RUI: developing bayesian statistical models incorporating mechanistic processes for ecological data

The National Science Foundation Division of Mathematical Sciences awarded Middlebury College $330,740 on September 1, 2026, under the Mathematical and Physical Sciences program (CFDA 47.049) to develop Bayesian statistical models for ecological data analysis.

The project develops three complementary Bayesian hierarchical models that estimate hidden ecological processes—including species interactions, animal movement, and population growth—from imperfect and incomplete observational data while rigorously quantifying uncertainty. The first model addresses zero-inflated, non-negative community abundance data and accounts for species interactions through a joint change-point approach. The second model characterizes demographic processes through longitudinal physiological measurements using a stochastic growth framework. The third model, described partially in the source, represents a third complementary modeling approach. These models integrate ecological knowledge with statistical inference methods related to machine learning and artificial intelligence to produce biologically meaningful insights and more reliable predictions supporting conservation, environmental management, and regulatory decision-making. The project also provides research opportunities to undergraduate students.

Performance runs through August 31, 2029, at Middlebury College in Middlebury, Vermont. This is a Research in Undergraduate Institutions (RUI) award, indicating the research integrates undergraduate education within the framework of a primarily undergraduate institution.

Abstract

Drawing reliable scientific conclusions from ecological data requires statistical models that account for both the biological processes governing ecological systems and the uncertainty inherent in imperfect observations. However, key ecological processes such as species interactions, animal movement, and population growth are typically unobserved, requiring statistical models to estimate these hidden processes from the available data. Additionally, ecological data are often noisy or incomplete, making IT difficult to distinguish true biological patterns from observational uncertainty. New statistical models are developed that combine ecological knowledge with available data to estimate hidden ecological processes while rigorously quantifying uncertainty. These models and inference methods are closely related to machine learning and artificial intelligence. Rather than relying on a one-size-fits-all approach, the project develops three complementary models tailored to different types of ecological data. Together, these models produce biologically meaningful insights and more reliable predictions to support conservation, environmental management, and regulatory decision-making in the regions where the data are collected. This project also provides research opportunities to undergraduate students. Statistical ecology develops probabilistic models to infer ecological processes from imperfect and incomplete data. The project develops three bayesian hierarchical models that link latent ecological states to observed data while explicitly incorporating biological mechanisms. The first is a joint change-point model for zero-inflated, non-negative community abundance data that accounts for species interactions. The second is a stochastic growth model for longitudinal terrapin physiological measurements that characterizes demographic processes through time. The third is a three-dimensional animal movement model that quantifies whether and how the movement behavior of a shark community changes following the construction of a subsea power line. Collectively, these models advance the field of statistical ecology by developing novel methodology for joint modeling of community abundance data and for characterizing both community-level and three-dimensional animal movement. The project extends current statistical modeling efforts that attempt to incorporate ecological mechanisms within a stochastic model. The models are developed to be computationally efficient and suitable for large, complex ecological datasets. The resulting methods improve inference and prediction for complex ecological systems while contributing broadly applicable bayesian methodology for ecological and environmental data analysis. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.