eBird Data Challenge

Examples

Here are some example analyses carried out in the past by the Bird Count India team to explore some of the data in eBird. In each case, we provide step-by-step procedures so that you can get a sense of the kind of data processing that was done to arrive at the result.

First things first!

The very first thing to consider, even before downloading the data, is whether you want to use the entire dataset for India. If you do, be warned that there are 4 million rows in the dataset, and you will not be able to open this file in Excel. If your software has a limit on the number of rows it can read (Excel’s limit is around 1 million), then it might be best to download a subset of the data, rather than all India. For example, you can download the data for only a single State or for a single species, or for a restricted date range. (Note that the download page gives the option to ‘include unvetted data’. For most purposes, it’s best to exclude such data.)

If you do want the entire India data, you will need to use a suitable software program to subset what you want. For example, one commonly used software platform for analysis and graphics is R. In R, you would open and subset the data using commands similar to those below.

dat <- read.delim("ebd_IN_prv_relAug-2016.txt", na.strings = c("NA", "", "null"), quote="")
## check the column names
names(dat)
## subset only data from Kerala
kl <- subset(dat, STATE_PROVINCE == "Kerala")
## check if number of rows are OK
nrow(kl)
## gives 889,898, just about OK for Excel!

If you have trouble opening the downloaded data, or are stuck at any stage, please let us know in the comments below, or on the Facebook event page.

Example question 1: What is the country-wide distribution of birding effort represented in eBird?

The motivation behind this question might be to assess which areas (let’s say Districts) have the most active eBirders, and also to identify gaps in bird information such that efforts can be made to fill them.

To answer this question, it is necessary to count up the number of lists per District. We might first want to create a new spreadsheet (or ‘dataframe’, in R) in which each list is represented by one row (in contrast to the raw data, in which each record is represented by one row, and therefore each list may have several rows). In Excel, this can be done using pivot tables.

Once this is done, we can tabulate the number of rows per District and then arrange in descending order, perhaps. If we want to display the results on a map, then we need to download map data for India showing administrative boundaries, match the District names from eBird to the map data; and then use software like QGIS to display a map of effort.

Examining this map could tell us where eBirders are most active, and conversely, where the major gaps in information are. See an example of this kind of analysis in an earlier post on birding gaps.

Example question 2: How much birding is required in a single spot in order to find all or at least most species?

In other words, how much birding effort is required to adequately document the birds of a place? For this, we might take a single location that has substantial birding effort, and look at how species numbers accumulate with effort. To put it another way, as the total amount of time spent birding in that location increases, what is the pattern of increase in the total number of species seen? We would expect there to be a rapid increase at first, and then gradually to stop increasing. The point (amount of time, number of lists) where new species stop being found might be considered adequate effort for that location.

To examine this, after choosing a particular location, we would want to order the birding lists in sequence (from earliest to latest, the checklist ID gives this), calculate the accumulated species seen and plot that versus the accumulated number of lists or the accumulated time spent birding.

To see an example of this done in the past, please look at an earlier analysis of repeated lists at a location.

Example question 3. How well are Important Bird and Biodiversity Areas (IBAs) in India covered in eBird?

IBAs are areas of particularly rich and/or threatened bird diversity. Clearly it is important to document and monitor the birdlife within them. Very few Indian IBAs have regular monitoring programmes; can birdwatchers uploading their birdlists to eBird contribute useful information?

For this, we first need a digital map of IBAs from India. Once we have the boundaries of different IBAs, we can check which eBird lists have been contributed from within those boundaries by looking up the latitudes and longitudes of the lists. From those eBird lists from within IBA boundaries, we can then calculate the amount of eBirding (in terms of lists and/or duration) from each IBA, and sort them to see which are heavily and which are poorly eBirded.

For a single IBA (or a set of them), we could also calculate the reporting frequency of each species. The reporting frequency for a particular species is the proportion of ‘complete’ lists that contain that species. The reporting frequency can be compared (with caution!) across IBAs and across seasons/years.

A brief analysis looked at these sorts of questions for IBAs before.

From across the World

eBird data have been used in a large number of research and conservation applications across the world, and cited in various scientific publications. Here we highlight two examples of interesting work that uses eBird data; both from North America.

In the first piece of work, an animated migration map of a large number of species was created by carrying out species distribution modelling on data submitted to eBird by participants. The map shows the movement of transcontinental migrants based on the aggregation of large amounts of information on observation of birds in space and time.

The second piece of work we highlight here is an assessment of how observation skills of individual birders increases over time. It looks at how individual birders at a particular place discover new species and whether this can be used as a measure of differences and changes in observation and identification skills. The conclusion is unsurprising: the more you watch birds, the better a birder (in terms of observation and identification skills) you can become!