Ben Raymond, Michael Sumner
- Usage overview
- Users: level of usage and expected knowledge
- Defining data sources
- Developer notes
- Data source summary
Often it’s desirable to have local copies of third-party data sets. Fetching data on the fly from remote sources can be a great strategy, but for speed or other reasons it may be better to have local copies. This is particularly common in environmental and other sciences that deal with large data sets (e.g. satellite or global climate model products). Bowerbird is an R package for maintaining a local collection of data sets from a range of data providers.
Bowerbird can be used in several different modes:
- interactively from the R console, to download or update data files on an as-needed basis
- from the command line, perhaps as a regular scheduled task
- programatically, including from within other R packages, scripts, or R markdown documents that require local copies of particular data files.
When might you consider using bowerbird rather than, say, curl or crul? The principal advantage of bowerbird is that it can download files recursively. In many cases, it is only necessary to specify the top-level URL, and bowerbird can recursively download linked resources. Bowerbird can also:
decompress downloaded files (if the remote server provides them in, say, zipped or gzipped form).
incrementally update files that you have previously downloaded. Bowerbird can be instructed not to re-download files that exist locally, unless they have changed on the remote server. Compressed files will also only be decompressed if changed.
Build up a configuration by first defining global options such as the destination on your local file system:
Bowerbird must then be told which data sources to synchronize. Let’s use data from the Australian 2016 federal election, which is provided as one of the example data sources:
my_source <- bb_example_sources("Australian Election 2016 House of Representatives data") ## add this data source to the configuration cf <- bb_add(cf, my_source)
Once the configuration has been defined and the data source added to it, we can run the sync process. We set
verbose = TRUE here so that we see additional progress output:
status <- bb_sync(cf, verbose = TRUE)
Congratulations! You now have your own local copy of your chosen data set. This particular example is fairly small (about 10MB), so it should not take too long to download. Details of the files in this data set are given in the
At a later time you can re-run this synchronization process. If the remote files have not changed, and assuming that your configuration has the
clobber parameter set to 0 (“do not overwrite existing files”) or 1 (“overwrite only if the remote file is newer than the local copy”) then the sync process will run more quickly because it will not need to re-download any data files.
Users can interact with bowerbird at several levels, with increasing levels of complexity:
Using bowerbird with data source definitions that have been written by someone else. This is fairly straightforward.
Writing your own data source definitions so that you can download data from a new data provider, but using an existing handler such as
bb_handler_rget. This is a little more complicated. You will need reasonable knowledge of how your data provider disseminates its files (including e.g. the source URL from which data files are to be downloaded, and how the data repository is structured). Be prepared to fiddle with
rgetsettings to accommodate provider-specific requirements (e.g. controlling recursion behaviour). The “Defining data sources” section below provides guidance and examples on writing data source definitions.
Writing your own handler function for data providers that do not play nicely with the packaged handlers (
bb_handler_earthdata). This is the trickiest, and at the time of writing we have not provided much guidance on how to do this. See the “Writing new data source handlers” section, below.
It is expected that the majority of users will fall into one of the first two categories.
A few example data source definitions are provided as part of the bowerbird package — see
bb_example_sources() (these are also listed at the bottom of this document).
Other packages provide data source definitions that can be used with bowerbird. The blueant package provides a suite of bowerbird data source definitions themed around Southern Ocean and Antarctic data, and includes a range of oceanographic, meteorological, topographic, and other environmental data sets.
The general bowerbird workflow is to build up a configuration with one or more data sources, and pass that configuration object to the
bb_sync function to kick off the download process. Each data source contains the details required by bowerbird to be able to fetch it, including a handler function that bb_sync will call to do the actual download.
bb_handler_rget function is a generic handler function that will be suitable for many data sources. Note that this
bb_handler_rget function is not intended to be called directly by the user, but is specified as part of a data source specification. The
bb_sync function calls
bb_handler_rget during its run, passing appropriate parameters as it does so.
bb_handler_rget is a wrapper around
bb_rget, which is a recursive file downloading utility. Typically, one only needs to define a data source in terms of its top-level URL and appropriate flags to pass to
bb_rget, along with some basic metadata (primarily intended to be read by the user).
Specifying a data source is done by the
bb_source function. This can seem a little daunting, so let’s work through some examples. Most of these examples are included in
Say we’ve found this bathymetric data set and we want to define a data source for it. It’s a single zip file that contains some ArcGIS binary grids. A minimal data source definition might look like this:
src1 <- bb_source( name = "Geoscience Australia multibeam bathymetric grids of the Macquarie Ridge", id = "10.4225/25/53D9B12E0F96E", doc_url = "https://doi.org/10.4225/25/53D9B12E0F96E", license = "CC-BY 4.0", citation = "Spinoccia, M., 2012. XYZ multibeam bathymetric grids of the Macquarie Ridge. Geoscience Australia, Canberra.", source_url = "http://www.ga.gov.au/corporate_data/73697/Macquarie_ESRI_Raster.zip", method = list("bb_handler_rget"))
The parameters provided here are all mandatory:
idis a unique identifier for the dataset, and should be something that changes when the data set is updated. Its DOI, if it has one, is ideal for this. Otherwise, if the original data provider has an identifier for this dataset, that is probably a good choice here (include the data version number if there is one)
nameis a human-readable but still unique identifier
doc_urlis a link to a metadata record or documentation page that describes the data in detail
licenseis the license under which the data are being distributed, and is required so that users are aware of the conditions that govern the usage of the data
citationgives citation details for the data source. It’s generally considered good practice to cite data providers, and indeed under some data licenses this is in fact mandatory
methodparameter is specified as a list, where the first entry is the name of the handler function that will be used to retrieve this data set (
bb_handler_rget, in this case) and the remaining entries are data-source-specific arguments to pass to that function (none here)
- and finally the
source_urltells bowerbird where to go to get the data.
Add this data source to a configuration and synchronize it:
This should have caused the zip file to be downloaded to your local machine, in this case into the
There are a few additional entries that we might consider for this data source, particularly if we are going to make it available for other users.
Firstly, having the zip file locally is great, but we will need to unzip it before we can actually use it. Bowerbird can do this by specifying a
For the benefit of other users, we might also add the
descriptionprovides a plain-language description of the data set, so that users can get an idea of what it contains (for full details they can consult the
doc_urllink that you already provided)
collection_sizeis the approximate disk space (in GB) used by the data collection. Some collections are very large! This parameter obviously gives an indication of the storage space required, but also the download size (noting though that some data sources deliver compressed files, so the download size might be much smaller)
data_groupis a descriptive or thematic group name that this data set belongs to. This can also help users find data sources of interest to them
access_functioncan be used to suggest to users an appropriate function to read these data files. In this case the files can be read by the
rasterfunction (from the
So our full data source definition now looks like:
src1 <- bb_source( name = "Geoscience Australia multibeam bathymetric grids of the Macquarie Ridge", id = "10.4225/25/53D9B12E0F96E", description = "This is a compilation of all the processed multibeam bathymetry data that are publicly available in Geoscience Australia's data holding for the Macquarie Ridge.", doc_url = "https://doi.org/10.4225/25/53D9B12E0F96E", license = "CC-BY 4.0", citation = "Spinoccia, M., 2012. XYZ multibeam bathymetric grids of the Macquarie Ridge. Geoscience Australia, Canberra.", source_url = "http://www.ga.gov.au/corporate_data/73697/Macquarie_ESRI_Raster.zip", method = list("bb_handler_rget"), postprocess = list("bb_unzip"), collection_size = 0.4, access_function = "raster::raster", data_group = "Topography")
This data source (Australian Election 2016 House of Representatives data) is provided as one of the example data sources in
bb_example_sources(), but let’s look in a little more detail here.
The primary entry point to this data set is an HTML index page, which links to a number of data files. In principle we could generate a list of all of these data files and download them one by one, but that would be tedious and prone to breaking (if the data files changed our hard-coded list would no longer be correct). Instead we can start at the HTML index and recursively download linked data files.
The definition for this data source is:
src2 <- bb_source( name = "Australian Election 2016 House of Representatives data", id = "aus-election-house-2016", description = "House of Representatives results from the 2016 Australian election.", doc_url = "http://results.aec.gov.au/", citation = "Copyright Commonwealth of Australia 2017. As far as practicable, material for which the copyright is owned by a third party will be clearly labelled. The AEC has made all reasonable efforts to ensure that this material has been reproduced on this website with the full consent of the copyright owners.", source_url = c("http://results.aec.gov.au/20499/Website/HouseDownloadsMenu-20499-Csv.htm"), license = "CC-BY", method = list("bb_handler_rget", level = 1, accept_download = "\\.csv$"), collection_size=0.01)
Most of these parameters will be familiar from the previous example, but the
method definition is more complex. Let’s look at the entries in the
method list (these are all parameters that get passed to the
level = 1specifies that
bb_rgetshould download recursively, but only recurse down one level (i.e. follow links found in the top-level
source_urldocument, but don’t recurse any deeper. If, say, we specified
level = 2, then
bb_rgetwould follow links from the top-level document as well as links found in those linked documents.)
accept_download = "\\.csv$"means that we only want to retrieve csv files.
Add this data source to a configuration and synchronize it:
Once again the data have been saved into a subdirectory that reflects the URL structure (
c:/temp/data/bbtest/results.aec.gov.au/20499/Website/Downloads). If you examine that directory, you will see that it contains around 50 separate csv files, each containing a different component of the data set.
You can immediately see that by using a recursive download, not only did we not need to individually specify all 50 of those data files, but if the data provider adds new files in the future the recursive download process will automatically find them (so long as they are linked from the top-level
The Earthdata system is one of NASA’s data management systems and home to a vast range of Earth science data from satellites, aircraft, field measurements, and other sources. Say you had a rummage through their data catalogue and found yourself wanting a copy of Sea Ice Trends and Climatologies from SMMR and SSM/I-SSMIS.
Data sources served through the Earthdata system require users to have an Earthdata account, and to log in with their credential when downloading data. Bowerbird’s
bb_handler_earthdata function eases some of the hassle involved with these Earthdata sources.
First, let’s create an account and get ourselves access to the data.
create an Earthdata login via https://wiki.earthdata.nasa.gov/display/EL/How+To+Register+With+Earthdata+Login if you don’t already have one
we need to know the URL of the actual data. The metadata record for this data set contains a “Download via HTTPS” link, which in turn directs the user to this URL: https://daacdata.apps.nsidc.org/pub/DATASETS/nsidc0192_seaice_trends_climo_v3/. That’s the data URL (which will be used as the
source_urlin our data source definition)
browse to the that data URL, using your Earthdata login to authenticate. When you use access an Earthdata application for the first time, you will be requested to authorize it so that it can access data using your credentials (see https://wiki.earthdata.nasa.gov/display/EL/How+To+Register+With+Earthdata+Login). This dataset is served by the NSIDC DAAC application, so you will need to authorize this application (either through browsing as just described, or go to ‘My Applications’ at https://urs.earthdata.nasa.gov/profile and add the application ‘nsidc-daacdata’ to your list of authorized applications)
You only need to create an Earthdata login once. If you want to download other Earthdata data sets via bowerbird, you’ll use the same credentials, but note that you may need to authorize additional applications, depending on where your extra data sets come from.
Now that we have access to the data, we can write our bowerbird data source:
src3 <- bb_source( name = "Sea Ice Trends and Climatologies from SMMR and SSM/I-SSMIS, Version 3", id = "10.5067/IJ0T7HFHB9Y6", description = "NSIDC provides this data set to aid in the investigations of the variability and trends of sea ice cover. Ice cover in these data are indicated by sea ice concentration: the percentage of the ocean surface covered by ice. The ice-covered area indicates how much ice is present; it is the total area of a pixel multiplied by the ice concentration in that pixel. Ice persistence is the percentage of months over the data set time period that ice existed at a location. The ice-extent indicates whether ice is present; here, ice is considered to exist in a pixel if the sea ice concentration exceeds 15 percent. This data set provides users with data about total ice-covered areas, sea ice extent, ice persistence, and monthly climatologies of sea ice concentrations.", doc_url = "https://doi.org/10.5067/IJ0T7HFHB9Y6", citation = "Stroeve J, Meier WN (2018) Sea Ice Trends and Climatologies from SMMR and SSM/I-SSMIS, Version 3. [Indicate subset used]. Boulder, Colorado USA. NASA National Snow and Ice Data Center Distributed Active Archive Center. doi:10.5067/IJ0T7HFHB9Y6. [Date Accessed].", source_url = c("https://daacdata.apps.nsidc.org/pub/DATASETS/nsidc0192_seaice_trends_climo_v3/"), license = "Please cite, see http://nsidc.org/about/use_copyright.html", authentication_note = "Requires Earthdata login, see https://wiki.earthdata.nasa.gov/display/EL/How+To+Register+With+Earthdata+Login . Note that you will also need to authorize the application 'nsidc-daacdata' (see 'My Applications' at https://urs.earthdata.nasa.gov/profile)", method = list("bb_handler_earthdata", level = 4, relative = TRUE, accept_download = "\\.(s|n|png|txt)$"), user = "your_earthdata_username", password = "your_earthdata_password", collection_size = 0.02, data_group = "Sea ice")
This is very similar to our previous examples, with these differences:
bb_handler_earthdata(whereas previously we used
bb_handler_earthdatais actually just a wrapper around
bb_handler_rget, but it takes care of some Earthdata-specific details like authentication using your Earthdata credentials
relative = TRUEmeans that
bb_rgetwill only follow relative links (i.e. links of the form
<a href="/some/directory/">...</a>, which by definition must be on the same server as our
source_url). Absolute links (i.e. links of the form
<a href="http://some.other.server/some/path">...</a>will not be followed. This is another mechanism to prevent the recursive download from downloading stuff we don’t want
- we have specified the file types that we want via the
Note that if you were providing this data source definition for other people to use, you would obviously not want to hard-code your Earthdata credentials in the
password slots. In this case, specify the credentials as empty strings, and also include
warn_empty_auth = FALSE in the data source definition (this suppresses the warning that
bb_source would otherwise give you about missing credentials):
<- bb_source( src3 name = "Sea Ice Trends and Climatologies from SMMR and SSM/I-SSMIS, Version 3", ... details as above...,user = "", password = "", warn_empty_auth = FALSE)
When another user wants to use this data source, they simply insert their own credentials, e.g.:
NASA’s Oceandata system provides access to a range of satellite-derived marine data products. The
bb_oceandata_handler can be used to download these data. It uses a two-step process: first it makes a query to the Oceancolour data file search tool (https://oceandata.sci.gsfc.nasa.gov/search/file_search.cgi) to find files that match your specified criterion, and then downloads the matching files.
Oceandata uses standardized file naming conventions (see https://oceancolor.gsfc.nasa.gov/docs/format/), so once you know which products you want you can construct a suitable file name pattern to search for. For example, “S*L3m_MO_CHL_chlor_a_9km.nc” would match monthly level-3 mapped chlorophyll data from the SeaWiFS satellite at 9km resolution, in netcdf format. This pattern is passed as the
search argument to the
bb_handler_oceandata handler function. Note that the
bb_handler_oceandata does not need a
source_url to be specified in the
Note that as of 1-Jan-2020, all Oceandata users require an EarthData login. You will need to provide your username and password in the
bb_source call, and you will also need to authorize the ‘OB.DAAC Data Access’ application in your EarthData account (see ‘My Applications’ at https://urs.earthdata.nasa.gov/profile).
Here, for the sake of a small example, we’ll limit ourselves to a single file (“T20000322000060.L3m_MO_PAR_par_9km.nc”, which is photosynthetically-active radiation from the Terra satellite in February 2000):
src4 <- bb_source( name = "Oceandata test file", id = "oceandata-test", description = "Monthly, 9km remote-sensed PAR from the MODIS Terra satellite", doc_url = "https://oceancolor.gsfc.nasa.gov/", citation = "See https://oceancolor.gsfc.nasa.gov/cms/citations", license = "Please cite", method = list("bb_handler_oceandata", search="T20000322000060.L3m_MO_PAR_par_9km.nc"), user = "your_earthdata_username", password = "your_earthdata_password", data_group = "Ocean data") ## add this source to a configuration and synchronize it: cf <- bb_config("c:/temp/data/bbtest") %>% bb_add(src4) status <- bb_sync(cf) ## and our local copy of this data file: status$files[]$file
Bowerbird hands off the complexities of recursive downloading to the
bb_rget utility. This allows bowerbird’s data source definitions to be fairly lightweight and more robust to changes by the data provider. However, one of the consequences of this approach is that bowerbird actually knows very little about the data files that it maintains, which can be limiting in some respects. It is not generally possible, for example, to provide the user with an indication of overall download progress (progress bar or similar) for a given data source because neither bowerbird nor
bb_rget actually know in advance how many files are in it or how big they are. Data sources do have a
collection_size entry, to give the user some indication of the disk space required, but this is only approximate (and must be hand-coded by the data source maintainer). See the ‘Reducing download sizes’ section below for tips on retrieving only a subset of a large data source.
- the synchronization process saves files relative to the
local_file_rootdirectory specified in
bb_rgetfunction saves files into a directory structure that follows the URL structure. For example, calling
bb_rget("http://www.somewhere.org/monkey/banana/dataset.zip")will save the local file
bb_rgetwill keep data files from different sources naturally separated into their own directories
Recursion is a powerful tool but will sometimes download much more than you really wanted. There are various methods for restricting the recursion:
if you want to include/exclude certain files from being downloaded, use the
*_followparameters control which links will be followed by the recursive spidering process, whereas the
*_downloadparameters control which data files will be downloaded
no_parent = TRUEprevents
bb_rgetfrom ascending to a parent directory during its recursion process, because if it did so it would likely be downloading files that are not part of the data set that we want (this is
TRUEby default). In some cases, though, you might want the recursion to ascend to a parent directory, and therefore need to override the default setting
no_check_certificate = TRUEwill allow a download from a secure server to proceed even if the server’s certificate checks fail. This option might be useful if trying to download files from a server with an expired certificate, but it is clearly a security risk and so should be used with caution
bb_rgetto pause for this number of seconds between successive retrievals. This option may help with servers that block multiple successive requests, by introducing a delay between requests
bb_rgetis not behaving as expected, try adding
debug = TRUE. This gives debugging output from the underlying
libcurlcalls (which is additional to the output obtained by calling
It’s up to you where you want your data collection kept, and to provide that location to bowerbird. A common use case for bowerbird is maintaining a central data collection for multiple users, in which case that location is likely to be some sort of networked file share. However, if you are keeping a collection for your own use, you might like to look at https://github.com/r-lib/rappdirs to help find a suitable directory location.
If the data source delivers compressed files, you will most likely want to decompress them after downloading. The postprocess options
bb_unzip, etc will do this for you. By default, these do not delete the compressed files after decompressing. The reason for this is so that on the next synchronization run, the local (compressed) copy can be compared to the remote compressed copy, and the download can be skipped if nothing has changed. Deleting local compressed files will save space on your file system, but may result in every file being re-downloaded on every synchronization run.
help("bb_unzip") for more information, including usage examples.
bb_cleanup postprocessing option can be used to remove unwanted files after downloading. See See
Some data providers require users to log in. The
authentication_note column in the configuration table should indicate when this is the case, including a reference (e.g. the URL via which an account can be obtained). For these sources, you will need to provide your user name and password, e.g.:
mysrc <- bb_example_sources("CMEMS global gridded SSH reprocessed (1993-ongoing)") mysrc$user <- "yourusername" mysrc$password <- "yourpassword" cf <- bb_add(cf, mysrc) ## or, using the pipe operator mysrc <- bb_example_sources("CMEMS global gridded SSH reprocessed (1993-ongoing)") %>% bb_modify_source(user = "yourusername", password = "yourpassword") cf <- cf %>% bb_add(mysrc)
Sometimes you might only want part of a data collection. Perhaps you only want a few years from a long-term collection, or perhaps the data are provided in multiple formats and you only need one. If the data source uses the
bb_handler_rget method, you can restrict what is downloaded by modifying the arguments passed through the data source’s
method parameter, particularly the
reject_download options. If you are modifying an existing data source configuration, you most likely want to leave the original method flags intact and just add extra flags.
Say a particular data provider uses predictable file name patterns that include the year information. It would be fairly easy to restrict ourselves to only the 2017 data using the
accept option. Here we use the
bb_modify_source helper function to do so:
mysrc <- bb_example_sources("CMEMS global gridded SSH reprocessed (1993-ongoing)") %>% bb_modify_source(user = "yourusername", password = "yourpassword", method = list(accept_follow = "/2017")) cf <- cf %>% bb_add(mysrc)
Alternatively, for data sources that are arranged in subdirectories, one could replace the
source_url with one or more that point to the specific subdirectories that are wanted.
If you have many data sources in your configuration, running the sync in parallel is likely to speed the process up considerably (unless your bandwidth is the limiting factor). A logical approach to this would be to split a configuration, with a subset of data sources in each (see
bb_subset), and run those subsets in parallel. One potential catch to keep in mind would be data sources that hit the same remote data provider. If they overlap overlap in terms of the parts of the remote site that they are mirroring, that might invoke odd behaviour (race conditions, simultaneous downloads of the same file by different parallel processes, etc).
An aspect of reproducible research is knowing which data were used to perform an analysis, and potentially archiving those data to an appropriate repository. Bowerbird can assist with this: see
bb_handler_rget R function provides a wrapper around
bb_rget that should be sufficient for many data sources. However, some data sources need a more elaborate method. Notes will be added here about defining new handler functions, but in the meantime look at
bb_handler_earthdata, which provide handlers for Oceandata and Earthdata data sources.
bb_summary function will produce a HTML or Rmarkdown summary of the data sources contained in a configuration object. If you are maintaining a data collection on behalf of other users, or even just for yourself, it may be useful to keep an up-to-date HTML summary of your repository in an accessible location. Users can refer to this summary to see which data are in the repository and some details about them.
Here is a
bb_summary of the example data source definitions that are provided as part of the bowerbird package:
For the Global Ocean - Multimission altimeter satellite gridded sea surface heights and derived variables computed with respect to a twenty-year mean. Previously distributed by Aviso+, no change in the scientific content. All the missions are homogenized with respect to a reference mission which is currently OSTM/Jason-2. VARIABLES
Authentication note: Copernicus Marine login required, see http://marine.copernicus.eu/services-portfolio/register-now/
Approximate size: 310 GB
House of Representatives results from the 2016 Australian election.
Approximate size: 0.01 GB
Documentation link: http://results.aec.gov.au/
Monthly remote-sensing chlorophyll-a from the SeaWiFS satellite at 9km spatial resolution
Authentication note: Requires Earthdata login, see https://urs.earthdata.nasa.gov/. Note that you will also need to authorize the application ‘OB.DAAC Data Access’ (see ‘My Applications’ at https://urs.earthdata.nasa.gov/profile)
Approximate size: 7.2 GB
Documentation link: https://oceancolor.gsfc.nasa.gov/
NSIDC provides this data set to aid in the investigations of the variability and trends of sea ice cover. Ice cover in these data are indicated by sea ice concentration: the percentage of the ocean surface covered by ice. The ice-covered area indicates how much ice is present; it is the total area of a pixel multiplied by the ice concentration in that pixel. Ice persistence is the percentage of months over the data set time period that ice existed at a location. The ice-extent indicates whether ice is present; here, ice is considered to exist in a pixel if the sea ice concentration exceeds 15 percent. This data set provides users with data about total ice-covered areas, sea ice extent, ice persistence, and monthly climatologies of sea ice concentrations.
Authentication note: Requires Earthdata login, see https://wiki.earthdata.nasa.gov/display/EL/How+To+Register+With+Earthdata+Login . Note that you will also need to authorize the application ‘nsidc-daacdata’ (see ‘My Applications’ at https://urs.earthdata.nasa.gov/profile)
Approximate size: 0.02 GB
Documentation link: https://nsidc.org/data/NSIDC-0192/versions/3
Weekly and monthly mean and long-term monthly mean SST data, 1-degree resolution, 1981 to present. Ice concentration data are also included, which are the ice concentration values input to the SST analysis
Approximate size: 0.9 GB
Documentation link: http://www.esrl.noaa.gov/psd/data/gridded/data.noaa.oisst.v2.html
A draft version of the Lake Superior Bathymetry was compiled as a component of a NOAA project to rescue Great Lakes lake floor geological and geophysical data, and make it more accessible to the public. No time frame has been set for completing bathymetric contours of Lake Superior, though a 3 arc-second (~90 meter cell size) grid is available.
Approximate size: 0.1 GB
Documentation link: https://www.ngdc.noaa.gov/mgg/greatlakes/superior.html