12 Data quality control

OBIS ignores records that do not meet a number of standards. For example, all species names need to be matched against an authoritative taxonomic register, such as the World Register of Marine Species. In addition, quality is checked against the OBIS required fields as well as against any impossible values. OBIS checks, rejects and reports the data quality back to the OBIS nodes, but never change records. The OBIS tier 2 nodes are responsible for the data quality and communicate errors back to the data providers. A number of QC tools are developed to help data providers and OBIS nodes:

For specific concerns regarding quality control checks or issues, please submit a GitHub ticket to the OBIS QC repository.

12.1 Why are records dropped?

Records can be dropped and therefore not published with your dataset for a number of reasons, including:

  • The species is not marine
  • The ‘scientificName’ or scientificNameID did not match with WoRMS
  • Issues with coordinates:
    • No coordinates given
    • decimalLatitude or decimalLongitude out of range
  • The coordinate is zero

For each dataset published, a quality report is generated where the number of dropped records and other quality issues will be flagged. Such reports can also be found when searching for data in OBIS. For example, if we searched for ‘Crustacea’ records, the following data quality report is given:

Number of Crustacean records dropped
Number of Crustacean records dropped

We can see that >110,222 Crustacean records have been dropped, mostly due to records missing coordinates or species being flagged as non-marine. Because species are determined as being marine by WoRMS, we acknowledge that sometimes species are marked as not_marine erroneously.For specific advice on this topic, see the common QC issues page.

To minimize the number of records dropped, be careful when formatting your data so that you are meeting the requirements.

12.2 How to conduct Quality Control

Once you have formatted your data for OBIS, or have received a formatted dataset, it is important to run quality control checks before publishing the dataset on the IPT. The following is a list of various tools you can use to help you perform quality checks on your data:

Conducting QC with obistools

Installing obistools requires the devtools package. Use the following code to install both packages:

install.packages("devtools")
devtools::install_github("iobis/obistools")

If you have difficulty installing obistools, please try updating your R packages, in particular the vctrs package. This can be done in RStudio in the Packages tab (“update” button) or by using the update.packages() command (you can choose which packages to update). If you cannot install obistools please reach out to and we will help you.

To use obistools to conduct quality control, you can follow the general order below. Please see the obistools GitHub for examples of how to use the functions.

  1. Check that the taxa names match with WoRMS
  2. Check that all required fields are present in the occurrence table
  3. Check coordinates
    • Plot them on a map to identify any points that appear outside the scope of the dataset obistools::plot_map. Using obistools:plot_map_leaflet() will additionally allow you to identify the row number for a particular data point.
    • Check that points are not on land obistools::check_onland
    • Ensure depth ranges are valid obistools::check_depth
  4. Check that the identifiers are present, unique, and appropriately correspond with each other obistools::check_eventids. You should also check the uniqueness of the occurrenceID field (e.g. using Hmisc::describe or simple code like length(occur$occurrenceID) == length(unique(occur$occurrenceID)))
  5. Check that eventDate is formatted properly obistools::check_eventdate
  6. Check for statistical outliers or other anomalies with Hmisc (below)

QC with R package Hmisc

The R package Hmisc has the function describe which can help you identify any discrepancies or outliers in your dataset.

It will summarize each of your variables for a given data field. This can help you quickly identify any missing data and ensure the number of unique IDs is correct. For example, in an Occurrence table with 1000 records, there should be 1000 unique occurrenceIDs.

library(Hmisc)
library(Hmisc)
data<-read.csv("example_data_occur.csv")
describe(data)
 
 12  Variables      407  Observations
------------------------------------------------------------------------------------------------------------------
CollectionCode 
       n  missing distinct    value 
     407        0        1  BIOFUN1 
                  
Value      BIOFUN1
Frequency      407
Proportion       1
------------------------------------------------------------------------------------------------------------------
eventID 
       n  missing distinct 
     407        0       27 

lowest : BIOFUN1_BF1A01 BIOFUN1_BF1A02 BIOFUN1_BF1A03 BIOFUN1_BF1A04 BIOFUN1_BF1A05
highest: BIOFUN1_BF1M3  BIOFUN1_BF1M4  BIOFUN1_BF1M6  BIOFUN1_BF1M8  BIOFUN1_BF1M9 
------------------------------------------------------------------------------------------------------------------
occurrenceID 
       n  missing distinct 
     407        0      407 

lowest : CSIC_BIOFUN1_1   CSIC_BIOFUN1_10  CSIC_BIOFUN1_100 CSIC_BIOFUN1_101 CSIC_BIOFUN1_102
highest: CSIC_BIOFUN1_95  CSIC_BIOFUN1_96  CSIC_BIOFUN1_97  CSIC_BIOFUN1_98  CSIC_BIOFUN1_99

This video shows how to use both obistools and Hmisc to conduct QC checks in R.