Processed data for the analysis of human mobility changes from COVID-19 lockdown on bird occupancy in North Carolina, USA

Description

Methods

The data source we used for bird data was eBird, a global citizen science project run by the Cornell Lab of Ornithology. We used the COVID-19 Community Mobility Reports by Google to represent the pause of human activities at the county level in North Carolina.  These data are publicly available and were last updated on 10/15/2022. We used forest land cover data from NC One Map that has a high resolution (1-meter pixel) raster data from 2016 imagery to represent canopy cover at each eBird checklist location. We also used the raster data of the 2019 National Land Cover Database to represent the degree of development/impervious surface at each eBird checklist location. All three measurements were used for the highest resolution that was available to use.

We downloaded the eBird Basic Dataset (EBD) that contains the 15 study species from February to June 2020. We also downloaded the sampling event data that contains the checklist efforts information. First, we used the R package Auk (version 0.6.0) in R (version 4.2.1) to filter data in the following conditions: (1) Date: 02/19/2020 - 03/29/2020; (2) Checklist type: stationary; (3) Complete checklist; (4) Time: 07:00 am - 06:00 pm; (5) Checklist duration: 5-20 mins; (6) Location: North Carolina. After filtering data, we used the zero fill function from Auk to create detection/non-detection data of each study species in NC. Then we used the repeat visits filter from Auk to filter eBird checklist locations where at least 2 checklists (max 10 checklists) have been submitted to the same location by the same observer, allowing us to create a hierarchical data frame where both detection and state process can be analyzed using Occupancy Modeling. This data frame was in a matrix format that each row represents a sampling location and the columns represent the detection and non-detection of the 2-10 repeat sampling events. 

For the Google Community Mobility data, we chose the "Workplaces" categoriy of mobility data to analyze the Anthropause effect because it was highly relevant to the pause of human activities in urban areas. The mobility data from Google is a percentage change compared to a baseline for each day. A baseline day represents a normal value for the day of the week from the 5-week period (01/03/2020-02/06/2020). For example, a mobility value of -30.0 for Wake County on Apr 15, 2020, means the overall mobility in Wake County on that day decreased by 30% compared to the baseline day a few months ago. Because the eBird data we used covers a wider range of dates rather than each day, we took the average value of mobility before lockdown, during lockdown, and after lockdown in each county in NC.

For the environmental variables, we calculated the values in ArcGIS Pro (version 3.1.0). We created a 200 m buffer at each eligible eBird checklist location. For the forest cover data, we used "Zonal Statistics as Table" to extract the percentage of forest cover at each checklist location's 200-meter circular buffer. For the National Land Cover Database (NLCD) data, we combined low-intensity, medium-intensity, and high-intensity development as development covers and used "Summarize Within" to extract the percentage of development cover using the polygon version of NLCD. We used a correlation matrix of the three predictors (workplace mobility, percent forest cover, and percent development cover) and found no co-linearity. Thus, these three predictors plus the interaction between workplace mobility and percent development cover were the site covariates of the Occupancy Models. For the detection covariates, four predictors were considered including time of observation, checklist duration, number of observers, and workplace mobility. These detection covariates were also not highly correlated. We then merged all data into an unmarked data frame using the "unmarked" R package (version 1.2.5). The unmarked data frame has eBird sampling locations as sites (rows in the data frame) and repeat checklists at the same sampling locations as repeat visits (columns in the data frame).