Veronica Lee

Date of Award

2022

Document Type

Thesis

Degree Name

Bachelors

Department

Natural Sciences

First Advisor

Skripnikov, Andrey

Area of Concentration

Statistics with Mueseum Studies

Abstract

In our thesis, we will model the relative abundance of ten selected bird species in peninsular Florida using data from eBird, a citizen science project developed by the Cornell Lab of Ornithology. We will implement strategies developed by the Cornell Lab to make best use of citizen science data in the modeling context, such as imposing structure on the data by filtering for effort on the behalf of eBird participants. We will begin by conducting generalized linear modeling using the quasi- Poisson, negative binomial, and zero-inflated Poisson distributions for the response variable. We will use covariates related to both environment and effort. In particular, we will consider a full suite of land cover covariates instead of the smaller, manually selected subset used by the Cornell Lab. Next, we will model certain covariates as having nonlinear relationships with the response using generalized additive modeling techniques. To further extend on the previous work of the Cornell Lab, we will then incorporate spatial dependence into the modeling task by performing hierarchical generalized linear modeling with a spatial conditional autoregressive structure for random effects. Finally, we will investigate our model fits using predictive performance metrics and effect displays for the relationships between certain covariates and relative abundance. We found the quasi-Poisson hierarchical generalized linear model with spatial random effects to have the best overall quality; it performed better in certain predictive metrics than comparable models that did not account for spatial dependence. Additionally, environmental covariates were generally found to be less statistically significant after accounting for spatial dependence in the modeling process.

Download

Share

COinS