Browsing by Author "Cooley, Daniel S., committee member"
Now showing 1 - 8 of 8
Results Per Page
Sort Options
Item Open Access A new post-processing paradigm? Improving high-impact weather forecasts with machine learning(Colorado State University. Libraries, 2018) Herman, Gregory Reid, author; Schumacher, Russ S., advisor; Barnes, Elizabeth A., committee member; van den Heever, Susan C., committee member; Cooley, Daniel S., committee member; Hamill, Thomas M., committee memberHigh-impact weather comes in many different shapes, sizes, environments, and storm types, but all pose threats to human life, property, and the economy. Because of the significant societal hazards inflicted by these events, having skillful forecasts of the risks with sufficient lead time to make appropriate precautions is critical. In order to occur, these extreme events require a special conglomeration of unusual meteorological conditions. Consequently, effective forecasting of such events often requires different perspectives and tools than routine forecasts. A number of other factors make advance forecasts of rare, high-impact weather events particularly challenging, including the lack of sufficient resolution to adequately simulate the phenomena dynamically in a forecast model; model biases in representing storms, and which often become increasingly pronounced in extreme scenarios; and even difficulty in defining and verifying the high-impact event. This dissertation systematically addresses these recurring challenges for several types of high-impact weather: flash flooding and extreme rainfall, tornadoes, severe hail, and strong convective winds. For each listed phenomenon, research to more concretely define the current state of the science in analyzing, verifying, and forecasting the phenomenon. From there, in order to address the aforementioned persistent limitations with forecasting extreme weather events, machine learning-based post-processing models are developed to generate skillful, calibrated probabilistic forecasts for high-impact weather risk across the United States. Flash flooding is a notoriously challenging forecast problem. But the challenge is rooted even more fundamentally with difficulties in assessing and verifying flash flooding from observations due to the complex combination of hydrometeorological factors affecting flash flood occurrence and intensity. The first study in this dissertation investigates the multi-faceted flash flood analysis problem from a simplified framework considering only quantitative precipitation estimates (QPEs) to assess flash flood risk. Many different QPE-to-flash flood potential frameworks and QPE sources are considered over a multi-year evaluation period and QPE exceedances are compared against flash flood observations and warnings. No conclusive "best" flash flood analysis framework is clearly identified, though specific strengths and weaknesses of different approaches and QPE sources are identified in addition to regional differences in optimal correspondence with observations. The next two-part study accompanies the flash flood analysis investigation by approaching forecasting challenges associated with extreme precipitation. In particular, more than a decade of forecasts from a convection-parameterized global ensemble, the National Oceanic and Atmospheric Administration's Second Generation Global Ensemble Forecast System Reforecast (GEFS/R) model, are used to develop machine learning (ML) models for probabilistic prediction of extreme rainfall across the conterminous United States (CONUS) at Days 2 and 3. Both random forests (RFs) and logistic regression models (LR) are developed, with separate models trained for each lead time and for eight different CONUS regions. Models use the spatiotemporal evolution of a host of different atmospheric fields as predictors in addition to select geographic and climatological predictors. The models are evaluated over four years of withheld forecasts. The models, and particularly the RFs, are found to compare very favorably with both raw GEFS/R ensemble forecasts and those from a superior global ensemble produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) both in terms of forecast skill and reliability. The trained models are also inspected to discern what statistical findings are identified through ML. Many of the findings quantify anecdotal knowledge that is already recognized regarding the forecast problem, such as the relative skill of simulated precipitation in areas where extreme precipitation events are associated with large-scale processes well resolved by the GEFS/R compared with areas where extreme precipitation predominantly occurs in association with convection in the warm-season. But more subtle spatiotemporal biases are also diagnosed, including a northern displacement bias in the placement of convective systems and a southern displacement bias in placing landfalling atmospheric rivers. The final extended study shifts weather phenomenon focus from extreme rainfall to severe weather: tornadoes, large hail, and severe convective winds. While both high-impact, the two classes of weather hazards share some commonalities and contrasts. While rainfall is directly forecast by dynamical weather models, most severe weather occurs on too small of spatial scales to be directly simulated by the same models. Consequently, unlike with extreme precipitation, when developing post-processed severe weather forecasts, there is no obvious benchmark for objectively determining whether and how much improvement the post-processing is yielding. A natural alternative, albeit much more stringent, benchmark is operational forecasts produced by human forecasters. Operational severe weather forecasts are produced by the Storm Prediction Center (SPC), but there is limited published verification of their outlooks quantifying their probabilistic skill. In the first part of this study, an extended record SPC severe weather outlooks were evaluated to quantitatively assess the state of operational severe weather forecasting, including strengths and weaknesses. SPC convective outlooks were found to decrease in skill with increased forecast lead time, and were most skillful for severe winds, with the worst performance for tornado outlooks. Many seasonal and regional variations were also observed, with performance generally best in the North and East and worst in the South and especially West. The second part of the study follows similar methodology to the extreme precipitation models, developing RF-based probabilistic forecast models forced from the GEFS/R for Days 1--3 across CONUS, analogous to the format in which SPC produces its convective outlooks. RF properties are inspected to investigate the statistical relationships identified between GEFS/R fields and severe weather occurrence. Like with the extreme precipitation model, RF severe weather forecasts are generated and evaluated from several years of withheld validation cases. These forecasts are compared alongside SPC outlooks and also blended with them to produce a combined forecast. Overall, by statistically quantifying relationships between the synoptic-scale environment and severe weather in a manner consistent with the community's physical understanding of the forecast problems, the RF models are able to demonstrate skill over SPC outlooks at Days 2 and 3, and can be blended with SPC outlooks to enhance skill at Day 1. Overall, multiple high-impact weather phenomena---extreme precipitation and severe weather---are investigated from verification, analysis, and forecasting standpoints. On verification and analysis, foundations have been laid both to improve existing operational products as well as better frame and contextualize future studies. ML post-processing models developed were highly successful in advancing forecast skill and reliability for these hazardous weather phenomena despite being developed from predictors of a coarse, dated dynamical model in the GEFS/R. The findings also suggest adaptability across a wide array of forecast problems, types of predictor inputs, and lead times, raising the possibility of broader applicability of these methods in operational numerical weather prediction.Item Open Access Acoustic tomography of the atmosphere using iterated unscented Kalman filter(Colorado State University. Libraries, 2012) Kolouri, Soheil, author; Azimi-Sadjadi, Mahmood R., advisor; Chong, Edwin K. P., committee member; Cooley, Daniel S., committee memberTomography approaches are of great interests because of their non-intrusive nature and their ability to generate a significantly larger amount of data in comparison to the in-situ measurement method. Acoustic tomography is an approach which reconstructs the unknown parameters that affect the propagation of acoustic rays in a field of interest by studying the temporal characteristics of the propagation. Acoustic tomography has been used in several different disciplines such as biomedical imaging, oceanographic studies and atmospheric studies. The focus of this thesis is to study acoustic tomography of the atmosphere in order to reconstruct the temperature and wind velocity fields in the atmospheric surface layer using the travel-times collected from several pairs of transmitter and receiver sensors distributed in the field. Our work consists of three main parts. The first part of this thesis is dedicated to reviewing the existing methods for acoustic tomography of the atmosphere, namely statistical inversion (SI), time dependent statistical inversion (TDSI), simultaneous iterative reconstruction technique (SIRT), and sparse recovery framework. The properties of these methods are then explained extensively and their shortcomings are also mentioned. In the second part of this thesis, a new acoustic tomography method based on Unscented Kalman Filter (UKF) is introduced in order to address some of the shortcomings of the existing methods. Using the UKF, the problem is cast as a state estimation problem in which the temperature and wind velocity fields are the desired states to be reconstructed. The field is discretized into several grids in which the temperature and wind velocity fields are assumed to be constant. Different models, namely random walk, first order 3-D autoregressive (AR) model, and 1-D temporal AR model are used to capture the state evolution in time-space . Given the time of arrival (TOA) equation for acoustic propagation as the observation equation, the temperature and wind velocity fields are then reconstructed using a fixed point iterative UKF. The focus in the third part of this thesis is on generating a meaningful synthetic data for the temperature and wind velocity fields to test the proposed algorithms. A 2-D Fractal Brownian motion (fBm)-based method is used in order to generate realizations of the temperature and wind velocity fields. The synthetic data is generated for 500 subsequent snapshots of wind velocity and temperature field realizations with spatial resolution of one meter and temporal resolution of 12 seconds. Given the location of acoustic sensors the TOA&rsquos are calculated for all the acoustic paths. In addition, white Gaussian noise is added to the calculated TOAs in order to simulate the measurement error. The synthetic data is then used to test the proposed method and the results are compared to those of the TDSI method. This comparison attests to the superiority of the proposed method in terms of accuracy of reconstruction, real-time processing and the ability to track the temporal changes in the data.Item Open Access Environmental controls and aerosol impacts on tropical sea breeze convection(Colorado State University. Libraries, 2020) Park, Jungmin, author; van den Heever, Susan C., advisor; Cooley, Daniel S., committee member; Kreidenweis, Sonia M., committee member; Miller, Steven D., committee member; Rasmussen, Kristen L., committee memberNearly half of the world's human population resides within 150 km of the ocean, and this coastal population is expected to continue increasing over the next several decades. The accurate prediction of convection and its impacts on precipitation and air quality in coastal zones, both of which impact all life's health and safety in coastal regions, is becoming increasingly critical. Thermally driven sea breeze circulations are ubiquitous and serve to initiate and support the development of convection. Despite their importance, forecasting sea breeze convection remains very challenging due to the coexistence, covariance, and interactions of the thermodynamic, microphysical, aerosol, and surface properties of the littoral zone. Therefore, the overarching goal of this dissertation research is to enhance our understanding of the sensitivity of sea breeze circulation and associated convection to various environmental parameters and aerosol loading. More specifically, the objectives are the following: (1) to assess the relative importance of ten different environmental parameters previously identified as playing critical roles in tropical sea breeze convection; and (2) to examine how enhanced aerosol loading affects sea breeze convection through both microphysical and aerosol-radiation interactions, and how the environment modulates these effects. In the first study, the relative roles of five thermodynamic, one wind, and four land/ocean-surface properties in determining the structure and intensity of sea breeze convection are evaluated using ensemble cloud-resolving simulations combined with statistical emulation. The results demonstrate that the initial zonal wind speed and soil saturation fraction are the primary controls on the inland sea breeze propagation. Two distinct regimes of sea breeze-initiated convection, a shallow and a deep convective mode, are also identified. The convective intensity of the shallow mode is negatively correlated by the inversion strength, whereas the boundary layer potential temperature is the dominant control of the deep mode. The processes associated with these predominant controls are analyzed, and the results of this study underscore possible avenues for future improvements in numerical weather prediction of sea breeze convection. The sea breeze circulation and associated convection play an important role in the transport and processing of aerosol particles. However, the extent and magnitude of both direct and indirect aerosol effects on sea breeze convection are not well known. In the second part of this dissertation, the impacts of enhanced aerosol concentrations on sea breeze convection are examined. The results demonstrate that aerosol-radiation-land surface interactions produce less favorable environments for sea breeze convection through direct aerosol forcing. When aerosol-radiation interactions are eliminated, enhanced aerosol loading leads to stronger over-land updrafts in the warm-phase region of the clouds through increased condensational growth and latent heating. This process occurs irrespective of the sea breeze environment. While condensational invigoration of convective updrafts is therefore robust in the absence of aerosol direct effects, the cold-phase convective responses are found to be environmentally modulated, and updrafts may be stronger, weaker, or unchanged in the presence of enhanced aerosol loading. Surface precipitation responses to aerosol loading also appear to be modulated by aerosol-radiation interactions and the environment. In the absence of the aerosol direct effect, the impacts of enhanced aerosol loading may produce increased, decreased, or unchanged accumulated surface precipitation, depending on the environment in which the convection develops. However, when aerosols are allowed to interact with the radiation, a consistent reduction in surface precipitation with increasing aerosol loading is observed, although the environment once again modulated the magnitude of this aerosol-induced reduction.Item Open Access Statistical modeling and computing for climate data(Colorado State University. Libraries, 2019) Hewitt, Joshua, author; Hoeting, Jennifer A., advisor; Cooley, Daniel S., committee member; Wang, Haonan, committee member; Kampf, Stephanie K., committee memberThe motivation for this thesis is to provide improved statistical models and approaches to statistical computing for analyzing climate patterns over short and long distances. In particular, information needs for water managers motivate my research. Statistical models and computing techniques exist in a careful balance because climate data are generated by physical processes that can yield computationally intractable statistical models. Simplified or approximate statistical models are often required for practical data analyses. Critically, model complexity is moderated as much by research needs and available data as it is by computational capabilities. I start by developing a weighted likelihood that improves estimates of high quantiles for extreme precipitation (i.e., return levels) from latent spatial extremes models. In my second project, I develop a geostatistical model that accounts for the influence of remotely observed spatial covariates. The model improves prediction of regional precipitation and related climate variables that are influenced by global-scale processes known as teleconnections. I make the model more accessible by providing an R package that includes visualization, estimation, prediction, and diagnostic tools. The models from my first two projects require estimating large numbers of latent effects, so their implementations rely on computationally efficient methods. My third project proposes a deterministic, quadrature-based computational approach for estimating hierarchical Bayesian models with many hyperparameters, including those from my first two projects. The deterministic method is easily parallelizable and can require substantially less computational effort than common stochastic alternatives, like Monte Carlo methods. Notably, my quadrature-based method can also improve the computational efficiency of other recent, fast, deterministic approaches for estimating hierarchical Bayesian models, such as the integrated nested Laplace approximation (INLA). I also make the quadrature-based method accessible through an R package that provides inference for user-specified hierarchical models. Throughout my thesis, I demonstrate how improved models, more efficient computational methods, and accessible software allow modeling of large, complex climate data.Item Open Access Statistical models for dependent trajectories with application to animal movement(Colorado State University. Libraries, 2017) Scharf, Henry R., author; Hooten, Mevin B., advisor; Cooley, Daniel S., committee member; Fosdick, Bailey K., committee member; Hobbs, N. Thompson, committee memberIn this dissertation, I present novel methodology to study the way animals interact with each other and the landscape they inhabit. I propose two statistical models for dependent trajectories in which depedencies among paths arise from pairwise relationships defined using latent dynamic networks. The first model for dependent trajectories is formulated in a discrete-time framework. The model allows researchers to make inference on a latent social network that describes pairwise connections among actors in the population, as well as parameters that govern the type of behavior induced by the social network. The second model for dependent trajectories is formulated in a continuous-time framework and is motivated primarily by reducing uncertainty in interpolations of the continuous trajectories by leveraging positive dependence among individuals. Both models are used in applications to killer whales. In addition to the two models for multiple trajectories, I introduce a new model for the movement of an individual showing a preference for areas in a landscape near a complex-shaped, dynamic feature. To facilitate estimation, I propose an approximation technique that exploits of locally linear structure in the feature of interest. I demonstrate the model for the movement of an individual responding to a dynamic feature, as well as the approximation technique, in an application to polar bears for which the changing boundary of Arctic sea ice represents the relevant dynamic feature.Item Open Access Stochastic analysis and probabilistic downscaling of soil moisture(Colorado State University. Libraries, 2018) Deshon, Jordan P., author; Niemann, Jeffrey D., advisor; Green, Timothy R., committee member; Cooley, Daniel S., committee memberMany applications require fine-resolution soil-moisture maps that exhibit realistic statistical properties (e.g., spatial variance and correlation). Existing downscaling models can estimate soil-moisture based on its dependence on topography, vegetation, and soil characteristics. However, observed soil-moisture patterns also contain stochastic variations around such estimates. The objectives of this research are to perform a geostatistical analysis of the stochastic variations in soil moisture and to develop downscaling models that reproduce the observed statistical features while including the dependence on topography, vegetation, and soil properties. Extensive soil-moisture observations from two catchments are used for the geostatistical analysis and model development, and two other catchments are used for model evaluation. The Equilibrium Moisture from Topography, Vegetation, and Soil (EMT+VS) model is used to downscale soil moisture, and the difference between the point measurements and the EMT+VS estimates are considered to be the stochastic variations. The stochastic variations contain a temporally stable pattern along with temporally unstable patterns. All of these patterns include spatially correlated and uncorrelated variations. Moreover, the spatial variance of the stochastic patterns increases with the mean moisture content. The EMT+VS model can reproduce the observed statistical features if it is generalized to include stochastic deviations from equilibrium soil moisture, variations in porosity, and measurement errors. It can also reproduce most observed properties if stochastic variations are inserted directly in its soil moisture outputs. These analyses and downscaling models provide insight into the nature of stochastic variations in soil moisture and can be further tested by application to other catchments and larger regions.Item Open Access Uncertainty in measuring seepage from earthen irrigation canals using the inflow-outflow method and in evaluating the effectiveness of polyacrylamide applications for seepage reduction(Colorado State University. Libraries, 2015) Martin, Chad Allen, author; Gates, Timothy K., advisor; Cooley, Daniel S., committee member; Bailey, Ryan T., committee memberSeepage losses from unlined irrigation canals account for a large fraction of the total volume of water diverted for agricultural use, and reduction of these losses can provide significant water quantity and water quality benefits. Quantifying seepage losses in canals and identifying areas where seepage is most prominent are crucial for determining the potential benefits of using seepage reduction technologies and materials. In recent years, polymers have been studied for their potential to reduce canal seepage, and the use of linear-anionic polyacrylamide (PAM) was studied as part of this analysis. To quantify seepage reduction, seepage rates must be estimated before and after application of linear-anionic polyacrylamide (LA-PAM). In this study, seepage rates from four earthen irrigation canals in the Lower Arkansas River Valley (LARV) of southeastern Colorado were estimated with repeated measurements using the inflow-outflow volume balance procedure. It is acknowledged that a significant degree of measurement error and variability is associated with using the inflow-outflow method; however, as is often the case, it was selected so that canal operations were not impacted and so that seepage studies could be conducted under normal flow conditions. To account for uncertainty related to using the inflow-outflow procedure, detailed uncertainty analysis was conducted by assigning estimated probability distribution functions to volume balance components then performing Monte Carlo simulation to calculate possible seepage values with associated probabilities. Based upon previous studies, it was assumed that flow rates could be measured with +/- 5% accuracy, evaporation at +/- 20% accuracy, and water stage within 0.04 to 0.06 feet (all over the 90% interpercentile range). Spatial and temporal variability in canal hydraulic geometry was assessed using field survey data and was incorporated into the uncertainty model, as were temporal variability in flow measurements. Monte Carlo simulation provided a range of seepage rates that could be expected for each inflow-outflow test based upon the pre-defined probable error ranges and probability distribution functions. Using the inflow-outflow method and field measurements directly for assessing variables, deterministic seepage rates were estimated for 77 seepage tests on four canals in the LARV. Canal flow rates varied between 25.8 and 374.2 ft³/s and averaged 127.9 ft³/s, while deterministic estimates of seepage varied between -0.72 and 1.53 (ft³/s) per acre of wetted perimeter with an average of 0.36 (ft³/s)/acre for all 77 tests. Deterministic seepage results from LA-PAM application studies on the earthen Lamar, Catlin, and Rocky Ford Highline canals in southeastern Colorado indicated that seepage could be reduced by 34-35%, 84-100%, and 66-74% for each canal, respectively. Uncertainty analysis was completed for 60 seepage tests on the Catlin and Rocky Ford Highline canals. To describe hydraulic geometry within the seepage test reaches of these canals, canal cross-sections were surveyed at 25 and 16 locations, respectively. Probability distribution functions were assigned to parameters used to estimate wetted perimeter and top width for each cross-section to account for measurement error and spatial uncertainty in hydraulic geometry. Probability distributions of errors in measuring canal flow rates and stage, and in calculating water surface evaporation also were accounted for. From stochastic analysis of these 60 seepage tests, mean values of estimated seepage were between -0.73 (ft³/s)/acre (gain) and 1.53 (ft³/s)/acre, averaging 0.32 (ft³/s)/acre. The average of the coefficient of variation values computed for each of the tests was 240% and the average 90th interpercentile range was 2.04 (ft³/s)/acre. For the Rocky Ford Highline Canal reaches untreated with LA-PAM sealant, mean values of canal seepage rates ranged from -0.26 to 1.09 (ft³/s)/acre, respectively, and averaged 0.44 (ft³/s)/acre. For reaches on the Catlin Canal untreated with LA-PAM, mean values of seepage ranged from 0.02 to 1.53 (ft³/s)/acre, respectively, and averaged 0.63 (ft³/s)/acre. For reaches on the Rocky Ford Highline Canal and Catlin Canal treated with LA-PAM, mean canal seepage rates values ranged from 0.25 to 0.57 (ft³/s)/acre, averaging 0.33 (ft³/s)/acre, and from -0.73 to 0.55 (ft³/s)/acre, averaging -0.01 (ft³/s)/acre, respectively. Comparisons of probability distributions for several pre- and post-PAM inflow-outflow tests suggest likely success in achieving seepage reduction with LA-PAM. Sensitivity analysis indicates that while the major effect on seepage uncertainty is error in measured flow rate at the upstream and downstream ends of the canal test reach, but that the magnitude and uncertainty of storage change due to unsteady flow also is a significant influence. Based upon the findings, recommendations for future seepage studies were provided, which have the ability to account for and reduce uncertainty of inflow-outflow measurements.Item Open Access Using slicing techniques to support scalable rigorous analysis of class models(Colorado State University. Libraries, 2015) Sun, Wuliang, author; Ray, Indrakshi, advisor; Bieman, James M., committee member; Malaiya, Yashwant K., committee member; Cooley, Daniel S., committee memberSlicing is a reduction technique that has been applied to class models to support model comprehension, analysis, and other modeling activities. In particular, slicing techniques can be used to produce class model fragments that include only those elements needed to analyze semantic properties of interest. However, many of the existing class model slicing techniques do not take constraints (invariants and operation contracts) expressed in auxiliary constraint languages into consideration when producing model slices. Their applicability is thus limited to situations in which the determination of slices does not require information found in constraints. In this dissertation we describe our work on class model slicing techniques that take into consideration constraints expressed in the Object Constraint Language (OCL). The slicing techniques described in the dissertation can be used to produce model fragments that each consists of only the model elements needed to analyze specified properties. The slicing techniques are intended to enhance the scalability of class model analysis that involves (1) checking conformance between an object configuration and a class model with specified invariants and (2) analyzing sequences of operation invocations to uncover invariant violations. The slicing techniques are used to produce model fragments that can be analyzed separately. An evaluation we performed provides evidence that the proposed slicing techniques can significantly reduce the time to perform the analysis.