpretable AI for skillful, trustworthy, and insightful prediction of seasonal-to-decadal climate variability. First, I show how interpretable neural networks can be used to optimize a simple forecasting method, analog forecasting. This approach highlights four precursor patterns for one-year forecasts of El Niño Southern Oscillation in the Tropical Pacific, West Pacific, Baja Coast region, and Tropical Atlantic. In addition, when making five-year forecasts of observed sea surface temperature variability in the North Atlantic, this optimized analog forecasting approach rivals the performance of an initialized decadal prediction system. Second, I design neural networks to learn patterns of internal variability and forced change. Using these neural networks, I perform climate change attribution for observed sea surface temperatures. Despite the unprecedented, record-high, global-mean sea surface temperature in 2023, our results suggest that much of this warming can be explained by internal variability, as anomalously cold conditions in 2021 and 2022 shifted to anomalously warm conditions in 2023. Third, I use neural networks to make decadal forecasts of the likelihood that annual-global-mean temperature exceeds 1.5˚C, a critical Paris Agreement temperature threshold. These forecasts predict that it is very likely that annual-global-mean temperature exceeds 1.5˚C in the next decade (2024-2033), serving as a harbinger for future climate change. These forecasts are consistent with dynamical initialized prediction systems, demonstrating that neural networks can provide skillful decadal forecasts at reduced computational expense. Neural networks are powerful tools for prediction, and facilitate deeper discovery of our chaotic, interconnected, predictable Earth.Item Open Access Investigating and mitigating errors in the remote sensing of maritime low clouds at night(Colorado State University. Libraries, 2024) Turner, Jesse, author; Miller, Steven D., advisor; Kummerow, Christian D., committee member; Smith, Ryan G., committee member; Noh, Yoo-Jeong, committee memberLow clouds are ubiquitous to the world's oceans, affecting aviation, maritime transportation, and the structure and dynamics of the broader atmospheric system. Understanding the diurnal properties and distributions of these clouds requires an observing system capable of spanning vast regions of ocean devoid of surface-based observations. Here, earth observation satellite imagery provides potentially valuable information on cloud coverage over the oceans. The brightness temperature difference (BTD) between the longwave infrared (e.g., 11 µm) and shortwave infrared (e.g., 3.9 µm) window band measurements is commonly used as a first-order bi-spectral test to identify low clouds over the ocean at night. Occasionally, unusual patterns of clear-sky features in this BTD occur, giving rise to spurious false-positive cloud measurements. These confusing signals are caused by nuances of the atmospheric and surface emission sensitivity at these two wavelengths. Ideally, positive values in the 11 µm - 3.9 µm BTD are caused by actual low clouds, owing to slightly higher emissivity at the longwave IR compared to the shortwave IR. However, a clear-sky environmental scenario can mimic this signal: a warm and moist air mass over a cold region of water. These same environmental conditions are conducive to advection fog formation, compounding the interpretation of conventional infrared-based cloud detection in these regions. Moonlight reflectance, when available from the Day/Night Band on the Visible/Infrared Imaging Radiometer Suite (VIIRS), can help to disentangle cases of actual vs. false low cloud (FLC). This research examines cases from the United States east coast, the Mexico south coast, and the large-scale Gulf Stream to investigate the physical causes of false cloud signals. Insight gained from this research can help forecasters and researchers determine which physical regions are prone to false alarms, and in complement, which regions offer higher confidence for cloud detection. Further, this study uses numerical model data and radiative transfer simulations to estimate the positive signals caused solely by air mass over cold water effects. This simulation method lends insight on the global extent and frequency of nighttime maritime low cloud overstatement. Knowledge of the patterns of false signals in the IR BTD provides opportunities to improve products that depend on the nighttime low cloud test, such as fog and visibility warnings, sea-surface temperature cloud masking, and cloud climatologies used for climate research. The simulation also provides a novel predictive tool for anticipating potential regions of both false alarm low cloud and regions prone to advection fog formation.Item Open Access Investigating the potential of meltwater as a local source of ice nucleating particles in the central Arctic summer(Colorado State University. Libraries, 2024) Mavis, Camille, author; Kreidenweis, Sonia, advisor; Creamean, Jessie, advisor; Pierce, Jeffrey, committee member; Peers, Graham, committee memberDue to climate change, the Arctic has crossed a threshold into positive feedbacks between sea-ice loss and increased absorption of solar radiation, causing warming up to four times the global average. Parameterizing the Arctic radiation budget to predict the new steady-state is paramount for guiding policies impacting future global socioeconomics and Arctic livelihoods. Arctic mixed-phase clouds (AMPCs) are a pillar in the feedback systems by modulating the surface energy budget, depending on the partitioning of cloudwater between ice and liquid phases that is sensitive to the concentration of ice nucleating particles (INPs) in the atmosphere. However, current observational gaps of central Arctic INP concentrations and sources may contribute to current challenges in resolving the controls on Arctic cloud ice content. The year-long expedition aboard the RV Polarstern from 2019 - 2020, entitled The Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC), was a highly coordinated interdisciplinary effort that provided a unique opportunity to observe INPs in the central Arctic. The Arctic summer is a unique period characterized by pristine aerosol conditions, in which emissions from local sources have an increased influence, potentially impacting the ubiquitous low-lying AMPCs. Thus, the summer is an ideal season for exploration of the potential importance of INPs from local sources, such as melt ponds. In this study, we used the Colorado State University (CSU) Ice Spectrometer and chemical treatments to determine the INP concentration and inferred composition in source samples of bulk sea water and meltwater from ponds and leads over the month of July. In addition, ambient aerosol filters were deployed both on the ship and on the ice, downwind of these meltwater features. We found that the concentration of INPs in meltwater was 10 times higher than in the mixed layer of the ocean, a surprising result since previous studies did not see a difference in the two source samples. The INPs in meltwater were capable of freezing at temperatures (T) ≥ −10 °C and were predominantly biological, based on our heating assay. Biological INPs capable of freezing at T ≥ −10 °C were present in 80 % of the on-ice aerosol samples. The alignment of slopes of the cumulative INP spectra between the meltwater and aerosol filter samples at T ≥ −15 °C suggested an influence from meltwater on the aerosol INPs at those temperatures. Similarities between aerosol INP sampled on the ice and on-board Polarstern suggested that the on-ice INP concentrations were likely influenced by a regional meltwater source signature, rather than being measurably impacted by a singular upwind pond. A relationship was observed between wind speed, supermicron particle counts, and on-ice aerosol INP populations active at warm (−15 °C) and cold (−25 °C) temperatures. A distinct on-ice aerosol sample containing no INPs active at T ≥ −15 °C was found to be influenced by southerly air over the ice-free ocean, emphasizing the potential impact meltwater may have as a unique source of warm temperature INPs in the central Arctic. These findings suggest that summertime central Arctic biological INP concentrations may increase if, as predicted, a spatio-temporal expansion of the melt season occurs in the near future. This increased INP concentration from local sources could impact central Arctic cloud microphysics, and thus their impact on the surface energy budget.Item Open Access Quantification of volatile organic compound emissions from unconventional oil and gas development(Colorado State University. Libraries, 2024) Zhang, Weixin, author; Collett, Jeffrey L., Jr., advisor; Pan, Da, committee member; Pierce, Jeffrey R., committee member; Ham, Jay M., committee memberOil and gas (O&G) development in the U.S. has accelerated in the past two decades, aided by unconventional extraction techniques including hydraulic fracturing and horizontal drilling. Potential environmental and health impacts of volatile organic compounds (VOCs) originating from O&G activities in populated regions have raised concerns. In Broomfield, Colorado, six new O&G well pads were approved for development in 2017 and an air monitoring program was established in October 2018 to collect weekly and later plume-triggered air samples. This study addresses the limited existing knowledge of activity-specific VOC emission rates from unconventional O&G development (UOGD), utilizing these observations and dispersion model simulations through emission inversion methods. Emissions are characterized from well drilling, hydraulic fracturing, coiled tubing/millout, flowback, and production operations. Substantial variations in average VOC emission rates, determined using weekly canister observations, are observed across different UOGD phases. Drilling and coiled tubing/millout operations exhibit the highest total VOC emission rates, attributed to hydrocarbon release from shale formations and drilling mud. In contrast, hydraulic fracturing gives lower emission rates, consistent with injection of fluids into the well during this operation, minimizing the probability of subsurface hydrocarbon emissions. Diesel-powered engines are identified as the primary ethyne sources during hydraulic fracturing. Production was characterized by lower VOC emission rates than pre-production phases but remains an important emission category due to its long duration (decades). Variations of emission rates within each phase highlight the complexity of factors and activities influencing emission rates, including, for example, vertical vs. horizontal drilling and periodic maintenance activities. VOC emission rates associated with drilling mud volatilization and hydraulic fracturing suggest that previously published emission estimates (EPA (2022), and Hecobian et al. (2019)) underestimate average VOC emission rates during these activities. Significantly lower emission rates during flowback compared to previous work (Hecobian et al., 2019) reveal how improved management practices, including tankless, closed-loop fluid handling systems have effectively reduced what used to be a dominant source of pre-production VOC emissions. Plume-triggered samples, capturing transient high-concentration plumes, reveal short-term VOC emission rates approximately an order of magnitude higher for drilling and flowback than determined from weekly samples. In the case of flowback, short-term emission pulses have been linked to periodic emptying of sand canisters used to trap fracking sand emerging from previously fracked wells.Item Open Access Effects of warming and stratospheric aerosol injection on tropical cyclone distribution and frequency in a high-resolution global circulation model(Colorado State University. Libraries, 2024) Feder, Andrew, author; Randall, David, advisor; Hurrell, James, committee member; Rugenstein, Jeremy, committee memberTropical cyclones (TCs) occur stochastically in any given TC season, with varying numbers and intensities within basins over time. Nevertheless, they arise out of fundamental laws of thermodynamics and fluid physics, and in recent years, as global circulation models (GCMs) have increased in spatial resolution, increasingly realistic TCs and TC distributions have emerged from them. Where prior research on TC climatologies has relied on proxies like Potential Intensity (PI) and synthetic storm models, the cyclones emerging from the dynamics of newer GCMs can now be analyzed directly, using native model variables. Such direct analysis may be particularly useful in studying possible global storm distributions under radically altered future climates, including high-emissions warming scenarios, and even those shaped by climate interventions. These interventions include various directed changes in global albedo, such as Stratospheric Aerosol Injection (SAI), with only limited precedent in the historical period. GCMs simulating realistic climate intervention scenarios, have not as of yet paired storm-resolving resolution with realistic intervention scenario construction. This has left gaps in our understanding as to how interventions might affect global storm/TC distributions, and whether ameliorating warming in this way could also substantially lessen related natural disaster risk profiles. In this paper, we utilize a new high-resolution model configuration to conduct experiments examining the effects of SAI, on tropical cyclones and global storm physics more broadly. These experiments are constructed based on prior work on SAI using the GLENS GCM ensemble (Tilmes et al. 2020; Danabasoglu 2019a,b). Our analysis centers on 3 10-year experiments conducted using 30-km grid spacing. These include a recent-past calibration run; the Intergovernmental Panel on Climate Change climate pathway SSP 8.5 (IPCC 2021), for the years 2090-2099, with no SAI; and SSP 8.5, with SAI having begun in 2020 to maintain a global temperature rise of no more than 1.5° C, also simulated for the years 2090-2099. With the resulting data sets, we deploy a novel TC-tracking algorithm to analyze resulting changes in storm tracks and properties. Based on our results for these different scenarios, we find that SAI, while in some ways restoring global storm patterns to a pre-warming state, may also create unique basin-scale TC distribution features and pose novel related hazards.Item Open Access Satellite observations of oceanic high-latitude drizzle using a combined radar-radiometer retrieval(Colorado State University. Libraries, 2024) Jones, Spencer R., author; Kummerow, Christian, advisor; Chiu, Christine, committee member; Chandrasekaran, Venkatachalam, committee member; Grassotti, Christopher, committee memberThe high latitude oceans are problematic for satellite estimations of precipitation due to the high frequency of occurrence of light drizzle and snowfall. Passive microwave radiometric observations are sensitive to integrated cloud water path and provide good sampling for robust statistics but have little skill in distinguishing precipitation onset from cloud water and cloud ice due to a lack of sensitivity to drop sizes when they are small. Spaceborne precipitation radars to date have lacked sensitivity to drizzle, and cloud radars have suffered from both the uncertainties inherent in Z-R relations and poor sampling due to nadir-only scans. This study combines coincident active and passive microwave observations from CloudSat's Cloud Profiling Radar (CPR) and the Advanced Scanning Microwave Radiometer (AMSR2) to resolve cloud and hydrometeor distribution parameters and to force consistency between the two independent sets of coincident observations. Consistency between the radar and radiometer is found by using an optimal estimation (OE) retrieval algorithm, a physics-based technique that simultaneously resolves the most likely atmospheric state given both radar and radiometer observations as well as a priori information. The OE algorithm uncertainties are estimated using a method that attempts to emulate the departure in observation space of retrieved states from the unknown true state. The focus on observational uncertainties and the accuracy obtained by using nondiagonal observational error covariance matrices allows the algorithm both to resolve states that are radiatively consistent and to reduce the level of nonuniqueness found in dealing with passive observations alone. The result is an estimation of drizzle frequency and intensity that are consistent with both the CPR and AMSR2 observations for the high latitude oceans. We find that zonal means of retrieved high-latitude drizzle below 0.25 mm hr-1 from these combined observations (0.263 mm day-1) falls slightly above those of CloudSat estimates (0.244 mm day-1), provided by the 2C-RAIN-PROFILE and 2C-SNOW-PROFILE products (Lebsock 2018; Wood and L'Ecuyer 2018), and far below that of radiometer-only estimates (0.920 mm day-1) provided by GPROF (Kummerow et al. 2015).Item Open Access The signature of the western boundary currents on tropospheric climate variability(Colorado State University. Libraries, 2024) Larson, James, author; Hurrell, James, advisor; Thompson, David, advisor; Willis, Megan D., committee memberOceanic western boundary currents play a crucial role in transporting heat poleward, thereby influencing the midlatitude climatological-mean climate and serving as an important role for midlatitude storm tracks that provide rainfall to land regions. It is not yet firmly established what role these oceanic currents play in influencing atmospheric variability. Characterized by the presence of mesoscale features such as oceanic eddies and sharp sea surface temperature (SST) gradients, the western boundary currents define a uniquely separate regime for air-sea interactions on climatic timescales relative to the rest of the ocean basins. In this study, simple but robust observational and modeling evidence reveals that anomalous precipitation and vertical motion co-vary with local SST anomalies in the western boundary currents, with a measurable influence extending into the upper troposphere. Periods of anomalously warm SSTs are associated with anomalous, co-located upward motion of > 0.02 Pa/s and precipitation anomalies of ~0.6 mm/day when averaged over a month. Yet, the standard resolution of most climate models, with grid cells on the order of 100 kilometers, fail to capture this co-variability. It is demonstrated that sharpening the horizontal resolution in both a climate model and in atmospheric reanalyses alters the spatial patterns both of sea surface temperature and of regional atmospheric processes. Given the significant influence of these western boundary currents on the broader regions surrounding them, climate projections conducted with grid cells coarser than 50 kilometers may overlook crucial processes.Item Embargo Changes in shortwave solar radiation under local and transported wildfire smoke plumes: implications for agriculture, solar energy, and air quality applications(Colorado State University. Libraries, 2024) Corwin, Kimberley A., author; Fischer, Emily, advisor; Pierce, Jeffrey, committee member; Chiu, Christine, committee member; Corr-Limoges, Chelsea, committee member; Burkhardt, Jesse, committee memberThe emission and transport of pollutants from wildfires is well-documented, particularly at the surface. However, smoke throughout the atmospheric column affects incoming shortwave solar radiation with potentially wide-ranging consequences. By absorbing and scattering light, smoke changes the amount and characteristics of shortwave radiation–a resource that controls plant photosynthesis, solar energy generation, and atmospheric photochemical reactions. In turn, these influence ecological systems as well as air quality and human health. This dissertation examines how wildfire smoke alters boundary layer and surface-level shortwave radiation in ways that are relevant for agricultural, energy, and air quality applications. First, I present an analysis of smoke frequency and smoke-driven changes in the total and diffuse fraction (DF) of photosynthetically active radiation (PAR; 400-700 nm) at the surface. I compare PAR and PAR DF on smoke-impacted and smoke-free days during the agricultural growing season from 2006 to 2020 using data from 10 ground-based radiation monitors and satellite-derived smoke plume locations. I show that, on average, 20% of growing season days are smoke-impacted and that smoke prevalence has increased over time (r = 0.60, p < 0.05). Smoke frequency peaks in the mid to late growing season (i.e., July, August), particularly over the northern Rocky Mountains, Great Plains, and Midwest. I find an increase in the distribution of PAR DF on smoke-impacted days, with larger increases at lower cloud fractions. On clear-sky days, daily average PAR DF increases by 10 percentage points when smoke is present. Spectral analysis of clear-sky days shows smoke increases DF (average: +45%) and decreases total irradiance (average: −6%) across six wavelengths measured from 368 to 870 nm. Optical depth measurements from ground and satellite observations both indicate that spectral DF increases and total spectral irradiance decreases with increasing smoke plume optical depth. My analysis provides a foundation for understanding smoke's impact on PAR, which carries implications for agricultural crop productivity under a changing climate. Second, I examine smoke's impact on two key measures used to assess a location's baseline solar resource availability for solar energy production: direct normal (DNI) and global horizontal (GHI) irradiance. I quantify smoke-driven changes in DNI and GHI at different spatial and temporal scales across the contiguous U.S. (CONUS) using radiative transfer model output and satellite-based smoke, aerosol, and cloud observations. Importantly, I expand the scale of previous studies on smoke and solar energy by including areas primarily affected by dilute, aged, transported smoke plumes in addition to areas with dense, fresh, local smoke plumes. I show that DNI and GHI decrease as smoke frequency increases at the state, regional, and national scale. DNI is more sensitive to smoke with sizable losses persisting downwind of fires. Although large reductions in GHI are possible close to fires, mean GHI declines minimally (< 5%) due to transported smoke. Overall, GHI–the main resource used for photovoltaic energy production–remains a relatively stable resource across most of CONUS even in extreme fire seasons, which is promising given U.S. solar energy goals. Third, I investigate smoke-driven changes in surface-level and boundary layer downwelling actinic flux (F↓)–a crucial component of determining the rate of photooxidation in the atmosphere. I present a case study of changes in F↓ at 550 nm (process validation) and 380 nm (NO2 photolysis) along a research flight through the California Central Valley during the 2018 Western Wildfire Experiment for Cloud Chemistry, Aerosol Absorption, and Nitrogen (WE-CAN) aircraft campaign. F↓ was measured onboard via the HIAPER Airborne Radiation Package (HARP), and I use the National Center for Atmospheric Research (NCAR) Tropospheric Ultraviolet and Visible (TUV) Radiation Model to compute F↓ under smoke-free and smoke-impacted conditions. Modeling F↓ with TUV facilitates calculating the change in F↓ and provides a means of assessing F↓ at altitudes not sampled by the aircraft, such as the ground. I find that the smoke-impacted F↓ from TUV aligns closely with HARP observations: all modeled fluxes are within 20% of measurements at 550 nm and 85% are within 20% of measurements at 380 nm. The average modeled-to-measured ratios (F ↓550=0.96; F ↓380=0.89) indicate that TUV minorly underestimates the observed F↓. On average, observed F↓380 decreased 26%, 17%, and 9% at 0-0.5 km, 0.5-1 km, and 1-1.5 km, respectively, while TUV estimates larger reductions of 41%, 26%, and 19% at the same altitudes. At the ground-level, I calculate a 47% decrease in F↓380 using TUV, which is likely an upper bound given the model slightly underestimates observations. As wildfire smoke increases with climate change, understanding how smoke aloft changes photochemistry is increasingly important for constraining future air quality.Item Open Access From surface to tropopause: on the vertical structure of the tropical cyclone vortex(Colorado State University. Libraries, 2024) DesRosiers, Alexander J., author; Bell, Michael M., advisor; Barnes, Elizabeth A., committee member; Rasmussen, Kristen L., committee member; Davenport, Frances V., committee memberThe internal vortex structure of a tropical cyclone (TC) influences intensity change. Beneficial structural characteristics that allow TCs to capitalize on favorable environmental conditions are an important determinant as to whether a TC will undergo rapid intensification (RI) or not. Accurately forecasting RI is a significant challenge and past work identified characteristics of radial and azimuthal structure of the tangential winds which favor RI, but vertical structure has received less attention. This dissertation aims to define vertical structure in a consistent manner to improve our understanding of how it influences intensity change in observed and modeled TCs, as well as discern when strong winds are more likely to reach the surface with potential for greater impacts. Part 1 investigates the height of the vortex (HOV) in observed TCs and its potential relationships with intensity and intensification rate. As a TC intensifies, the tangential wind field expands vertically and increases in magnitude. Past work supports the notion that vortex height is important throughout the TC lifecycle. The Tropical Cyclone Radar Archive of Doppler Analyses with Recentering (TC-RADAR) dataset provides kinematic analyses for calculation of HOV in observed TCs. Analyses are azimuthally-averaged with tangential wind values taken along the radius of maximum winds (RMW). A threshold-based technique is used to determine the HOV. A fixed-threshold HOV strongly correlates with current TC intensity. A dynamic HOV (DHOV) metric quantifies vertical decay of the tangential wind normalized to its maximum at lower levels with reduced intensity dependence. DHOV exhibits a statistically significant relationship with TC intensity change with taller vortices favoring intensification. A tall vortex is always present in observed cases meeting a pressure-based RI definition in the following 24-hr period, suggesting DHOV may be useful to intensity prediction. In Part 2, numerical modeling simulations are utilized to discern mechanisms responsible for the observed relationships in Part 1. Vertical wind shear (VWS) can tilt the TC vortex by misaligning the low- and mid-level circulation centers which prevents intensification until realignment occurs. Both observed and simulated TCs with small vortex tilt magnitudes possess DHOV values consistent with those observed prior to RI. In aligned TC intensification, DHOV and intensity have a mutually increasing relationship, indicating the metric provides useful information about vertical structure in both tilted and aligned TCs. Vertical vortex growth during RI is sensitive to internal processes which strengthen the TC warm core in the upper-levels of the troposphere. Comparison of a TC simulated in the presence of a concentrated upper-level jet of VWS to a control simulation in quiescent flow indicates that disruption of intensification in the upper levels limits vortex height and intensity without appreciable low- to mid-level tilt. Part 3 focuses on decay of the TC wind field as it encounters friction near the surface in the planetary boundary layer (PBL). Surface winds are important to operational TC intensity estimation, but direct observations within the PBL are rare. Forecasters use reduction factors formulated with wind ratios (WRs) from winds observed by aircraft in the free troposphere and surface winds. WRs help reduce stronger winds aloft to their expected weaker values at the surface. Asymmetries in the TC wind field such as those induced by storm motion can limit the accuracy of static existing WR values employed in operations. A large training dataset of horizontally co-located wind measurements at flight level and the surface is constructed to train a neural network (NN) to predict WRs. A custom loss function ensures the model prioritizes accurate prediction of the strongest wind observations which are uncommon. The NN can leverage relevant physical relationships from the observational data and predict a surface wind field in real-time for forecasters with greater accuracy than the current operational method, especially in high winds.Item Open Access Climate model error in the evolution of sea surface temperature patterns affects radiation and precipitation projections(Colorado State University. Libraries, 2024) Alessi, Marc J., author; Rugenstein, Maria A.A., advisor; Barnes, Elizabeth A., committee member; Maloney, Eric D., committee member; Willis, Megan D., committee memberAtmosphere-ocean general circulation models (AOGCMs) are the primary tool climate scientists use in predicting the effects of climate change. While they have skill in reproducing global-mean temperature over the historical period, they struggle to replicate recently observed sea surface temperature (SST) trend patterns. In this dissertation, we quantify the impact of potential future model error in SST pattern trends on projections of global-mean temperature and Southwest U.S. (SWUS) precipitation. We primarily use a Green's function (GF) approach to identify which SST regions are most relevant for changes in these variables. Our findings demonstrate significant sensitivity of both global-mean temperature and SWUS precipitation to the pattern of sea surface warming, meaning that a continuation of AOGCM error in SST trend patterns adds uncertainty to climate projections which are currently not accounted for. In Chapter 1, we quantify the relevance of future model error in SST to global-mean temperature projections through convolving a GF with physically plausible SST pattern scenarios that differ from the ones AOGCMs produce by themselves. We find that future model error in the pattern of SST has a significant impact on projections, such as increasing total model uncertainty by 40% in a high-emissions scenario by 2085. A reversal of the current cooling trend in the East Pacific over the next few decades could lead to a period of global-mean warming with a 60% higher rate than currently projected. These SST pattern scenarios work through a destabilization of the shortwave cloud feedback to affect temperature projections. In Chapter 2, we focus on near-term projections of precipitation in the SWUS. The observed decrease in SWUS precipitation since the 1980s and heightened drought conditions since the 2000s have been linked to a cooling sea surface temperature (SST) trend in the Equatorial Pacific. Notably, climate models fail to reproduce this observed SST trend, and they may continue doing so in the future. In this chapter, we assess the sensitivity of SWUS precipitation projections to future SST trends using a GF approach. Our findings reveal that a slight redistribution of SST leads to a wetting or drying of the SWUS. A reversal of the observed cooling trend in the Central and East Pacific over the next few decades would lead to a period of wetting in the SWUS. In Chapter 3, we analyze SWUS precipitation sensitivity to SST patterns on long timescales (7+ years) according to a GF approach and a convolutional neural network (CNN) approach. The GF and CNN identify different SST regions as having greater influence on SWUS precipitation: the GF highlights the Central Pacific known from theory to be relevant, while the CNN highlights the South-Central Pacific. To determine if the South-Central Pacific has a physically meaningful and so far overlooked influence on SWUS precipitation, rather than just a statistical relationship, we force an atmosphere-only climate model with an SST anomaly inspired by an Explainable Artificial Intelligence (XAI) method. We find that SSTs in the South-Central Pacific influence SWUS precipitation through an atmospheric bridge dynamical pathway, justifying the CNN's sensitivity physically. The fact that we cannot fully trust the evolution of SST patterns in AOGCMs has many implications for the field of climate science and for how the world's governments and organizations respond to global warming. It is critical for climate change adaptation and mitigation assessments to consider this previously unaccounted for uncertainty in climate projections. Climate scientists can do this by developing SST pattern storylines based on theory, observations, and our understanding of the ocean-atmosphere system. If we fail to communicate known uncertainties for both global-mean and regional projections, the world could lose faith in the climate science community, resulting in less of a global response to climate change.