Public health disease surveillance consists of the systematic collection, analysis, and interpretation of data on chronic and infectious diseases to detect changes in disease incidence or emerging disease clusters.

A major limitation in most of the currently employed health surveillance methods used in epidemiological studies is the assumption that the regional observations are independently distributed and it is not straightforward to account for risk factors in the surveillance procedure. In this research, we study a set of spatial modeling approaches to account for risk factors and spatial autocorrelation in spatiotemporal surveillance.  In particular, simultaneous autoregressive (SAR) models, a class of spatial regression models popular in the econometrics literature, are extensively studied within the classical space-time cumulative sum (CUSUM) change detection framework for detecting changes in the spatially distributed count data. A case study is presented to illustrate the methodology for prospective detection of COVID-19 clusters in Florida Panhandle census tracts while accounting for social vulnerability of underlying communities.

Space–time surveillance of count data subject to linear trends 

Large-scale disasters, in particular, concurrently occurring geophysical hazards and pandemics, disproportionately affect disadvantaged communities. The knowledge of factors that describe a community’s social vulnerability against disasters is crucial to identify populations most likely to need support before, during, and after a disaster and determine how resources might be allocated for mitigation and response. 

In this research, using publicly available spatial social vulnerability data and daily Covid-19 case counts at the U.S. census-tract level, we assess the association between social vulnerability and the pandemic spread during a concurrently occurring hurricane in Florida. A multilevel mixed-effects modeling methodology is proposed to understand how the census-tract infection count trajectories (daily new cases) are associated with the demographic and socioeconomic characteristics of communities and a disruptive hurricane occurring during the study period. The proposed multi-level modeling of disaster and pandemic data has practical implications for emergency management and public health policy for coordinating resource allocation efforts between state, region, and community.

Understanding Spatiotemporal Variation of Social Vulnerabilities from Longitudinal Hurricane-Pandemic Data: A Multilevel Model of Covid-19 Pandemic During Hurricane Sally in Florida 

A spatial regression and clustering method for developing place-specific social vulnerability indices using census and social media data 

Transportation professionals widely use Negative Binomial distribution-based safety performance functions (SPFs) for predictive crash analysis. To assist the Florida Department of Transportation practitioners the overall goal of this project was to provide thresholds of certain SPF parameters that ensures reliable crash predictions are obtained consistently. A multilevel statistical model was constructed to quantify the SPF prediction errors for various context classifications and to compute the minimum Average Annual Daily Traffic (AADT) threshold to guarantee the SPF prediction errors are consistently smaller than a desired reliability metric.

Predictions of the remaining safe useful life of structural and mechanical components is crucial for cost effective management of critical infrastructure. In this research we develop a probabilistic approach for predicting the remaining useful life of composite structures using embedded Lamb wave sensor data. A principal component regression model is constructed to extract damage sensitive features of a Lamb wave sensor signal and establish a relation between the features and measured composite delamination areas. A Wiener process is then developed to model the random delamination growth and predict future damage and failure time probability. Method is illustrated on real sensor data collected under run-to-failure fatigue loading conditions.

Remaining Useful Life Estimation with Lamb-Wave Sensors Based on Wiener Process and Principal Components Regression 

Statistical updating of finite element model with Lamb wave sensing data for damage detection problems 

An important challenge in global positioning system (GPS) - based localization is the non-line-of-sight or multi-path problem, commonly encountered in urban environments. When the direct line-of-sight to a satellite is blocked by a building, its GPS signal may reach the receiver on the ground via reflections from buildings, resulting in large errors or outliers in the pseudorange measurements.

In this research, we propose new unbiased estimators of the noise covariances in a factor-graph formulation when sensor data is contaminated with outliers and Gaussianity of data is not satisfied, particularly applicable to the GPS-based localization problem under non-line-of-sight conditions. Main contributions are the extending of the unbiased variance estimators to the existing factor-graph problem in the literature to the multi-path problem, the study of nonlinear vehicle motion models, and the investigation of the performance of the approach with a real GPS data.

Robust error estimation based on factor-graph models for non-line-of-sight localization