The semiarid region between the North African Savanna and Sahara Desert, known as the Sahel, experienced dramatic multidecadal precipitation variability in the 20th century that was unparalleled in the rest of the world, including devastating droughts and famine in the early 1970s and 80s. Accurate predictions of this region’s hydroclimate future are essential to avoid future disasters of this kind, yet simulations from state of the art general circulation models (GCMs) do a poor job of simulating past Sahel rainfall variability, and don’t even agree on whether future precipitation will increase or decrease under global warming. Furthermore, climate scientists are still not in agreement about whether anthropogenic emissions played an important role relative to natural variability in dictating past Sahel rainfall change. Because the climate system is complex and coupled, it is difficult to determine which processes should be considered causal drivers of circulation changes and which should be considered part of the climate response, and therefore many theories for monsoon rainfall variability coexist in the literature. It is difficult to evaluate these competing theories because observational studies generally cannot be interpreted causally, but simulated experiments may not represent the dynamics of the real world. The Coupled Model Intercomparison Project (CMIP) provides a wealth of data in which GCMs maintained at research institutions worldwide perform similar experiments, allowing the researcher to reach conclusions that are robust to differences in parameterization between GCMs. The scientific community has been using a wide range of statistical techniques to analyze this data, and each has notable limitations. This dissertation explores two statistical techniques for leveraging CMIP to explore the drivers and mechanisms of historical Sahel rainfall variability: analysis of ensemble-mean responses to prescribed variables, and causal inference. In ‎Chapter 1, we give an overview of the climatology and variability of Sahel rainfall and present relevant physical theory. In ‎Chapter 2, we examine the roles of various types of anthropogenic forcing in observations and coupled simulations, using a 3-tiered multi-model mean (MMM) to extract robust climate signals from CMIP phase 5 (CMIP5). We examine “20th century” historical and single-forcing simulations—which separate the influence of anthropogenic aerosols, greenhouse gases (GHG), and natural radiative forcing on global coupled ocean-atmosphere system, and were specifically designed for attribution studies—as well as pre-Industrial control simulations, which only contain unforced internal climate variability, to investigate the drivers of simulated Sahel precipitation variability. The comparison of single-forcing and historical simulations highlights the importance of anthropogenic and volcanic aerosols over GHG in generating forced Sahel rainfall variability that reinforces the observed pattern, with anthropogenic aerosols alone responsible for the low-frequency component of simulated variability. However, the forced MMM only accounts for a small fraction of observed variance. A residual consistency test shows that simulated internal variability cannot explain the residual observed multidecadal variability, and points to model deficiency in simulating multidecadal variability in the forced response, internal variability, or both. In ‎Chapter 3, we investigate the causes for discrepancies in low-frequency Sahel precipitation variability between these ensembles and for model deficiency in reproducing observations. In the most recent version of CMIP – phase 6 of the Coupled Model Intercomparison Project (CMIP6) – the differences between observed and simulated variability are amplified rather than reduced: CMIP6 still grossly underestimates the magnitude of low-frequency variability in Sahel rainfall, but unlike CMIP5, historical mean precipitation in CMIP6 does not even correlate with observed multi-decadal variability. We continue to use a MMM to extract robust climate signals from simulations, but now additionally include sea surface temperature (SST) as a mediating variable in order to test the proposed physical processes. This partitions all influences on Sahel precipitation variability into five components: (1) teleconnections to SST; (2) atmospheric and (3) oceanic variability internal to the climate system; (4) the SST response to external radiative forcing; and (5) the “fast” (not mediated by SST) precipitation response to forcing. Though the coupled simulations perform quite poorly, in a vast improvement from previous ensembles, the CMIP6 atmosphere-only ensemble is able to reproduce the full magnitude of observed low-frequency Sahel precipitation variance when observed SST is prescribed. The high performance is due entirely to the atmospheric response to observed global SST – the fast response to forcing has a relatively small impact on Sahel rainfall, and only lowers the performance of the ensemble when it is included. Using the previously-established North Atlantic Relative Index (NARI) to approximate the role of global SST, we estimate that the strength of simulated teleconnections is consistent with observations. Applying the lessons of the atmosphere-only ensemble to coupled settings, we infer that both coupled CMIP ensembles fail to explain low-frequency historical Sahel rainfall variability mostly because they cannot explain the observed combination of forced and internal variability in SST. Though the fast response is small relative to the simulated response to observed SST variability, it is influential relative to simulated SST variability, and differences between CMIP5 and CMIP6 in the simulation of Sahel precipitation and its correlation with observations can be traced to differences in the simulated fast response to forcing or the role of other unexamined SST patterns. In this chapter, we use NARI to approximate the role of global SST because it is considered by some to be the best single index for estimating teleconnections to the Sahel. However, we show that NARI is only able to explain half of the high-performing simulated low-frequency Sahel precipitation variability in the atmospheric simulations with prescribed global SST. Statistical techniques commonly applied in the literature cannot distinguish between correlation and causality, so we cannot analyze the response of Sahel rainfall to global SST in more depth without atmospheric CMIP simulations targeted at every ocean basin of interest or a new method. In ‎Chapter 4, we turn to a novel technique called causal inference to qualify the notion that NARI can adequately represent the role of global SST in determining Sahel rainfall. We apply a causal discovery algorithm to CMIP6 pre-Industrial control simulations to determine which ocean basins influence Sahel rainfall in individual GCMs. Though we find that state of the art causal discovery algorithms for time series still struggle with data that isn’t generated specifically for algorithm evaluation, we robustly find that NARI does not mediate the full effect of global SST variability on Sahel rainfall in any of the climate simulations. This chapter lays the foundation for future work to fully-characterize the dependence of Sahel precipitation on individual ocean basins using the non-targeted simulations already available in CMIP – an approach which can be validated by comparing the composite results to the interventional historical simulations that are available. Furthermore, we hope this chapter will guide algorithm improvement efforts that are needed to increase the performance and usefulness of time series causal discovery algorithms on climate data.

No description available.

No description available.

The design of an I/O Controller capable of processing data in real time in a tactical training simulator is presented. The controller consists of two microprocessor systems that communicate with peripherals by means of programmed I/O, and with the host computer by Direct Memory Access (DMA) and a serial RS232 link. This thesis addresses both the hardware and software aspects of the controller design.

While attempting to solve optimization problems, the lack of an explicit mathematical expression of the problem may preclude the application of the standard methods of optimization which prove valuable in an analytical framework. In such situations, computer simulations are used to obtain the mean response values for the required settings of the independent variables. Procedures for optimizing on the mean response values, which are in turn obtained through computer simulation experiments, are called simulation-optimization techniques. The focus of this work is on the simulation-optimization technique of response surface methodology (RSM). RSM is a collection of mathematical and statistical techniques for experimental optimization. Correlation induction strategies can be employed in RSM to achieve improved statistical inferences on experimental designs and sequential experimentations. Also, the search procedures currently employed by RSM algorithms can be improved by incorporating gradient deflection methods. This dissertation has three major goals: (a) develop analytical results to quantitatively express the gains of using the common random number (CRN) strategy of variance reduction over direct simulation (independent streams or IS strategy) at each stage RSM, (b) develop a new RSM algorithm by incorporating gradient deflection methods in existing RSM algorithms, and (c) to conduct extensive empirical studies to quantify: (i) the use of eRN strategy over direct simulation in a standard RSM algorithm, and (ii) the gains of the new RSM algorithm over a standard existing RSM algorithm. / Ph. D.

Simulation is one of the most important tools for system performance evaluation in communication networks as well as many other areas. However, simulation is computationally intensive. A traditional sequential simulation of a complex model or a rare-event system may require days or even weeks of computer execution time. Therefore, simulation often becomes a bottleneck of a performance study. With the growing availability of multiprocessor computing systems (e.g., tightly-coupled parallel computers or distributed networks of workstations), parallel simulation, which parallelizes a simulation program for execution on multiple processors, becomes an attractive means to reduce simulation execution time. With few exceptions, existing parallel simulation algorithms can be broadly classified into four methods: multiple replication, time-parallel, parallel regenerative, and space-parallel. Each method is associated with some advantages and limitations. We study these methods and propose a number of parallel simulation algorithms for a class of communication network systems modeled by queueing systems. In multiple replication simulation, each processor simulates a replication of the target simulation model independently. Due to the lack of a prior: knowledge about the steady state conditions, an arbitrarily selected initial state is often used for each simulation run. This can result in significant bias in the simulation outcome. To reduce this initial transient bias, we propose a polling initialization technique, in which a pilot simulation is used to find 'good' initial states that are representative of the steady-state conditions. Time-parallel simulation obtains parallelism by partitioning the time domain of the simulation model into a number of batches. Each batch is computed by a processor independently. Time-parallel simulation has not been fully explored by the research community partly because finding the exact initial states for the batches is often challenging and problem dependent. We develop two approximation time-parallel simulation algorithms for acyclic networks of loss G/G/1/K and G/D/1/K queues. These algorithms exploit unbounded parallelism and can achieve near-linear speedup when the number of arrivals is large. Two other time-parallel approaches are also proposed for Markov chains. For more general simulation models, an approximation approach that uses a substate matching technique is presented. Parallel regenerative simulation exploits parallels in by partitioning the simulation trajectory into a number of regeneration cycles. The amount of parallelism relies on the regeneration frequency of the model. In practice a regeneration state that has a short expected regeneration cycle length often does not exist in the target simulation model. As a result, a sufficient number of observations can not be obtained in a finite simulation interval. To overcome this constraint, we propose a partial regeneration algorithm that uses a substate matching technique to increase the number of observations. When the memory requirement of the target simulation models exceeds the storage capacity of a single processor, space-parallel simulation is an appropriate method. In space-parallel simulation, the target simulation model is decomposed into a number of components such that each component contains a disjoint subset of the model state variables. Each component is mapped into a logical process which is responsible for computing the trajectory corresponding to the component over the simulation time interval. An important class of space-parallel simulation is the conservative simulation, in which each logical process can proceed processing an event only if the process ensures that no event. will arrive later with a smaller timestamp. A number of previous experimental studies have suggested that lookahead, a capability that allows a simulation to look into the simulation time future, plays an important role in the performance of the conservative simulation. Although the performance of conservative simulation has been the interest in many previous studies, there has been a lack of formal arguments to quantify the impact of lookahead to conservative simulation performance. To address this question, we develop stochastic models to study the relationship between the amount of lookahead and the simulation performance with respect to different model topologies. We show that for closed simulation models, the simulation execution time is proportional to the amount of lookahead. For open models, on the other hand, lookahead is effectively useless when the simulation length is long. / Ph. D.

This thesis provides statistical analysis methods and a validation procedure for conducting this statistical analysis, under the common random number (CRN) correlation-induction strategy. The proposed statistical analysis provides estimates for the unknown parameters that are needed for validating the model. While conducting this statistical analysis, we make some key assumptions. Validation comprises of a three-stage statistical procedure. The first stage tests for the multivariate normality,the second stage tests the structure of the covariance matrix between responses, and the third stage tests for the adequacy of the proposed model. The statistical analysis and validation procedures are illustrated with an example of a hospital simulation study. / Master of Science

The overall objective of this research is the development of a set of guidelines, or constructs, which will assist in the formulation of a simulation language for bulk material transportation systems. The formulation of these guidelines necessitated a thorough analysis of two particular areas; these being the simulation analysis procedure and bulk material transportation systems. For comprehension of the nature of simulation languages, Part II presents each step of the analysis procedure as examined from the language's perspective. Supporting this analysis, Part III presents a detailed review of selected simulation languages which are currently available to the systems analyst. Bulk material transportation systems are presented in Part IV. These systems are discussed in detail from the viewpoint of the mode of transportation, the transportation medium, and the type of bulk materials transported. The functional specifications, or constructs, for a bulk material transportation simulation language are presented in Part V. These specifications are categorized according to the following areas; the system being described, the language form, and computer considerations. Utilizing these constructs a simulation language may be developed for subsequent use by bulk material transportation systems analysts which shall be a more appropriate choice for simulating their systems than any language currently available to them. / M.S.

M.S.

M.S.