Overview
We build and maintain predictive statistical and machine learning models that answer pressing applied questions about species, habitats, wetlands, water quality, and ecological systems. Our work spans the full modeling lifecycle — from problem framing and data assembly through model training, validation, uncertainty quantification, and deployment — always with a clear eye on how model products will actually be used in decisions, permitting, and policy. We select methods to fit the problem rather than defaulting to any single approach, drawing on classical statistical tools alongside modern machine learning where each is appropriate.
Selected Projects
Threatened & Endangered Species Habitat Models
NC Department of Transportation — ATLAS Project
Since 2018, we have served as a statistical advisor for the development and maintenance of species habitat models. We guided the determination of appropriate methods given available data. For a subset of terrestrial and aquatic listed species, we performed Random Forest models to predict the probability of suitable habitat and then apply risk-based thresholds. Our approach was informed by the constraints imposed by the presence-only, rare event data structure. To validate model performance, we designed a web-based expert elicitation strategy and a field-based biological sampling protocol. The final species models feed into the ATLAS environmental screening platform, enabling NCDOT project teams to identify potential biological conflicts earlier, evaluate trade-offs between alternatives more efficiently, and automate generation of required documentation. We also advise on appropriate model interpretation and application across this range of ATLAS tools.
High-Resolution Wetland Occurrence & Type Modeling
NC Department of Transportation — ATLAS Program
We are refining a machine learning pipeline to predict wetland occurrence and type at high spatial resolution across all 100 counties in North Carolina. The work involves refactoring code architecture for memory efficiency, training and tuning models against high-resolution environmental input features, extending the approach from occurrence to wetland type classification, and producing deployable datasets with full documentation and metadata for integration into the ATLAS interface.
Bat Culvert Use: Study Design & Statistical Modeling
NC Department of Transportation
We provided statistical study design and modeling support for a multi-year NCDOT investigation of bat use of roadway culverts statewide. Historically, most bat surveys had been conducted in larger, concrete structures expected to have a high probability of bats. Thus, the existing data were unsuited to answer questions that required a broader, more representative sample, such as "Can we say that certain types or sizes of culverts have zero bats?" or "Does the type or size of culvert influence the probability of bat presence?". Our work involved (1) integrating data from multiple pipe, culvert, and historical bat survey databases; (2) designing a statistically rigorous sampling strategy to fill knowledge gaps and answer key questions about use frequency and culvert characteristics; and (3) building predictive models. Analytic challenges included fixed and random effects, non-linear and interactive relationships among variables, and the rarity of bat detections. We delivered a Year 1 interim analysis in 2024 and are preparing the final report.
Falls Lake Water Quality Statistical Modeling
Upper Neuse River Basin Association (UNRBA) / Brown & Caldwell
We contributed a suite of statistical models for Falls Lake, a drinking water reservoir subject to ongoing regulatory scrutiny over nutrient loading and water quality standards. Our work primarily included trend analyses of reservoir condition (e.g., regulated nutrients, oxygen, and sediments) and Bayesian networks to support scenario analyses of how changing inputs and lake processes could impact the attainment of designated uses and water quality standards. These and other smaller analyses and data visualizations provided a technical basis for ongoing regulatory discussions where the effects of nutrient loading and climate change are difficult to disentangle.