occurrencelist() not returning all the gbif records for a species

[ropensci]

occurrencelist() not returning all the gbif records for a species

[sckott]

Hey @dmcglinn , answer in a second

[sckott]

Note that occurrencelist and occurencelist_many now return S3 objects, so you gotta use gbifdata to get the data (or convert it yourself I guess).

So, the problem is that the search is saying I want exactly "Aristolochia serpentaria", when what it seems like you want is that, but with variants, right?

Try this:

library(rgbif)
out <- gbifdata(occurrencelist(scientificname = 'Aristolochia serpentaria*', coordinatestatus = TRUE, maxresults = 1000))

unique(out$taxonName)

[1] Aristolochia serpentaria l. Aristolochia serpentaria   
Levels: Aristolochia serpentaria Aristolochia serpentaria l.

nrow(out)

[1] 96

Notice the asterisk after the taxon name, and that you get two names returned, one with l. , presumably for Linnaeus.

Gives 96 georeferenced records though, where GBIF gives 116 (GBIF does give 431 records as you said, but not all have lat/long data)

[sckott]

Hi again @dmcglinn Here's the responsible line in the code https://github.com/ropensci/rgbif/blob/master/R/methods.r#L44

It removes rows that have NA's for both lat and long. And 20 of the 116 records have zeros for both lat and long, even on the GBIF site, see here http://data.gbif.org/ws/rest/occurrence/list?scientificname=Aristolochia%20serpentaria*&coordinatestatus=TRUE

So those zeros get converted to NA's and removed in rgbif since I assumed that people wouldn't be interested in records without lat/long data.

What do you think?

[dmcglinn]

Hey @schamberlain thanks for the help and speedy replies on these issues. It does look like adding the '*' to the species name so that variants were returned was the primary issue, but also changing coordinatestatus to FALSE increased the number of returns as well. The query:

occurrencelist(scientificname = 'Aristolochia serpentaria*', coordinatestatus = FALSE, maxresults = 1e6)

returns 306 records which is identical to

http://data.gbif.org/ws/rest/occurrence/list?scientificname=Aristolochia%20serpentaria*&coordinatestatus=FALSE

but these queries do not return the full 431 items that a normal gbif species query on Aristolochia serpentaria returns. It appears that this may be do the fact that a GBIF query returns a broader range of names, specifically

library(rgbif)

out = occurrencelist(scientificname = 'Aristolochia serpentaria*', coordinatestatus = FALSE, maxresults = 1e6)

unique(out$taxonName)

[1] "aristolochia serpentaria"                                    
[2] "Aristolochia serpentaria L."                                 
[3] "Aristolochia serpentaria"                                    
[4] "Aristolochia serpentaria L. var. hastata (Nutt.) Duchartre"  
[5] "Aristolochia serpentaria L. var. hastata (Nuttall) Duchartre"
[6] "Aristolochia serpentaria var. hastata Duch."                 
[7] "Aristolochia serpentaria var. serpentaria"                   
[8] "Aristolochia serpentaria BARTON"  

whereas the GBIF query at http://data.gbif.org/ returns these names as well as synonym names such as:

Aristolochia hastata
Aristolochia nashii
Aristolochia convolvulacea Small
Aristolochia serpentaria var. hastata (Nutt.) Duchartre

[dmcglinn]

I checked that those additional names were indeed synonyms here: http://www.itis.gov/

[sckott]

Interesting. So GBIF.org is giving back synonyms as well as actual matches of the query string, whereas their API does not do that. Let me see if there is a parameter that we could fiddle with to get exactly what they give back.

[sckott]

The API docs says

scientificname - count only records where the scientific name matches that supplied - this is based on the scientific name found in the original record from the data provider and does not make use of extra knowledge of possible synonyms or of child taxa. For these functions, use taxonconceptkey.

and

taxonconceptkey - return only records which are for the taxon identified by the supplied numeric key, including any records provided under synonyms of the taxon concerned, and any records for child taxa (e.g. all genera and species within a family).

So the taxonconceptkey parameter does give back synonyms. However, from a user perspective, you would first have to get the taxonconceptkey, which is not ideal.

I'm guessing GBIF.org gets a taxonconceptkey based on your search, then looks up synonyms - but doesn't do this with the API - weird.

[dmcglinn]

Yea that's unfortunate, but I suppose one solution is the following?

library(rgbif)

gbifkey = taxonsearch(scientificname='Aristolochia serpentaria')$gbifkey
name_lkup = taxonget(key = as.numeric(as.character(gbifkey)))
sciname = as.character(subset(name_lkup, select='sciname', subset= rank == 'species' | rank == 'variety')[ ,1])
## to include variates add '*'
sciname = paste(sciname, '*', sep='')

out = occurrencelist_many(scientificname = sciname, coordinatestatus = FALSE, maxresults = 1e6)

out
$NumberFound
[1] 1845

However this now returns many more records than the original gbif.org query.

[sckott]

Hmmm, was trying getting synonyms from ITIS, and feeding those in to GBIF, but GBIF has different synonyms! Anyway, would be nice if GBIF had a synonyms API.

[dmcglinn]

The problem with the approach I proposed is that it does not guarantee that duplicate records are not returned. Does
occurrencelist return a unique record identifier field we could filter results on to ensure lack of duplication?

[sckott]

going out for a bit...

[dmcglinn]

just posted this pull request to include unique id's with the query results: #29

[dmcglinn]

Once #29 is merged the following query will return the same number of results as the GBIF web portal:

library(rgbif)

gbifkey = taxonsearch(scientificname='Aristolochia serpentaria')$gbifkey
name_lkup = taxonget(key = as.numeric(as.character(gbifkey)))

sciname = unique(as.character(subset(name_lkup, select='sciname',
                   subset= rank == 'species' | rank == 'variety')[ ,1]))

sciname = paste(sciname, '*', sep='')

out = occurrencelist_many(scientificname = sciname, maxresults = 1e6)

out 
$NumberFound
[1] 431

431 results matches the number of results returned when you do a simple web query for this species.

[sckott]

merged your pull, thanks for that!

What do you think @dmcglinn ? Should functions try to match exactly what happens in the GBIF web interface? Or not?

[dmcglinn]

I think you should provide the option for this with a new function, see my suggested solution in #30

The primary benefit in my mind is that if someone doesn't want to do the work of sorting out synonymy on their own and then querying each name individually you can provide the option of using GBIF's internal synonym mapping to complete the query. There is also the added benefit of the similarity between the web interface and the R query but that seems relatively minor (you'll probably just get less users complaining that something may have gone wrong). However, more functions in the package results in more effort maintaining so you may ultimately decide its not worth it.

[sckott]

Thanks for the new function!

Right, we should definitely strive to make it easier for users, which your function does.

I would like to have just one function that does everything with the occurencelist endpoint, but I imagine that is too difficult b/c there is a lot going on there. Another thing not included is the ability to specify many values for the same parameter, discussed here #28 . Hoping that they will change that since it's a lot of waste to used named params over and over again.

[sckott]

closing this for now