Using the rdataretriever to quickly analyze Breeding Bird Survey data

Introduction

The Breeding Bird Survey of North America (BBS) is a widely used dataset for understanding geospatial variation and dynamic changes in bird communities. The dataset is a continental scale community science project where thousands of birders count birds at locations across North America. It has been used in hundreds of research projects including research on biodiversity gradients (Hurlbert and Haskell 2003), ecological forecasting (Harris, Taylor, and White 2018), and bird declines (Rosenberg et al. 2019).

However working with the Breeding Bird Survey data can be challenging because it is composed of roughly 100 different files that need to be cleaned and then combined in multiple ways. These initial phases of the data analysis pipeline can require hours of work for even experienced users to understand the detailed layout of the data and either manually assemble it or write code to combine the data. The data structure and location also changes regularly meaning that code for this work that is not regularly tested quickly stops working.

This vignette demonstrates how using the rdataretriever can eliminate hours of work on data cleaning and restructing allowing researchers to quickly begin addressing interesting scientific questions. To do this is demonstrates analyzing the BBS data to evaluate correlates of biodiversity (in the form of species richness).

Load R Packages

We start by loading the necessary packages. In addition to the rdataretriever in this demo we’ll also use DBI, RSQLite and dplyr to work with the data, raster for working with environmental data, and ggplot2 for visualization.

library(rdataretriever)
library(DBI)
library(dplyr)
library(dbplyr)
library(raster)
library(ggplot2)

Install And Connect To The Breeding Bird Survey Data

First we’ll update the rdataretriever to make sure we have the newest data processing recipes in case something about the structure or location of the dataset has changed. Centrally updated data recipes reproducible research with this dataset because something changes every year when the newest data is released meaning that any custom code for processing the data stops working. When this happens the retriever recipe is updated and after running get_updates() data analysis code continues to run as it always has.

Next install the BBS data into an SQLite database named bbs.sqlite.

We could also load the data straight into R (using rdataretriever::fetch('breed-bird-survey')), store it as flat files (CSV, JSON, or XML), or load it into other database management systems (PostgreSQL, MariaDB, MySQL). The data is moderately large (~1GB) so SQLite represents a nice compromise between efficiently conducting the first steps in the data manipulation pipeline while requiring no additional setup or expertise. The large number of storage backends makes the rdataretriever easy to integrate into existing data processing workflows and to implement designs appropriate to the scale of the data with no additional work.

Having installed the data into SQLite we can then connect to the database to start analyzing the data.

bbs_db <- dbConnect(RSQLite::SQLite(), 'bbs.sqlite')

The two key tables for this analysis are the surveys and sites tables, so let’s create connections to those tables.

surveys <- tbl(bbs_db, "breed_bird_survey_counts")
sites <- tbl(bbs_db, "breed_bird_survey_routes")

The surveys table holds the data on how many individuals of each species are sampled at each site. The sites table holds information on where each site is which we’ll use to link the data to environmental variables.

Analyze The Data

To calculate the measure of biodiversity, which is species richness or the number of species, we’ll use dplyr to determine the number of species observed at each site in a recent year.

rich_data <- surveys %>%
  filter(year == 2016) %>%
  group_by(statenum, route) %>%
  summarize(richness = n()) %>%
  collect()
rich_data

The data is now smaller than the original ~1 GB, so we used collect to load the summarized data directly into R.

Next we need to get environmental data for each site, which we’ll get from the worldclim dataset.

bioclim <- getData('worldclim', var = 'bio', res = 10)

To extract the environmental data we first make our sites data spatial and add them to our map.

sites <- as.data.frame(sites)
sites_spatial <- SpatialPointsDataFrame(sites[c('longitude', 'latitude')], sites)

We can then extract the environmental data for each site from the bioclim raster and add it to the data on biodiversity.

bioclim_bbs <- extract(bioclim, sites_spatial) %>%
  cbind(sites)
richness_w_env <- inner_join(rich_data, bioclim_bbs)
richness_w_env

Now let’s see how richness relates to the precipitation. Annual precipition is stored in bio12.

ggplot(richness_w_env, aes(x = bio12, y = richness)) +
  geom_point(alpha = 0.5) +
  labs(x = "Annual Precipitation", y = "Number of Species")

It looks like there’s a pattern here, so let’s fit a smoother through it.

ggplot(richness_w_env, aes(x = bio12, y = richness)) +
  geom_point(alpha = 0.5) +
  geom_smooth()

This shows that there is low bird biodiversity in really dry areas, biodiversity peaks at intermediate precipitations, and then drops off at the highest precipitation values.

If we wanted to use this kind of information to inform conservation decisions at the state level, could look at the patterns within each state after filtering to ensure enough data points.

richness_w_env_high_n <- richness_w_env %>%
  group_by(statenum) %>%
  filter(n() >= 50)

ggplot(richness_w_env_high_n, aes(x = bio12, y = richness)) +
  geom_point() +
  geom_smooth() +
  facet_wrap(~statenum, scales = 'free') +
  labs(x = "Annual Precipitation", y = "Number of Species")

Looking back at this demo there is only one line of code directly involving the rdataretriever (other than installation). This demonstrates the strength that the rdataretriever brings to the the early phases of the data acquistion and processing pipeline by distilling those steps to a single line that provides the data in a ready-to-analyze form so that researchers can focus on the analysis of the data itself.

Conclusion

Thanks to the rdataretriever we can generate meaningful information about large scale bird biodiversity patterns in about 15 minutes. If we’d been working with the raw BBS data we would likely have spent hours manipulating and cleaning data before we could start this analysis.