Stat Colloquium: Dr. Ali Arab

Title:Modeling and Forecasting Emerging Events across Space and Time

Abstract: Modeling the dynamics of spatio-temporal processes is often challenging, particularly under data sparsity, as is frequently the case during the early stages of an emerging process. For example, modeling the spread of a vector-borne infectious disease during its initial emergence can be especially difficult due to limited observations and incomplete knowledge of the underlying disease dynamics. These challenges are particularly relevant to emerging and re-emerging epidemics, as well as other rapidly evolving crises.

Moreover, data sparsity can lead to inefficient statistical inference, and unreliable or ineffective forecasts. Such challenges are common when modeling rare or emerging ecological, environmental, epidemiological, and social processes, particularly those occurring in previously unaffected regions, under unfamiliar conditions, or following abrupt changes. Furthermore, the urgency of modeling and forecasting these processes, especially during conflict- or climate-driven crises, often introduces additional constraints. These include limited availability of relevant predictors, delays in data collection and processing, and a lack of sufficiently fine-resolution data. Conventional statistical models may struggle to adequately capture the dynamics of these processes, particularly when observations are sparse, the underlying mechanisms are poorly understood, or the processes exhibit substantial spatial and temporal heterogeneity.

In this presentation, we discuss strategies for addressing these challenges in modeling and forecasting emerging spatio-temporal processes. These include integrating conventional and organic data sources, incorporating scientific knowledge through mechanistic Physics-based modeling frameworks, employing zero-modified hierarchical models to accommodate sparse observations, and leveraging transfer learning to improve parameter estimation and predictive performance. We illustrate these approaches through case studies involving emerging spatio-temporal processes..