Search

Artificial intelligence (AI) and machine learning (ML) present some application opportunities and challenges that can be framed as learning problems. The performance of machine learning models depends on algorithms and the data. Moreover, learning algorithms create a model of reality through learning and testing with data processes, and their performance shows an agreement degree of their assumed model with reality. ML algorithms have been successfully used in numerous classification problems. With the developing popularity of using ML models for many purposes in different domains, the validation of such predictive models is currently required more formally. Traditionally, there are many studies related to model evaluation, robustness, reliability, and the quality of the data and the data-driven models. However, those studies do not consider the concept of the applicability domain (AD) yet. The issue is that the AD is not often well defined, or it is not defined at all in many fields. This work investigates the robustness of ML classification models from the applicability domain perspective. A standard definition of applicability domain regards the spaces in which the model provides results with specific reliability. The main aim of this study is to investigate the connection between the applicability domain approach and the classification model performance. We are examining the usefulness of assessing the AD for the classification model, i.e. reliability, reuse, robustness of classifiers. The work is implemented using three approaches, and these approaches are conducted in three various attempts: firstly, assessing the applicability domain for the classification model; secondly, investigating the robustness of the classification model based on the applicability domain approach; thirdly, selecting an optimal model using Pareto optimality. The experiments in this work are illustrated by considering different machine learning algorithms for binary and multi-class classifications for healthcare datasets from public benchmark data repositories. In the first approach, the decision trees algorithm (DT) is used for the classification of data in the classification stage. The feature selection method is applied to choose features for classification. The obtained classifiers are used in the third approach for selection of models using Pareto optimality. The second approach is implemented using three steps; namely, building classification model; generating synthetic data; and evaluating the obtained results. The results obtained from the study provide an understanding of how the proposed approach can help to define the model's robustness and the applicability domain, for providing reliable outputs. These approaches open opportunities for classification data and model management. The proposed algorithms are implemented through a set of experiments on classification accuracy of instances, which fall in the domain of the model. For the first approach, by considering all the features, the highest accuracy obtained is 0.98, with thresholds average of 0.34 for Breast cancer dataset. After applying recursive feature elimination (RFE) method, the accuracy is 0.96% with 0.27 thresholds average. For the robustness of the classification model based on the applicability domain approach, the minimum accuracy is 0.62% for Indian Liver Patient data at r=0.10, and the maximum accuracy is 0.99% for Thyroid dataset at r=0.10. For the selection of an optimal model using Pareto optimality, the optimally selected classifier gives the accuracy of 0.94% with 0.35 thresholds average. This research investigates critical aspects of the applicability domain as related to the robustness of classification ML algorithms. However, the performance of machine learning techniques depends on the degree of reliable predictions of the model. In the literature, the robustness of the ML model can be defined as the ability of the model to provide the testing error close to the training error. Moreover, the properties can describe the stability of the model performance when being tested on the new datasets. Concluding, this thesis introduced the concept of applicability domain for classifiers and tested the use of this concept with some case studies on health-related public benchmark datasets.

The concept of Shifting the Balance of Care was first introduced to NHS Scotland in 2005 through the Kerr Report. The key messages from the report were to: ensure sustainable and safe local services, which are supported by the right skills, change the emphasis of care into the community, provide preventative reactive care, and fully integrate the system to tackle the changes, use technology more effectively, and involve the public in finding solutions to change. Following the report, a framework was developed which highlighted and prioritised eight areas of improvement. These areas for improvement are the focus by which this research examines if Operational Research (OR), specifically OR models, can have a positive impact in Shifting the Balance of Care. The research utilises underlying OR methodologies and methods and provides evidence from the literature of the ability of nine selected models to facilitate the Shift in the Balance of Care. A contributing factor to the research is the barriers to implementation of OR models into the NHS. With reference to the literature, the common barriers to implementation of OR models are categorised and used to provide direction to modellers where implementation barriers are more prevalent in some models than in others. The research also provides empirical evidence of three selected models’ (the Lean Methodology, Process Mapping and Simulation, developed over two Case Studies) ability to address and influence the prioritised Improvement Areas, with the addition of a newly developed model: SoApt. The development of SoApt follows the Principles of Model Development derived as a guide to modellers who wish to develop a new model. SoApt is also empirically explored in a Case Study and provides some evidence of the models ability to aid Decision-makers, faced with limited budgets, to choose between options which will Shift the Balance of Care. OR methods and methodologies are examined to ascertain the Roles of Models for each model explored in the Case Studies. Examination of the Roles of Models against the Improvement Areas provided evidence of a models’ ability to address more than one of the priority areas and that models can be used together or sequentially. In addition, with reference to OR methods and methodologies, a theoretical Evaluation Framework is proposed which suggests the User and User Satisfaction is key to the evaluation of a model’s success; positive experiences of the User and Use of the model may help to eliminate some of the barriers to implementation.

The dissertation contains essays concerning the linkages between macroeconomy and financial market or the conduct of monetary policy via DSGE modelling. The dissertation contributes to the questions of fitting macroeconomic models to the data, and so contributes to our understanding of the driving forces of fluctuations in macroeconomic and financial variables. Chapter one offers an introduction to my thesis and outlines in detail the main results and methodologies. In Chapter two I introduce a statistical measure for model evaluation and selection based on the full information of sample second moments in data. A model is said to outperform its counterpart if it produces closer similarity in simulated data variance-covariance matrix when compared with the actual data. The "distance method" is generally feasible and simple to conduct. A flexible price two-sector open economy model is studied to match the observed puzzles of international finance data. The statistical distance approach favours a model with dominant role played by the expectational errors in foreign exchange market which breaks the international interest rate parity. Chapter three applies the distance approach to a New Keynesian model augmented with habit formation and backward-looking component of pricing behaviour. A macro-finance model of yield curve is developed to showcase the dynamics of implied forward yields. This exercise, with the distance approach, reiterate the inability of macro model in explaining yield curve dynamics. The method also reveals remarkable interconnection between real quantity and bond yield slope. In Chapter four I study a general equilibrium business cycle model with sticky prices and labour market rigidities. With costly matching on labour market, output responds in a hump-shaped and persistent manner to monetary shocks and the resulting Phillips curve seems to radically change the scope for monetary policy because (i) there are speed limit effects for policy and (ii) there is a cost channel for monetary policy. Labour reforms such as in mid-1980s UK can trigger more effective monetary policy. Research on monetary policy shall pay greater attention to output when labour market adjustments are persistent. Chapter five analyzes the link between money and financial spread, which is oft missed in specification of monetary policy making analysis. When liquidity provision by banks dominates the demand for money from the real economy, money may contain information of future output and inflation due to its impact on financial spreads. I use a sign-restriction Bayesian VAR estimation to separate the liquidity provision impact from money market equilibrium. The decomposition exercise shows supply shocks dominate the money-price nexus in the short to medium term. It also uncovers distinctive policy stance of two central banks. Finally Chapter six concludes, providing a brief summary of the research work as well as a discussion of potential limitations and possible directions for future research.

The valley wind system belongs to the most characteristic circulations of the at- mospheric boundary layer in mountainous terrain. It dominates surface winds at the valley bottom, contributes to the vertical air mass exchange with the free tropo- sphere, and plays a crucial role in the horizontal air mass exchange with neighboring plains and valleys. The accurate prediction of the valley wind system is therefore crucial for forecasting the weather in complex terrain. In this study, we assess the ability of the ICON model to resolve along-valley winds in the Inn Valley, by comparing the simulation to measurements from a char- acteristic valley wind day during the CROSSINN campaign (CROSSINN = Cross- valley flow in the Inn Valley). The three-dimensional observations allow us to eval- uate the simulation beyond classical statistical validation techniques and to further investigate the underlying processes (process-based evaluation). Particularly, the thermal forcing is investigated using radiosounding-derived valley-plain tempera- ture and pressure gradients. Besides the model evaluation, we provide a compre- hensive and detailed overview of the observed development of the valley boundary layer during the selected day (13 September 2019, the 8th intensive observational period (IOP8) of CROSSINN). At 1 km horizontal grid spacing, the ICON model can simulate the daytime up- valley flow well, including its strength, timing, and vertical structure (for the Inn Valley with ridge-to-ridge distance of 10 km or 10 dx). However, during the night the model is not able to correctly represent valley-plain temperature and pressure contrasts, thus deviating from the observed nighttime down-valley flow. We identify three main challenges for the simulation of IOP 8. The first challenge is the observed westerly down-valley flow during the night and morning of IOP 8, which may in- volve a scale interaction with the synoptic westerly flow. The second challenge is the complex interplay between the up-valley wind, the slope circulation, and the cross-valley vortex, which affects the detachment of the up-valley wind from the surface and its transition to higher altitudes.The third challenge pertains to the sta- ble boundary layer and valley cold air pool during the night after IOP 8. The model limitations in representing these processes result in the failure to simulate thermal forcing and the connected down-valley flow in the second night. The main limitations of the model include the mis-representation of synoptic flow (limited predictability), smaller-scale thermally-driven flows (limited resolu- tion), and cold air pools (stable boundary layer issues). In summary, this study contributes to a better understanding of along-valley winds under non-idealized conditions, and provides ICON’s first process-based model evaluation on thermally driven flows.

In model selection, we seek a balance between goodness-of-fit and generalizability, for which model complexity is key. Fitting Propensity (FP) has been suggested as an ideal measure of complexity that refers to a model’s inherent flexibility to fit diverse patterns of data, all else being equal. Assessing the FP of item response theory (IRT) models requires random and uniform sampling of all item response patterns for a set of items and fitting the sampled data to one or more models repeatedly many times. The model fit information across the replications is summarized for each model and examined. In the case of multiple models, comparisons between models are also made. Computational issues due to the high-dimensional discrete space involved in the generation of random datasets have rendered it infeasible to investigate FP for more than a handful of dichotomously scored items under the conventional full information (FI) approach of the multinomial framework.This study turns to limited information (LI) methods as an alternative, capitalizing on the fact that IRT models can be realized as contingency tables using marginal probabilities. LI methods use information from only the lower-order, usually univariate and bivariate, margins of IRT models as opposed to full response patterns. Thus, they not only significantly reduce the number of response probabilities to be generated in the first place but can also make model estimation computationally simpler. The computational gain afforded by the proposed LI approach opens doors for investigating the FP of more complex IRT modeling schemes which traditionally require many more response patterns. A data-generating algorithm founded on classical literature on sampling contingency tables with fixed margins along with sequential importance sampling (SIS) of contingency tables is introduced for random and uniformly distributed sampling across all univariate and bivariate margins of items. To estimate the data consisting of solely the lower-order margins, a pairwise marginal maximum likelihood (PMML) estimator tailored to fit a wide variety of IRT models is introduced. Lastly, the feasibility of the proposed LI data generation algorithm and estimation approach to assess the FP of IRT models is tested under various combinations of data sampling and estimation methods.

Wetland impacts in the Mid-Atlantic USA are frequently mitigated via wetland creation in former uplands. Regulatory approval requires a site-specific water budget that predicts the annual water level regime (hydroperiod). However, many studies of created wetlands indicate that post-construction hydroperiods frequently are not similar to impacted wetland systems. My primary objective was to evaluate a water budget model, Wetbud (Basic model), through comparison of model output to on-site water level data for two created forested wetlands in Northern Virginia. Initial sensitivity analyses indicated that watershed curve number and outlet height had the most leverage on model output. Addition of maximum depth of water level drawdown greatly improved model accuracy. I used Nash-Sutcliffe efficiency (NSE) and root mean squared error (RMSE) to evaluate goodness of fit of model output against site monitoring data. The Basic model reproduced the overall seasonal hydroperiod well once fully parameterized, despite NSE values ranging from -0.67 to 0.41 in calibration and from -4.82 to -0.26 during validation. For RMSE, calibration values ranged from 5.9 cm to 12.7 cm during calibration and from 8.2 cm to 18.5 cm during validation. My second objective was to select a group of "design target hydroperiods" for common Mid-Atlantic USA wetland types. From > 90 sites evaluated, I chose four mineral flats, three riverine wetlands, and one depressional wetland that met all selection criteria. Taken together, improved wetland water budget modeling procedures (like Wetbud) combined with the use of appropriate target hydroperiod information should improve the success of wetland creation efforts.

The land surface is a key component in the earth’s climate system, and is one of the main contributors to uncertainty in climate projections. Fluxes of energy, momentum, water and carbon between the land surface and the atmosphere provide important feedbacks in the climate system. Land surface models (LSMs) have been a key tool in understanding land surface fluxes, the processes that drive them, and their impacts on climate change. Recent work by Best et al. (2015) has demonstrated that the performance of modern LSMs is much worse than expected, and this has acted as a catalyst, forcing the land surface community to re-evaluate models and the methods used to evaluate them. This thesis contributes to this effort by attempting to address the question *how predictable are land surface fluxes* in two ways: by analysing the performance of LSMs, and by addressing the predictability of measured fluxes in the FLUXNET dataset directly using empirical modelling. I first explore the performance of LSMs and show that the Best et al. (2015) results are not due to poor methodology; and further, performance problems appear to be at least partly shared between models. Next, I assess the predictability of land surface fluxes from meteorological variables by using empirical models on a broader set of FLUXNET data. This work pinpoints a number of key variables that provide independent information suitable for predicting fluxes. It also describes a set of empirical models of staged complexity that are suitable as benchmarks for LSMs. Finally, I investigate the variability of predictability within the FLUXNET data, and the degree to which it is determined by site characteristics. Predictability is shown to be affected by site aridity, but is otherwise less determined by site characteristics than expected. This thesis shows that substantial performance gains can be made in land surface predictability, with further gains likely with better exploitation of newly available observational data. The methods described provide a basis for improving the standardisation and rigour of LSM evaluation, which will ultimately lead to improvements to land surface models, and the climate projections that are based on them.

Biological functions that use amino acids (AA) are limited by AA supply. This concept was likened to staves in a barrel, where the shortest stave determines the barrel's ability to hold water (Mitchell and Block, 1946). Inaccuracies in models that predict nutrient supply and requirements of dairy cows result in inefficient feeding, as under-prediction of requirements results in deficiency, and over-prediction results in excess. To avoid limitations in production due to AA deficiencies, protein is fed in quantities that likely exceed requirements. Overfeeding of AA results in increased expenses for producers and increased N excretion to the environment, providing economic and environmental incentives to increase N-efficiency. Work presented in the following chapters evaluated the impact of AA supplementation on milk production in dairy cattle, and evaluated the PREP10 and NRC (2001) nutrient requirement model predictions. In two feeding trials (Chapter 2 and Chapter 3), low protein diets did not result in decreased milk production, indicating that protein requirements were overestimated. Although supplementation of AA did not increase milk production, low protein diets resulted in greater N-efficiency, especially when supplemented with Histidine. Evaluation of the PREP10 and NRC (2001) models (Chapter 4) used production data from the literature to identify deficiencies in prediction equations, and found that correction of model bias would considerably reduce prediction errors. Model inaccuracies affect the inefficiency of dairy cow feeding, and must be evaluated to improve feed efficiency. Such improvements could act synergistically with AA supplementation to more closely match nutrient supply to requirements.

The main objective of this thesis is to evaluate the simulation of heatwaves in climate models and to better understand their relevant physical drivers. The thesis objective is motivated by the desire for climate science to provide more actionable information to decision-makers and planners on climate extremes. This thesis employs multiple state-of-the-art data products. CMIP5 coupled climate models were evaluated in terms of simulating large scale circulation patterns and their representation of heatwaves. Heatwaves were calculated from a gridded observationally-based Australian temperature dataset and compared against heatwaves in climate models. Reanalysis data was included to examine circulation and land surface conditions that drive heatwaves in different regions of Australia. Future projections of these driving conditions were then assessed in multiple climate models to characterize uncertainty. Evaluating the representation of large scale circulation patterns over the Australian region, including the frequency, persistence and transition states of these patterns, revealed a relatively large diversity of model performance across these metrics. The spatial patterns of heatwave trends and climatology in CMIP5 also presented a diverse range of patterns and model skill. There was some association between model skill in representing large scale circulation and skill in representing the heatwave climatology. Models that represented circulation best were typically higher resolution models which also provide a reasonable representation of the heatwave climatology; models that represent circulation poorly were typically low resolution models with a poor representation of the heatwave climatology. Examination of the linkages between large scale circulation patterns and heatwave occurrence in reanalysis revealed that the land surface state often modulated the circulation heatwave association. Analysis of future projections in CMIP5 revealed uncertainty concerning the magnitude of the frequency changes in relevant atmospheric circulation features and concerning the direction of change in relevant land surface conditions. Reducing uncertainty in how these driving conditions are projected to change is necessary to improve the robustness of regional heatwave projections at scales relevant for planning and decision-making.

Atmospheric blocking describes an important part of the weather variability inherent to the mid-latitudes imposing an anomalous strong, quasi stationary anticyclonic circulation on the upper-level flow. Blocking leads to a reversal of the customary upper-level westerly flow and deflects the jet stream as well as migratory weather systems north and southward. Owing to the long persistence and anomalous flow regime, blocking is associated with severe heat waves in summer and cold spells in winter. Despite its vast impact on regional weather and society, average skills in predicting blocking initiation and continuation are observed across many numerical weather prediction models. In this study, the representation of atmospheric blocking in the new general circulation model ICON is investigated by means of a blocking climatology, an analysis of blocking characteristics and a mean state assessment. To this end, a Northern Hemisphere blocking climatology is derived by identifying blocking as regions of anomalous low vertically averaged potential vorticity (PV) values directly below the tropopause. Three 5-year model runs forced with sea surface temperatures and sea ice fields from the ERA-Interim reanalysis for the period of 2001 to 2005 are integrated (AMIP setup). Results from increasing horizontal resolutions (80, 40 and 20 km) are considered and compared to ERA-Interim observations for the period of 1979 to 2014. Three preferred regions of blocking activity are identified, two at the end of the Atlantic and Pacific storm track and a third over northern Russia. Good agreement of the spatial blocking patterns is found, especially in the core region of blocking activity. Main regions of deviation are confined to the mid-latitudes ranging from the Pacific to the Atlantic ocean. Statistics show that blocking duration is overestimated, while blocking size is systematically underestimated in ICON. The atmospheric mean state of the troposphere is generally simulated well, whereas large biases emerge in the stratosphere. A cool bias in the lower stratosphere leads to a PV dipole which is responsible for globally lifting the tropopause north of 50° N. A potential link between the Madden-Julian oscillation and the PV distribution in the Pacific region is suggested, leading to a shift of the customary Pacific PV trough. The strong mean state bias as well as the overestimated duration of blocking in the low resolution configuration are mainly responsible for the mediocre blocking representation in the mid-latitude Pacific sector. An improved mean state with increasing horizontal resolution is observed in the Pacific, resulting in better blocking depiction. However, in the Atlantic sector, no improvement of the 20 km run over the intermediate configuration is found. This points to the importance of different processes in blocking representation such as orographic or diabatic effects in the Atlantic sector.

Wheat yields for New South Wales, Australia simulated by the APSIM agricultural model are used as a case study to investigate how the suitability of climate datasets for input into complex impact models can be assessed. Climate forcing datasets for APSIM are generated from the output of multiple climate models using three different types of method using unprocessed climate output, statistically correcting climate model output and perturbing observations of the recent climate with simulated future climate changes. The application of multiple processing methods to multiple climate model simulations generates a range of different climate datasets. Simulated impact changes are sensitive to differences in future changes in the climate between these datasets. In addition, simulated impact changes can be sensitive to differences in the representation of the recent climate. These arise because different climate models have different errors in their simulations of the real climate and different processing methods correct these to different extents. The thesis takes an impact-based approach to assessing different climate datasets. It compares wheat yields for a recent time period from APSIM simulations forced with datasets derived from climate models with yields from APSIM simulations forced with climate observations and addresses the question: To what extent can impact-based assessment be used to evaluate climate forcing datasets for a complex crop model? While the impact-based approach reveals those combinations of climate models and processing methods that result in large errors in simulated recent yields, there is no strong overall relationship between errors in simulated recent yields and future changes in yields. The approach is therefore not useful in determining which climate models and processing methods give the most plausible simulated future changes in yields. This thesis therefore concludes that, to sample uncertainty, impact studies should ideally apply multiple processing methods to many climate models.

An important aspect of empirical research is the construction of a model that represents the data. In psychological and educational measurement, models are typically evaluated regarding their ability to fit well to the observed data. Philosophers of science have long recognized that goodness-of-fit to the realized data is an insufficient metric of a model’s usefulness; models should also be appraised regarding their generalizability to unseen data. Frequentist statistics, Bayesian inference, and information theory seem to offer philosophically and methodologically dissimilar perspectives on model evaluation. However, this dissertation develops a simple theoretical framework that integrates these three perspectives. Within this framework, the information-theoretic principle of minimum description length is explored in the context of item response theory modeling. The findings reveal that complexity in item response theory is defined not by the number of freely estimated parameters in a model, but by its functional form. The frequentist, Bayesian, and information-theoretic approaches are then utilized in evaluating the usefulness of a unidimensional 3-parameter logistic model of item response data from the Program for International Student Assessment. Philosophical ramifications, future research directions, and implications for educational and psychological measurement are discussed.

Previous authors have developed many different types of water pipe condition and failure models using the various methodologies available.  Contrary, current utilities are struggling to maintain their current water infrastructure system, due to the lack of effective prediction tools at hand.  The gap between the methodologies available in academic research and the tools available to current water utilities needs to be addressed.  This paper presents a fuzzy inference prediction model used to forecast the performance rating of individual drinking water pipeline sections (node to node) in which utilities can easily apply to their drinking water infrastructure system. Prior to the development of a prediction model, a through literature and current practice review is completed detailing and summarizing all the available mathematical models.  Following, an infrastructure overview is presented detailing the various pipe materials, lifecycle and failure modes and mechanisms.  A data structure is also detailed which lists all parameters that affect the condition and/or performance of a pipeline.  All of these tools are successfully used to develop a fuzzy inference performance model. The fuzzy inference performance model is considered novel in that it considers close to 30 pipe parameters.  Moreover, the performance model is applied using the Western Virginia Water Authority (WVWA) and the Washington Suburban Sanitary Commission (WSSC) databases to evaluate and verify the predicting results.  Lab testing of several pipe samples is also used to evaluate the model.  The testing consists of a ring bearing test which is used to calculate the rupture modulus of the pipe.  Comparing the original vs. the current rupture modulus can determine the remaining strength of the pipe.  The remaining strength can then be used to assess the performance results predicted by the fuzzy inference model. Further a framework is set forth which utilizes the model's predicted performance ratings to develop deterioration curves which can be used as a tool to forecast and plan future inspection, repair, rehabilitation and replacement of water pipelines.  The deterioration model is made up of a Markov chain approach coupled with a non-optimization technique.

Using General Circulation Models (GCMs) for tropical cyclone studies is challenging due to the relatively small size of the storms, the intense convection and a host of scale interactions. However, with the advancement of computer architectures, GCMs are becoming capable of running at high horizontal resolutions with grid spacings of less than 60 km. As a result, high-resolution GCMs are becoming a tool of choice to evaluate tropical cyclones in current and future climate conditions. This raises questions concerning the fidelity of GCMs for tropical cyclone assessments. The physical and dynamical components of GCMs need to be evaluated to assess their reliability for tropical cyclone studies. An idealized tropical cyclone test case for high-resolution GCMs is developed and implemented in aqua-planet mode with constant sea surface temperatures. The initial conditions are based on an analytic initial vortex seed that is in gradient-wind and hydrostatic balance and intensifies over a 10-day period. The influence of the model parameterization package on the development of the tropical cyclone is assessed. In particular, different physics parameterization suites are investigated within the National Center for Atmospheric Research's Community Atmosphere Model CAM, including physics versions 3.1, 4 and 5. The choice of the CAM physics suite has a significant impact on the evolution of the idealized vortex into a tropical cyclone. In addition, a test case of intermediate complexity is introduced. Therein it is suggested that a GCM dynamical core be paired with simple moist physics to test the evolution of the test vortex. This simple-physics configuration includes important driving mechanisms for tropical cyclones, including surface fluxes, boundary layer diffusion and large-scale condensation. The impact of the CAM dynamical core (the resolved fluid flow component) on the tropical cyclone intensity and size is evaluated. In particular, the finite-volume, spectral element, Eulerian spectral transform and semi-Lagrangian spectral transform dynamical cores are utilized. The simple-physics simulations capture the dominant characteristics of tropical cyclones and are compared to the CAM 5 full physics results for each dynamical core. The research isolates the impact of the physical parameterizations, numerical schemes and uncertainties on the evolution of the cyclone in CAM.

The purpose of this study was to determine if control charts are an effective tool to identify trends in forest growth and yield model accuracy. Accurate forest growth and yield models are important for projecting future forest composition. However, environmental factors have the potential to make forest growth models created from historic data inaccurate. Control charts in this study determine if forest growth predictions fall within confidence limits established for historic growth at a number of points in time. Two data sets were used in this study: the first was a Continuous Forest Inventory (CFI) from three tracts at the University of Tennessee Cumberland Research Station and the second data set was Forest Inventory and Analysis data collected by the U.S. Forest Service. The CFI plots represented a stand level data set measured every 5 years from 1962-1977 and revisited for a re-measurement in 2009. The FIA plots were a regional data with subsets of plots measured annually from 1999-2008. The FIA data set was limited to plots of the oak/hickory forest type from Tennessee, Alabama, and Georgia. Two forest growth and yield models were used to predict growth: (1) WinYield and (2) Forest Vegetation Simulator (FVS). The two different data sets were used with both FVS and WinYield to evaluate control charts using different models ad at different spatial and temporal scales. The data sets were also subset by site index, stand age, stocking percent, aspect, and species composition to determine if control charts could identify changes in model accuracy for forests subjected to different growing conditions. The CFI and FIA data had short growth predictions and control charts indicated that there were no trends affecting accuracy. The CFI data also had a long growth prediction of 32 years and the control charts found that the predictions using WinYield and FVS were inaccurate, indicating that there may be a trend causing inaccuracy in the model.

Three-dimensional air quality models are vital tools for air quality research. The models are under continuous development as the knowledge and understanding of atmospheric processes improve. Once a model has been evaluated retrospectively against available observations, sensitivity studies can be conducted to determine possible causes for discrepancies between the model and observations and to assess potential impact of future emission changes. This research will evaluate the performance of two air quality models, the Community Multiscale Air Quality (CMAQ) model and the Comprehensive Air Quality Model with extensions (CAMx), at a fine scale over the southeastern United States in January and July 2002. The air quality in the southeast is of particular interest because of the high NH3 emissions from agriculture which play a key role in the nutrient and nitrogen cycle, act to neutralize acids in the air, and participate in the formation of PM2.5. The baseline simulations are completed at a horizontal grid spacing of 4-km using both CMAQ and CAMx. The evaluation of meteorological variables, chemical concentrations, wet and dry deposition, column mass, and visibility is completed using available observations from surface measurements and satellite data. The evaluation protocol involves analysis through domain-wide statistics, spatial distribution, and temporal variations. Additionally, two sensitivity studies are conducted. In order to assess the model sensitivity to horizontal grid spacing, a sensitivity study evaluates the performance of CMAQ at 12-, 4-, and 1.33-km horizontal grid spacings. The second sensitivity study evaluates the sensitivity of CMAQ to emission reductions of SO2, NOx, and agricultural-livestock NH3. In the baseline simulation, O3 and PM2.5 are overpredicted by both models in January and underpredicted by both models in July, with CAMx predicting higher values than CMAQ in both months. The overprediction by the models in January is likely influenced by simulated weaker vertical mixing than what occurred in the true atmosphere. In July, underestimated emissions of precursor species or overpredicted wet deposition may be contributing to the underpredicted O3 and PM2.5. The spatial distribution of the adjusted gas ratio indicates that the regions of high NH3 emissions in the eastern NC and northeastern GA are NH3-rich and reductions of NH3 alone would do little to reduce PM2.5 pollution, which will be further evaluated in the second sensitivity study. When compared to satellite data, CMAQ shows good agreement for CO and NO2 in both months, with larger biases for O3 and AOD. These discrepancies may be due to uncertainties in the boundary conditions of the model and the calculation of AOD from the model predictions, as well as assumptions made in the satellite retrieval algorithms. CMAQ shows some improvement for O3 and PM2.5 using finer grid spacings in January, but no improvement in July. One cause for this seasonal variation is the increased mass removed through wet deposition at finer scales in both months. This reduces the overprediction of PM species in January, but exacerbates the underprediction in July. Additional factors impacting the sensitivity to horizontal grid resolution are the sensitivity of the meteorology model and emissions. Through the emission sensitivity study, it was found that NH3 emission reductions have a larger impact than SO2 or NOx emission reductions on PM2.5 in January, while SO2 emission reductions result in the largest decrease in PM2.5 in July. Combining NH3 emission reductions with the projected SO2 and NOx emission reductions act to reduce PM2.5 more than SO2 and NOx emission reductions alone in both January and July.