search

Fireandplants / plant_gbif

This repository is for data and scripts related to plant species distribution across the globe using the Global Biodiversity Information Facility (GBIF) dataset.
4 stars 2 forks source link

readme

plant_gbif

This repository is for data and scripts for matching taxon names to GBIF occurrence data. The master branch represents the current project investigating abandonment of the arbuscular mycorrhizal mutualisms across angiosperms. It currently mirrors the myco-evol-project branch. The plant_gbif repo contains branches for other related projects with similar workflows.

Note: one thing I (Schwilk) have not cleaned up is that all of Schwilk's scripts (python or R) assume that the "scripts" directory is the working directory. McGlinn's scripts assume that "." is the working directory.

Current release: DOI

Matching GBIF occurrence records

Matching the taxon names with plant occurrence records in the Global Biodiversity Information Facility (GBIF) dataset.

These scripts read text files as utf-8 and immediately treat as unicode internally. All matching and comparisons work on unicode internally. All written output is encoded as utf-8. This has a slight speed penalty but is worth it and necessary as there are unicode characters in the GBIF data and some in other taxon name sources.

A note on scientific names: The input data sources only inlcude latin binomials as the species names. These are not full scientific names weith authors nor do any taxa in the two original data sources (Tank et al tree and mycorrizal state state information data) include an infraspecific epithets. When we do name epxansion to synonyms and match against gbif data, we do include infraspecific epithets. We ignore authors when matching to species ebcause that information is missing.

Analysis steps / walkthrough

Much code for steps 1-2 are in the taxon-name-utils repository. That repository aims to create a set of general and useful name matching and synonym finding tools. The code in this repo assumes that both repos are cloned in the same parent directory --- these scripts modify the python module search path to include the taxon-name-utils/scripts directory.

1. Extract all possible name binomials in the full GBIF occurrence data

First, we must extract all possible taxon binomials from the full GBIF Plantae data. Schwilk downloaded the the full GBIF Plantae data on August 23, 2022 (https://doi.org/10.15468/dl.s4fvvn). This is 323,717,877 occurrence records. This data, stored as a compressed zip file is not in the git repository, but is referred to by the extract_gbif_names.py script. To rerun these analysis, the user must supply this file and modify the file location in the extract_gbif_names.py script. Other large data such as these are stored in the /data/ directory of the repo which is ignored by git (see .gitignore).

python extract_gbif_names.py

This will create the names list. The current version is ../query_names/gbif-occurrences-names_151014.csv. This is all unique scientific names in the GBIF Plantae occurrences data.

2. Prepare taxon name lists

Run

prepare_names_lists.sh

What it does: There are 5284 plant taxa with mycorrizal state information from Maherali 2016: ../query_names/myco_species . We expand this out to those names plus all synonyms using taxon-names-tools/synonymize.py. We use World Flora Online v.2022.07 Jul. 12, 2022. (see https://github.com/schwilklab/taxon-name-utils).

This will create a names list, ../query_names/myco_species_expanded. This expanded names list includes each name in the original list and all synonyms according to The World Flora Online backbone data

Then the script produces parsed versions of all names using a modified version of Cam Webb's parsenames gawk script from the taxon-tools package. Finally, the script lines up the original name as a single string alongside the parsed version for all three names files. The main useful result are two of these: ../query_names/myco_species_expanded_both and ../query_names/gbif-occurrences-names_220823_both

3. Conduct fuzzy name matching

This step creates a lookup table that associates every possible taxon binomial in the full GBIF Plantae occurrence database with its match in the expanded canonical names list created in step 1. The code uses fuzzy_match.py from the taxon-name-utils repository to do matching based on a combination of Levenshtein distances and Jaro-Winkler distances. The matching algorithm is

python make_myco_gbif_fuzzy_lookup.py

For every "expanded name" (canonical names and synonyms), go through all names in GBIF data and first match genus and then specific epithet. If there is no exact match, find the closest genus match that is within a Levenshtein distance of 2 and, within that genus, find the closest specific epithet within a Levenshtein distance of 3. "Closest" is defined by Jaro-Winkler similarity. The resulting table is ../query_names/gbif_myco_lookup_220823.csv. The threshold distances hard-coded in the script above over-match by design. Therefore, this table requires a bit of cleaning in R to throw out a few false-positive matches followed by a manual check.

This will overmatch so we now drop any suspect matches according to the following algorithm: 1) Any fuzzy match for which both names are listed in The World FLora Online database. 2) We conduct some tests to mark likely good matches asthose whose specific epithet invovles only a latin gender change, common typing mistakes and alternative spellings (caespitosa vs cespitosa, sylvestris, sylvestris). All fuzzy matches are then checked manually to avoid incorrect matches (match that involves a known change of meaning (eg "micro" vs macro" or Latin diminutives). Some of this could be automated as most false positives and likely good amtches follow a few common typing transpostions and spelling variations. But a manual cehck is safest. 3) The remaining suspect matches are hand checked and most marked for removal. The automatic parts of the steps above (steps 1 and 2) are executed by clean_gbif2canonical.R. This produces the file for manual editing gbif_myco_lookup_220826_cleaned.csv

Conduct the manual fixes by adding "TRUE" to the manual_remove column to remove incorrect matches. ON 2022-08-26, Schwilk removed 141 incorrect matches, leaving 844-141 = 703 fuzzy matches most of which appear to be spelling alternatives or data entry errors. Overall, that leaves 24991 names extracted from gbif that can be matched to the expanded canonical names (canonical plus synonyms). We need to convert matched names (we ignored author portion of name) back to the full names so we can match precisely with the gbif scientificName field. To do that run

finish_gbif2canonical.R

to finalize the manual marks and to create a lookup table that includes the full strings for the canonical names as well as the full gbif scientificName field (with author) needed in the next step. The final lookup table has 24991 unique canonical+synonym names matched to 29472 gbif full scientificName field names. The larger number of gbif scientificNames is beacause we match on full scientific names but omit authors because of many minor variations in how authors are specified in gbif data from heterogenous sources. The created file is gbif_myco_lookup_220826_final.csv

4. Extract matching records from the GBIF Plantae data

This step reads line by line through all GBIF occurrence records and extracts those for which 1) there is a latitude and longitude, and 2) for which the scientificName field matches a name in gbif_myco_lookup_220826_final.csv

python3 extract_matched_gbif_occurrences.py

Total records scanned = 323717877
Total matches found   = 189072622

The result is saved as a large tab-separated file, current version is data/myco-gbif-occurrences_extracted_.csv. This is our full species occurrence data, but it WILL have records with untrustworthy coordinates, it will include observations from horticultural plants (eg NY City Parks!) and therefore needs further cleaning. This file uncompressed is 46 GB and its MD5sum is bc5a6584d47670f45c2e927db4a960d4. I then compressed that file using xz compression to about 3 GB to send to Dan McGlinn.

Data cleaning and GIS queries

The next steps of the data processing removed possibly erroneous data, queried various GIS databases for environmental variables, and lastly summarized the spatial and environmental conditions at the species level.

The script to run this entire process is ./scripts/geog_filter/run_all.R

GBIF records were removed from the dataset using ./scripts/geog_filter/geog_filter.R if they were found to have any of the following attributes:

The script ./scripts/geog_filter/climate_query.R carried out the following environmental queries using GIS databases.

The WorldClim database was queried for mean annual temperature and annual precipitation data (Hijmans et al. 2005).

The International Soil Reference and Information Centre (ISRIC) world soil information database was used for total soil Nitrogen (N) (Batjes 2012).

The Global Gridded Soil Phosphorus Distribution Maps database from the Oak Ridge Oak Ridge National Laboratory Distributed Active Archive Center for labile inorganic phosphorus (P) content (Yang et al. 2014).

Lastly, the script ./scripts/geog_filter/climate_summary.R was used to compute the median and 95% quantile of all the enviornmental and spatial variables at the species level.

References

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. and Jarvis, A., 2005. Very high resolution interpolated climate surfaces for global land areas. International journal of climatology, 25(15), pp.1965-1978.

Batjes, N.H., 2012. ISRIC-WISE Derived Soil Properties on a 5 by 5 Arc-minutes Global Grid (Version 1.2). ISRIC–World Soil Information, Wageningen, The Netherlands.

Yang, X., W.M. Post, P.E. Thornton, and A. Jain. 2014. Global Gridded Soil Phosphorus Distribution Maps at 0.5-degree Resolution. Data set. Available on-line [http://daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1223