Feederwatch Analysis

I completed this analysis with one other person as our final project for Cornell's INFO 2950: Introduction to Data Science course in Fall 2022. The aim was to use Python to analyze data from the Cornell Lab of Ornithology's Project Feederwatch - an citizen science intiative to gather observations of avian species at birdfeeders across North America - to answer several research questions:

How does local housing density influence the distribution of avian species commonly observed at backyard feeders, on both a broad continental scale and on a smaller New York specific scale?

• Part A: Is observed North American bird species diversity at backyard feeders independent of housing density in the immediate area?
• Part B: Do taxonomic groups in New York that share biological characteristics (like preferred diet/habitat) also display marked preferences for feeders with different housing densities, or are they observed at constant rates across backyards with different housing densities?

We used three main datasets to address these questions. The first was a dataset composed of observations (rows) of bird sightings at backyard feeders between November 2020 and April 2021 across the US and Canada.

The second dataset held attributes regarding location/environment details based on a unique location id, which could be joined to the location id in Dataset 1.

The third dataset held species name translations. This dataset, when joined with Dataset 1 on species_code, allowed us to access the scientific name of each observed species, which then allowed us to investigate data by taxonomic group as in Research Question Part B.

For Research Question Part A, we discovered that there was in fact some association between US-Canada avian species diversity and a feeder's local housing density. This relationship could be approximately modeled with a linear regression y = 0.291 - 0.013x₁ - 0.024 x₂ - 0.126 x₃, where y represented the proportion of the total species for a region being found in a housing-density-specific area and where the baseline condition was a feeder in a rural area. Each x represented a housing density level ( x₁ = suburban/rural, x₂ = suburban, and x₃ = urban). The overall trend suggested by our findings was that species diversity for a region (i.e. a U.S. state) decreased on average with increasing housing density (more urban areas tended to have fewer species observed at feeders).

For Research Question Part B, we focused on two main taxonomic groups commonly observed at backyard feeders in New York with disparate life characteristics - Accipiter (soaring birds of prey such as hawks) and Junco (small sparrows). After adjusting for the different levels of total observations taking place across different areas, we discovered that there were statistically significant housing density preferences for each genus.

Overall, we illustrated that there is a statistically significant association between housing density on broad-scale avian biodiversity and on genus-specific observation frequencies.