Browsing by Author "Ebert-Uphoff, Imme, committee member"

Now showing 1 - 10 of 10

Open Access
Assessing the state-dependency of infrared satellite precipitation errors
(Colorado State University. Libraries, 2022) Goldenstern, Eric, author; Kummerow, Christian, advisor; Chiu, Christine, committee member; Ebert-Uphoff, Imme, committee member
The sensing and prediction of precipitation remains at the forefront of weather forecasting, building upon centuries of measurement and study. While in-situ and ground-based methodologies such as rain gauges and weather radars provide the best assessments of precipitation, they are prone to sampling issues and coverage gaps both over challenging terrain and in developing areas of the world. As a result, the use of remote sensing methodologies, namely satellites, have allowed for the expansion of precipitation measurement to encompass nearly the entire Earth. However, unlike rain gauges, satellites are incapable of directly sensing precipitation; rather, they must infer it from the spectral information that can be captured from space through a mathematical framework known as a retrieval. While satellite precipitation retrievals are a boon to the meteorological community due to their ability to fill in these coverage gaps, their indirect nature inevitably gives rise to errors in the measurements themselves. Furthermore, these errors have historically been specific to their training area and are not directly comparable to the errors in other areas. Therefore, this thesis aims to begin disentangling these errors into more generalizable metrics through known information about the measurements themselves and the environmental state being observed. To do this, a neural-network style retrieval algorithm was developed using infrared and lightning data from the Geostationary Operational Environmental Satellite – 16 (GOES-16) to create a validation statistics study. The error from this retrieval, selected to be its bias statistic, was then analyzed both in the context of the satellite data and ancillary meteorological data. From these analyses, it was shown that an understanding of the satellite data allows for limited reproducibility of the retrieval bias tendencies across multiple areas of study, and that ancillary environmental information can shed additional light on how these errors are influenced by the underlying meteorological state. Though this thesis does not create an exact, quantitative methodology for such an assessment, it does provide a direction in which a framework can be established to predict precipitation uncertainties for a more global perspective.
Open Access
Design of a gait acquisition and analysis system for assessing the recovery in a classical murine model of Parkinson's disease
(Colorado State University. Libraries, 2015) Damale, Pranav, author; Chong, Edwin K. P., advisor; Tjalkens, Ronald, committee member; Ebert-Uphoff, Imme, committee member
Gait deficits are important clinical symptoms of Parkinson's disease (PD). Data focusing on gait can be used to measure recovery of motor impairments in rodents with systemic dopamine depletion. This thesis presents a design for a gait acquisition and analysis system able to capture paw strikes of a mouse, extract their positions and timing data, and report quantitative gait metrics to the operator. These metrics can then be used to evaluate the gait changes in mice. This work presents the design evaluation of the system, from initial cellphone captured video concepts through prototyping and testing to the final implementation. The system utilizes a GoPro camera, optimally lit walkway design, image processing techniques to capture footfalls, and algorithms for their quantitative assessment. The results gained from live animal study with methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP)-induced murine model of PD and treated with 1,1-bis(3'-indolyl)-1-(p-chlorophenyl)methane (C-DIM12) are presented, and it is shown how the quantitative measurements can be used to determine healthy, injured, and recovering gait.
Open Access
From neuro-inspired attention methods to generative diffusion: applications to weather and climate
(Colorado State University. Libraries, 2024) Stock, Jason, author; Anderson, Chuck, advisor; Ebert-Uphoff, Imme, committee member; Krishnaswamy, Nikhil, committee member; Sreedharan, Sarath, committee member
Machine learning presents new opportunities for addressing the complexities of atmospheric science, where high-dimensional, sparse, and variable data challenge traditional methods. This dissertation introduces a range of algorithms, motivated specifically by the intricacies of weather and climate applications. These challenges complement those that are fundamental in machine learning, such as extracting relevant features, generating high-quality imagery, and providing interpretable model predictions. To this end, we propose methods to integrate adaptive wavelets and spatial attention into neural networks, showing improvements on tasks with limited data. We design a memory-based model of sequential attention to expressively contextualize a subset of image regions. Additionally, we explore transformer models for image translation, with an emphasis on explainability, that overcome the limitations of convolutional networks. Lastly, we discover meaningful long-range dynamics in oscillatory data from an autoregressive generative diffusion model---a very different approach from the current physics-based models. These methods collectively improve predictive performance and deepen our understanding of both the underlying algorithmic and physical processes. The generality of most of these methods is demonstrated on synthetic data and classical vision tasks, but we place a particular emphasis on their impact in weather and climate modeling. Some notable examples include an application to estimate synthetic radar from satellite imagery, predicting the intensity of tropical cyclones, and modeling global climate variability from observational data for intraseasonal predictability. These approaches, however, are flexible and hold potential for adaptation across various application domains and data modalities.
Open Access
GREMLIN: GOES radar estimation via machine learning to inform NWP
(Colorado State University. Libraries, 2023) Hilburn, Kyle Aaron, author; Miller, Steven D., advisor; Kummerow, Christian D., committee member; Barnes, Elizabeth A., committee member; Ebert-Uphoff, Imme, committee member; Alexander, Curtis R., committee member
Imagery from the Geostationary Operational Environmental Satellite (GOES) has been a key element of U.S. operational weather forecasting since 1975. The latest generation, the GOES-R Series, offers new capabilities to support the need for high-resolution rapidly refreshing imagery for situational awareness. Despite the well demonstrated value to human forecasters, usage of GOES imagery in data assimilation (DA) for initializing numerical weather prediction (NWP) has been limited, particularly in cloudy and precipitating scenes. By providing a rich and powerful library of nonlinear statistical tools, artificial intelligence (AI) / machine learning (ML) enables new approaches for connecting models and observations. The objective of this research is to develop techniques for assimilating GOES-R Series observations in precipitating scenes for the purpose of improving short-term convective-scale forecasts of high-impact weather hazards. The hypothesis of this dissertation is that by harnessing the power of ML, the new GOES-R capabilities can be used to create equivalent radar reflectivity suitable for initializing convection in high-resolution NWP models. Chapter 1 will present a proof-of-concept that ML can be used as an observation operator for GOES-R to simulate Multi-Radar Multi-Sensor (MRMS) composite reflectivity data and thereby initialize convection in NOAA's Rapid Refresh and High-Resolution Rapid Refresh (RAP/HRRR). Development of the GREMLIN (GOES Radar Estimation via Machine Learning to Inform NWP) convolutional neural network (CNN) will be described. This includes the creation of a hierarchy of open source datasets, and will emphasize the importance of the neural network loss function in focusing the attention of the network on the most important meteorological features. Explainable AI (XAI) tools are applied to GREMLIN to discover three primary strategies employed by the network in making predictions, highlighting the unique ability of CNNs to utilize spatial context in satellite imagery. The results of retrospective Rapid Refresh Forecast System (RRFS) forecasts will be described, which show that GREMLIN can produce more accurate short-term forecasts than using real radar data over areas of the U.S. with poor radar coverage. In Chapter 2, the Interpretable GREMLIN model is developed to elucidate the nature of the spatial context utilized by CNNs to make accurate predictions. This clarity is accomplished by moving the inner workings of the CNN out into a feature engineering step and replacing the neural network with a linear regression model. This exposes the effective input space of the CNN and establishes well defined relationships between inputs and outputs, which provides guarantees on how the model will respond to novel inputs. Despite a 24x reduction in the number of trainable parameters, the interpretable model has similar accuracy as the original CNN. Using the interpretable model, five additional physical strategies missed by XAI are discovered. The pros and cons of interpretable model development and implications for generalizability, consistency, and trustworthy AI will be discussed. Finally, Chapter 3 will extend this research for the development of Global GREMLIN, discussing the challenges and opportunities. GREMLIN is validated for regimes outside of the training dataset, and regime dependence is quantified in terms of temperature and moisture. The impacts of additional predictors and advanced ML architectures, and the derivation of uncertainty estimates that will be needed for new DA approaches in RRFS, will be discussed. Current efforts to implement GREMLIN on NOAA's GeoCloud, which will make GREMLIN available to a broader base of users, will be described.
Open Access
Near-cloud aerosol retrieval and three-dimensional radiative transfer using machine learning
(Colorado State University. Libraries, 2021) Yang, Chen-Kuang, author; Chiu, Christine, advisor; Kummerow, Christian D., committee member; Miller, Steven D., committee member; Ebert-Uphoff, Imme, committee member
According to the most recent report of the Intergovernmental Panel on Climate Change, aerosols remain one of the largest sources of uncertainty in estimating and interpreting the Earth's changing energy budget. To reduce the uncertainty, an advanced understanding of aerosol optical properties and aerosol-cloud interaction is needed, which has largely relied on (but is not limited to) passive satellite observations. Current aerosol retrieval methods require a separation between cloud-free and cloudy regions, but this separation is often ambiguous. Three-dimensional (3D) cloud radiative effects can extend beyond the physical boundaries and enhance the reflectance in adjacent cloud-free regions as far as 10 km from clouds. Aerosol optical properties cannot be accurately retrieved without considering the 3D cloud radiative effect in this so-called "twilight" or "transition" zone, which denotes the area between cloud-free and cloudy regions. Indeed, most contemporary retrievals discard these regions, making it impossible to estimate the aerosol radiative effects in this zone. To help break the deadlock, 3D cloud radiative effects must be incorporated into the retrieval methods, and two approaches are proposed in this work, both leveraging machine learning techniques. The first approach accounts for 3D cloud radiative effects by building a 3D shortwave radiative transfer emulator as the forward model for the retrieval methods. Our emulator was trained by cumulus scenes generated from large eddy simulations and radiation fields calculated from 3D radiative transfer, to predict downward and upward flux profiles at a 500 m horizontal resolution and 30 m vertical resolution. From a case drawn from the testing dataset, our emulator captures the spatial pattern of the surface downwelling flux (e.g., shadows and illuminations), and the associated PDF has a remarkable similarity to the synthetic truth. In addition, compared to 1D calculation, our 3D emulator improves the root-mean-square-error by a factor of 6. For the flux and heating rate profiles, our emulator is much superior to the 1D calculation for the cloudy column. The application of this 3D radiative transfer emulator to numerical weather modeling or large-eddy simulations type of model is beyond the scope of the current work to develop an aerosol retrieval algorithm, but the possibility exists to do so. While the promising results from the emulator make it possible to conduct 3D RT retrieval methods, this approach still faces ambiguity in separating cloud-free and cloudy pixels. Here, we present a new retrieval algorithm for aerosol optical depth (AOD) in the vicinity of clouds which contains two unique features. First, it does not require pre-separation of aerosols and clouds. Second, it incorporates 3D radiative effects, allowing us to provide accurate aerosol retrievals near clouds. The AOD retrieval uncertainty of this method in the cloud-free region is (0.0004 ± 4% AOD), which is much better than the (0.03 ± 5% AOD) retrieval uncertainty in NASA Aerosol Robotic Network (AERONET). This method shows skill of predicting AOD over the near-cloud regions, and its validity was confirmed by using one of the explainable artificial intelligence methods to demonstrate that the model's decisions are supported by radiative transfer theory. Finally, a case study using MODIS observations shed light on how this new method can be applied to real world observation, possibly leading to new scientific insight on aerosol structure and aerosol-cloud interaction.
Open Access
On the certainty framework for causal network discovery with application to tropical cyclone rapid intensification
(Colorado State University. Libraries, 2022) DeCaria, Michael, author; van Leeuwen, Peter Jan, advisor; Chiu, Christine, committee member; Barnes, Elizabeth, committee member; Ebert-Uphoff, Imme, committee member
Causal network discovery using information theoretic measures is a powerful tool for studying new physics in the earth sciences. To make this tool even more powerful, the certainty framework introduced by van Leeuwen et al. (2021) adds two features to the existing information theoretic literature. The first feature is a novel measure of relative strength of driving processes created specifically for continuous variables. The second feature consists of three decompositions of mutual information between a process and its drivers. These decompositions are 1) coupled influences from combinations of drivers, 2) information coming from a single driver coupled with a specific number of other drivers (mlinks), and 3) total influence of each driver. To represent all the coupled influences, directed acyclic hypergraphs replace the standard directed acyclic graphs (DAGs). The present work furthers the interpretation of the certainty framework. Measuring relative strength is described thermodynamically. Two-driver coupled influence is interpreted using DAGs, introducing the concept of separability of drivers' effects. Coupled influences are proved to be a type of interaction information. Also, total influence is proved to be nonnegative, meaning the total influences constitute a nonnegative decomposition of mutual information. Furthermore, a new reference distribution for calculating self-certainty is introduced. Finally, the framework is generalized for variables that are continuous with one discrete mode, for which partial Shannon entropy is introduced. The framework was then applied to the rapid intensification of Hurricane Patricia (2015). The hourly change in maximum tangential windspeed was used as the target. The four drivers were outflow layer (OL) maximum radial windspeed (uu), boundary layer (BL) radial windspeed at radius of maximum wind (RMW) (ul), equivalent potential temperature at BL RMW (θe), and the temperature difference between the OL and BL (ΔT). All variables were azimuthally averaged. The drivers explained 45.5% of the certainty. The certainty gain was 35.8% from θe, 24.5% from ΔT, 24.0% from uu, and 15.7% from ul. The total influence of θe came mostly from inseparable effects, while the total influence of uu came mostly from separable effects. Physical mechanisms, both accepted in current literature and suggested from this application, are discussed.
Open Access
The Madden Julian oscillation and tropical-extratropical teleconnections
(Colorado State University. Libraries, 2019) Tseng, Kai-Chih, author; Barnes, Elizabeth A., advisor; Maloney, Eric D., advisor; Randall, David A., committee member; Ebert-Uphoff, Imme, committee member
The Madden Julian Oscillation (MJO) excites strong variations in extratropical circulations that have important implications for subseasonal-to-seasonal (S2S) prediction. In particular, certain MJO phases are characterized by a consistent modulation of geopotential height patterns in the North Pacific and North America. Although the MJO's influence in the downstream weather has been widely explored in previous studies, the relationship between robust MJO teleconnection patterns and model prediction skills has received little attention. In this study, the reanalysis data and ensemble hindcasts from numerical weather forecast system are used to quantify the influence of robust MJO teleconnection on model prediction skills. By calculating the pattern consistency of MJO teleconnection, the ability of MJO convection to modulate extratropical weather is quantified over different time lags and phases. The diagnostic result demonstrates that the robust MJO teleconnection in specific MJO phases/lags are also characterized by excellent agreement in the prediction of geopotential height anomalies across model ensemble members at forecast lead of up to 3 weeks. The mechanisms that lead some MJO phases to have more consistent teleconnections than others are examined by using a linear baroclinic model (LBM). The simulation results show that MJO phases 2, 3, 6 and 7 consistently generate Pacific-North America like (PNA-like) pattern on S2S timescales while other phases do not. By employing a Rossby wave source analysis, the result shows that a dipole-like Rossby wave source patterns on each side of the jet in MJO phase 2, 3, 6 and 7 can increase the pattern consistency of teleconnection due to the constructive interference of similar teleconnection signals. On the other hand, the symmetric patterns of Rossby wave source in other phases can dramatically reduce the pattern consistency due to destructive interference. The consistency of MJO teleconnections is also characterized by an interannual variability. During the El Niño years, the pattern consistency is dramatically decreased compared to the La Niña years. Employing the numerical experiments in LBM and applying a Rossby wave ray tracing algorithm, we demonstrate two factors largely determine the interannual variability of MJO teleconnection consistency. During El Niño years, the eastward extension of subtropical jet and a less-dipole like Rossby wave source pattern on each side of the jet dramatically decrease poleward propagating wave signals. By contrast, the competing effect between these two factors results in modest changes in pattern consistency during La Niña years. Thus, the observed consistency of MJO teleconnections is much smaller during El Niño years than La Niña years. The dynamics associated with the pattern consistency of MJO teleconnection are addressed in the first half of this work. What is still unclear, however, is the importance of the accumulated influence of past MJO activity on these results. To examine the importance of past MJO phases in determining future states of extratropical circulations, a LBM and one of the simplest machine learning algorithm: logistic regression are used. By increasing the complexity of logistic regressions with additional informational about past MJO phases, we show that 15 additional lags before lag 0 play a dominant role in determining the future state of MJO teleconnections. This result is supported by the numerical LBM simulations. We further demonstrate that this 15-day span is characterized by a phase/lead time dependent feature, which is relevant to the dynamics of MJO teleconnections and explained in this work. Ultimately, a particular emphasis is placed on the role of model MJO in influencing the winter climatol- ogy of extratropical circulations. The MJO is known for consistently modulating the extratropical weather. In addition, simulating the MJO continues to be a challenge for many state-of-art climate models, and it is unclear the extent to which these biases in the MJO may cause biases in midlatitude variability. By analyzing 22 climate model simulations from the Coupled Model Intercomparison Project Phase 5 (CMIP5) and the reanalysis data, we demonstrate that one of leading variability of daily geopotential height is associated with MJO activity, and can be identified without prior knowledge of MJO in both observations and CMIP5 data. This shows the dominant role of MJO in modulating extratropical circulations. However, due to this strong relationship between MJO and extratropical circulations, the model biases in the MJO convection is also reflected in the wintertime climatology of extratropical circulations.
Open Access
The role of Earth system interactions in large-scale atmospheric circulation and climate
(Colorado State University. Libraries, 2023) Yook, Simchan, author; Thompson, David W. J., advisor; Ravishankara, A. R., committee member; Hurrell, James, committee member; Ebert-Uphoff, Imme, committee member
The complex interactions among different components of the Earth system play a key role in governing the climate variability through various physical processes. For example, an interaction between the fluctuations in one component of the Earth system and associated variations in another component of the Earth system can either amplify or dampen the climate variability depending on the nature of their two-way feedback mechanisms. Thus, understanding the role of various physical interactions among components of the Earth system is critical to understand the changes in climate as well as to reduce the uncertainty in future climate projections. This dissertation focuses on discovering the key processes and interactions among different components of the Earth system on the climate variability using observations and model hierarchies. In Part 1, the interactions between the atmospheric circulation and western North Pacific SST anomalies are explored in two sets of simulations: 1) a simulation run on a coupled atmosphere-ocean general circulation model (GCM), and 2) a simulation forced with prescribed, time-evolving SST anomalies over the western North Pacific. The results support the interpretation of the observed lead/lag relationships between western North Pacific Sea Surface Temperature (SST) anomalies and the atmospheric circulation, and provide numerical evidence that SST variability over the western North Pacific has a demonstrable effect on the large-scale atmospheric circulation throughout the North Pacific sector. In Part 2, the role of moist lapse rate in altering the temperature variability under climate change is explored. To reduce the complexity of the problem, the changes in the temperature variance under global warming are first analyzed in the simplest version of model hierarchy: a single column Rapid Radiative Transfer Model with a simplified convective adjustment. Similar analyses were repeated with varying model hierarchies with additional complexities: a global general circulation model in global Radiative Convective Equilibrium (RCE) setting with fixed SST, and fully coupled Earth system models. The results highlight the role of moist lapse rate as a potential constraint for climate variability in the tropical atmosphere simulated by different model hierarchies. In Part 3, the effects of coupled chemistry-climate interactions on the amplitude and structure of stratospheric temperature variability are quantified in two numerical simulations: A "free running" simulation that includes fully coupled chemistry-climate interactions; and a "specified chemistry" version of the model forced with prescribed chemical composition. The results indicate that the inclusion of coupled chemistry-climate interactions increases the internal variability of temperature by a factor of ~two in the lower tropical stratosphere through dynamically driven ozone-temperature feedbacks. The results highlight the fundamental role of two-way feedbacks between the atmospheric circulation and chemistry in driving climate variability in the lower stratosphere. In Part 4, the effects of coupled chemistry-climate interactions on the large-scale atmospheric circulation are further explored based on two observational case studies of the Antarctic ozone holes of 2020 and 2021. The 2020 and 2021 were marked by two of the largest Antarctic ozone holes on record. It has been demonstrated that the ozone holes of 2020 and 2021 were associated with large changes in the atmospheric circulation consistent with the climate impacts of Antarctic ozone depletion. The ozone holes were also unusual for their associations with aerosol burdens due to two extraordinary events: the Australian wildfires of early 2020 and the eruption of La Soufriere in 2021. The results provide suggestive evidence that injections of both wildfire smoke and volcanic emissions into the stratosphere can lead to hemispheric-scale changes in surface climate. This dissertation provides a detailed look at the complex aspects of the coupled interactions among different components of the Earth system and their roles on climate variability and large-scale dynamics. To clarify the role of the different physical processes contributing to the climate responses, this study performed a comprehensive analysis based on observations as well as a series of numerical experiments run on different configurations of climate model hierarchies. The findings herein improve our understanding of different Earth system interactions and their influences on global climate and large-scale atmospheric dynamics.
Open Access
Towards using neural networks for geoscientific discovery
(Colorado State University. Libraries, 2020) Toms, Benjamin A., author; Barnes, Elizabeth A., advisor; Ebert-Uphoff, Imme, committee member; Hurrell, James W., committee member; Thompson, David W. J., committee member
How can we use computational methods to extract physically meaningful patterns from geoscientific data? This question has been asked in some form for decades within the geoscientific community, with many landmark discoveries resulting from the novel application of computational methods to a geoscientific dataset. For example, the Madden-Julian Oscillation was discovered through Fourier transforms of tropical time-series, while the defining structures of the Northern Hemispheric annular modes were first captured using principal component analysis. These discoveries rooted in computational methods have since driven decades of geoscientific research and innovation, and are only two of among many similar examples. It is therefore clear that computational science and geoscience are inextricably intertwined, and so the continued advancement of both fields in tandem is beneficial to future geoscientific discovery. Many methods exist to discover patterns within geoscientific data, although each is limited by its own set of assumptions. The most common assumption is that of linearity, which oftentimes conflicts with our understanding that the earth system can be both dynamically and statistically nonlinear. However, a recently popularized subset of methods within the computer science community known as neural networks can identify nonlinear patterns and are therefore potentially powerful tools for geoscientific discovery. Neural networks learn how to map one dataset to another using a combination of nonlinear relationships, and are generalizable to a broad range of tasks including forecasting and identifying patterns within images. Regardless of the application, a common limitation of neural networks has been the difficulty to understand how and why they make their decisions. Therefore, while they have been used in geoscience for more than two decades, they have mostly been applied when accuracy is valued more than understanding, such as for making forecasts. Within this dissertation, we first propose a framework for how neural networks can be used for geoscientific discovery by applying recently invented methods from the computer science community. We focus on methods that explain which aspects of the input dataset are useful for the neural network when making connections to the output dataset. This framework enables physical interpretations of how and why neural networks make decisions, since the geoscientist that designs the neural network is likely familiar with the physical meaning of each input. In the first study of the dissertation, we outline the framework and apply it to two simple tasks to ensure the neural network interpretations abide by our current understanding of the earth system. The interpretable neural networks successfully identify the pattern of the El Niño Southern Oscillation and oceanic patterns that lend seasonal predictability, which lends confidence that the framework is reliable. In the second study, we then further test the methods by applying them to a more spatially and temporally complex oscillation called the Madden-Julian Oscillation (MJO). The interpretable neural networks correctly identify the known spatial structures and seasonality of the MJO, and also suggest that the MJO is nonlinear and expresses its nonlinearity through the uniqueness of each event. The final study assesses whether the proposed framework can be used to identify predictable patterns of earth-system variability within climate models through its application to decadal predictability. We find that the interpretable neural networks identify known modes of oceanic decadal variability that contribute to predictability of continental surface temperatures. The interpretations can also be used to identify distinct regimes of predictability, wherein spatially and temporally unique oceanic modes contribute predictability for the same location at different times. From a broader perspective, these studies suggest that neural networks are a viable tool for geoscientific discovery and are particularly useful given their ability to capture nonlinear, time-evolving patterns. It is likely that new neural network algorithms and methods for their interpretation will continue to be developed by the computer science community, and so this research provides a guideline for how such methods can be gainfully applied within the geosciences.
Open Access
Using mathematical techniques to leverage domain knowledge in image analysis for earth science
(Colorado State University. Libraries, 2023) Ver Hoef, Lander, author; Adams, Henry, advisor; King, Emily J., advisor; Hagman, Jess Ellis, committee member; Ebert-Uphoff, Imme, committee member
When presented with the power of modern machine learning techniques, there is a belief that we can simply let these algorithms loose on the data and see what they can find, unconstrained by human choice or bias. While such approaches can be useful, they are (of course) not fully free of bias or choice. Moreover, by utilizing the deep store of knowledge built up by scientific domains over decades or centuries, we can make architectural choices in our machine learning algorithms that focus the learning on features that we already know are important and informative, leading to more efficient, explainable, and interpretable methods. In this work, we present three examples of this approach. In the first project, to make use of the knowledge that texture is an important attribute of clouds, we use tools from topological data analysis focusing on the texture of satellite imagery, which leads to an effective and highly interpretable classifier of mesoscale cloud organization. This project resulted in a paper that has been published as a journal article. In the second project, we compare a rotationally invariant convolutional neural network against a conventional CNN both with and without data augmentation in their performance and behaviors on the task of predicting the major and minor axes lengths of storms in forecast data. Finally, in the third project, we explore three different techniques from harmonic analysis to enhance the signature of gravity waves in satellite imagery.