Skip to content

annakrystalli/seabirddietDB

Repository files navigation

seabirddietDB

Travis build status Codecov test coverage

The goal of seabirddietDB is to provide access and tools to interact with a database of seabird diets collected around the British Isles.

Installation

You can install the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("annakrystalli/seabirddietDB")

Data

For more details on the dataset, check out the full documentation

Example

To access the data simply load the package. The data is then available

library(seabirddietDB)
seabirddiet
#> # A tibble: 2,857 x 33
#>       id  year startyear endyear multiyear location latitude longitude
#>    <int> <dbl>     <dbl>   <dbl> <lgl>     <chr>       <dbl>     <dbl>
#>  1     1  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  2     2  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  3     3  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  4     4  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  5     5  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  6     6  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  7     7  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  8     8  1983      1983    1983 FALSE     East An…     52.5      1   
#>  9     9  1983      1983    1983 FALSE     East An…     52.5      1   
#> 10    10  1983      1983    1983 FALSE     East An…     52.5      1   
#> # … with 2,847 more rows, and 25 more variables: pred_common_name <chr>,
#> #   pred_species <chr>, pred_rank <chr>, pred_aphia_id <int>,
#> #   pred_valid_name <chr>, pred_valid_aphia_id <int>,
#> #   pred_breeding_status <chr>, pred_age_group <chr>,
#> #   prey_orig_descr <chr>, prey_taxon <chr>, prey_rank <chr>,
#> #   prey_aphia_id <int>, prey_valid_name <chr>, prey_valid_aphia_id <int>,
#> #   prey_size <chr>, prey_age_group <chr>, freq_occ <dbl>, freq_num <dbl>,
#> #   freq_biomass <dbl>, sample_size <dbl>, sample_type <chr>, ref_n <dbl>,
#> #   ref_ids <chr>, source <chr>, notes <chr>

A version with more formal data types/structures in the columns (ie geographic information stored as sf, categorical data as factors etc) is also available. Note that you will need package sf installed to load this version of the data.

class(seabirddiet_)

Helpers

metadata

A number of metadata datasets are included with the package: attributes, references and classification. Load them as you would one of the datasets, e.g. :

references
#> # A tibble: 97 x 2
#>    ref_id ref_valid                                                      
#>    <chr>  <chr>                                                          
#>  1 REF001 Anderson et al 2014 Ibis 156, 23-34                            
#>  2 REF002 Bailey et al 1991 ICES Mar. Sci. Symp., 193, 209-216           
#>  3 REF003 Baker et 1999 JNCC Report 289, 1-51                            
#>  4 REF004 Blake BF 1984 J. Experimental. Mar.Biol. Ecol. 76, 89-103      
#>  5 REF005 Blake et al 1955 Estuarine, Coastal & Shelf Science 20, 559-568
#>  6 REF006 Blake et al 1980 Unpubl NCC Report                             
#>  7 REF007 Blake et al 1985 Estuarine, Coastal & Shelf Science 20, 559-568
#>  8 REF008 Bull  et al 2001 JNCC Report, No. 303                          
#>  9 REF009 Bull  et al 2001 JNCC Report, No. 315                          
#> 10 REF010 Bull et al 2004 Ardea 92, 43-52                                
#> # … with 87 more rows

List

List predators

sbd_predators()
#> [1] "Alca torda"                "Fratercula arctica"       
#> [3] "Fulmarus glacialis"        "Morus bassanus"           
#> [5] "Phalacrocorax aristotelis" "Puffinus puffinus"        
#> [7] "Rissa tridactyla"          "Uria aalge"

List prey

sbd_prey(verbose = TRUE)
#> # A tibble: 129 x 5
#>    prey_taxon    prey_rank prey_aphia_id prey_valid_name   prey_valid_aphi…
#>    <chr>         <chr>             <int> <chr>                        <int>
#>  1 Acanthephyra  genus            107018 Acanthephyra                107018
#>  2 Acanthephyra… species          107581 Acanthephyra pel…           107581
#>  3 Actinopteryg… class             10194 Actinopterygii               10194
#>  4 Agonidae      family           125588 Agonidae                    125588
#>  5 Agonus catap… species          127190 Agonus cataphrac…           127190
#>  6 Alloteuthis … species          153131 Alloteuthis subu…           153131
#>  7 Ammodytes la… species          146485 Hyperoplus lance…           126756
#>  8 Ammodytes ma… species          126751 Ammodytes marinus           126751
#>  9 Ammodytes to… species          126752 Ammodytes tobian…           126752
#> 10 Ammodytidae   family           125516 Ammodytidae                 125516
#> # … with 119 more rows

Filter data

filter predator

sbd_filter(pred_species = "Fratercula arctica")
#> # A tibble: 615 x 33
#>       id  year startyear endyear multiyear location latitude longitude
#>    <int> <dbl>     <dbl>   <dbl> <lgl>     <chr>       <dbl>     <dbl>
#>  1     1  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  2     2  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  3     3  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  4     4  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  5     5  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  6     6  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  7     7  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  8     8  1983      1983    1983 FALSE     East An…     52.5      1   
#>  9     9  1983      1983    1983 FALSE     East An…     52.5      1   
#> 10    10  1983      1983    1983 FALSE     East An…     52.5      1   
#> # … with 605 more rows, and 25 more variables: pred_common_name <chr>,
#> #   pred_species <chr>, pred_rank <chr>, pred_aphia_id <int>,
#> #   pred_valid_name <chr>, pred_valid_aphia_id <int>,
#> #   pred_breeding_status <chr>, pred_age_group <chr>,
#> #   prey_orig_descr <chr>, prey_taxon <chr>, prey_rank <chr>,
#> #   prey_aphia_id <int>, prey_valid_name <chr>, prey_valid_aphia_id <int>,
#> #   prey_size <chr>, prey_age_group <chr>, freq_occ <dbl>, freq_num <dbl>,
#> #   freq_biomass <dbl>, sample_size <dbl>, sample_type <chr>, ref_n <dbl>,
#> #   ref_ids <chr>, source <chr>, notes <chr>

filter metrics

sbd_filter(pred_species = "Fratercula arctica", metrics = "freq_biomass")
#> # A tibble: 219 x 31
#>       id  year startyear endyear multiyear location latitude longitude
#>    <int> <dbl>     <dbl>   <dbl> <lgl>     <chr>       <dbl>     <dbl>
#>  1   291  1973      1973    1973 FALSE     Isle of…     56.2     -2.56
#>  2   292  1973      1973    1973 FALSE     Isle of…     56.2     -2.56
#>  3   293  1973      1973    1973 FALSE     Isle of…     56.2     -2.56
#>  4   298  1973      1973    1973 FALSE     Isle of…     56.2     -2.56
#>  5   299  1973      1973    1973 FALSE     Isle of…     56.2     -2.56
#>  6   300  1974      1974    1974 FALSE     Isle of…     56.2     -2.56
#>  7   301  1974      1974    1974 FALSE     Isle of…     56.2     -2.56
#>  8   308  1974      1974    1974 FALSE     Isle of…     56.2     -2.56
#>  9   309  1974      1974    1974 FALSE     Isle of…     56.2     -2.56
#> 10   310  1975      1975    1975 FALSE     Isle of…     56.2     -2.56
#> # … with 209 more rows, and 23 more variables: pred_common_name <chr>,
#> #   pred_species <chr>, pred_rank <chr>, pred_aphia_id <int>,
#> #   pred_valid_name <chr>, pred_valid_aphia_id <int>,
#> #   pred_breeding_status <chr>, pred_age_group <chr>,
#> #   prey_orig_descr <chr>, prey_taxon <chr>, prey_rank <chr>,
#> #   prey_aphia_id <int>, prey_valid_name <chr>, prey_valid_aphia_id <int>,
#> #   prey_size <chr>, prey_age_group <chr>, freq_biomass <dbl>,
#> #   sample_size <dbl>, sample_type <chr>, ref_n <dbl>, ref_ids <chr>,
#> #   source <chr>, notes <chr>

filter prey

sbd_filter(prey_taxon = c("Cottidae", "Actinopterygii"))
#> # A tibble: 124 x 33
#>       id  year startyear endyear multiyear location latitude longitude
#>    <int> <dbl>     <dbl>   <dbl> <lgl>     <chr>       <dbl>     <dbl>
#>  1   360  1982      1982    1982 FALSE     Isle of…     56.2     -2.56
#>  2   366  1983      1983    1983 FALSE     Isle of…     56.2     -2.56
#>  3   374  1984      1984    1984 FALSE     Isle of…     56.2     -2.56
#>  4   899  1990      1990    1990 FALSE     Isle of…     56.2     -2.56
#>  5   911  1992      1992    1992 FALSE     Isle of…     56.2     -2.56
#>  6  1055  2006      2006    2006 FALSE     Isle of…     56.2     -2.56
#>  7  1197  2009      2006    2011 TRUE      Anglesey     53.3     -4.43
#>  8  1201  2009      2006    2011 TRUE      Bempton      54.1     -0.17
#>  9  1214  2009      2006    2011 TRUE      Bullers…     57.4     -1.82
#> 10  1218  2009      2006    2011 TRUE      Burravoe     60.6     -1.05
#> # … with 114 more rows, and 25 more variables: pred_common_name <chr>,
#> #   pred_species <chr>, pred_rank <chr>, pred_aphia_id <int>,
#> #   pred_valid_name <chr>, pred_valid_aphia_id <int>,
#> #   pred_breeding_status <chr>, pred_age_group <chr>,
#> #   prey_orig_descr <chr>, prey_taxon <chr>, prey_rank <chr>,
#> #   prey_aphia_id <int>, prey_valid_name <chr>, prey_valid_aphia_id <int>,
#> #   prey_size <chr>, prey_age_group <chr>, freq_occ <dbl>, freq_num <dbl>,
#> #   freq_biomass <dbl>, sample_size <dbl>, sample_type <chr>, ref_n <dbl>,
#> #   ref_ids <chr>, source <chr>, notes <chr>

filter year

sbd_filter(year = 1973:1976)
#> # A tibble: 207 x 33
#>       id  year startyear endyear multiyear location latitude longitude
#>    <int> <dbl>     <dbl>   <dbl> <lgl>     <chr>       <dbl>     <dbl>
#>  1     1  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  2     2  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  3     3  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  4     4  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  5     5  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  6     6  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  7     7  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#>  8    11  1974      1974    1974 FALSE     Fair Is…     59.6     -1.63
#>  9    12  1974      1974    1974 FALSE     Fair Is…     59.6     -1.63
#> 10    13  1974      1974    1974 FALSE     Fair Is…     59.6     -1.63
#> # … with 197 more rows, and 25 more variables: pred_common_name <chr>,
#> #   pred_species <chr>, pred_rank <chr>, pred_aphia_id <int>,
#> #   pred_valid_name <chr>, pred_valid_aphia_id <int>,
#> #   pred_breeding_status <chr>, pred_age_group <chr>,
#> #   prey_orig_descr <chr>, prey_taxon <chr>, prey_rank <chr>,
#> #   prey_aphia_id <int>, prey_valid_name <chr>, prey_valid_aphia_id <int>,
#> #   prey_size <chr>, prey_age_group <chr>, freq_occ <dbl>, freq_num <dbl>,
#> #   freq_biomass <dbl>, sample_size <dbl>, sample_type <chr>, ref_n <dbl>,
#> #   ref_ids <chr>, source <chr>, notes <chr>

filter multiple

sbd_filter(year = 1973:1976, pred_species = "Uria aalge", prey_taxon = "Gadidae")
#> # A tibble: 3 x 33
#>      id  year startyear endyear multiyear location latitude longitude
#>   <int> <dbl>     <dbl>   <dbl> <lgl>     <chr>       <dbl>     <dbl>
#> 1  1333  1973      1973    1973 FALSE     Co. Ker…     52.2     -9.75
#> 2  1699  1973      1973    1973 FALSE     Skomer       51.6     -5.29
#> 3  1704  1975      1975    1975 FALSE     Skomer       51.6     -5.29
#> # … with 25 more variables: pred_common_name <chr>, pred_species <chr>,
#> #   pred_rank <chr>, pred_aphia_id <int>, pred_valid_name <chr>,
#> #   pred_valid_aphia_id <int>, pred_breeding_status <chr>,
#> #   pred_age_group <chr>, prey_orig_descr <chr>, prey_taxon <chr>,
#> #   prey_rank <chr>, prey_aphia_id <int>, prey_valid_name <chr>,
#> #   prey_valid_aphia_id <int>, prey_size <chr>, prey_age_group <chr>,
#> #   freq_occ <dbl>, freq_num <dbl>, freq_biomass <dbl>, sample_size <dbl>,
#> #   sample_type <chr>, ref_n <dbl>, ref_ids <chr>, source <chr>,
#> #   notes <chr>

Plot data

There are additional helpers for interactive plotting of data

sbd_plot_predators(year = 1973)


To cite use

citation("seabirddietDB")
#> 
#> To cite package 'seabirddietDB' in publications use:
#> 
#>   Anna Krystalli, Agnes Olin, James Grecian and Ruedi Nager
#>   (2019). seabirddietDB: Seabird Diet Database. R package version
#>   0.0.1. https://github.com/annakrystalli/seabirddietDB
#> 
#> A BibTeX entry for LaTeX users is
#> 
#>   @Manual{,
#>     title = {seabirddietDB: Seabird Diet Database},
#>     author = {Anna Krystalli and Agnes Olin and James Grecian and Ruedi Nager},
#>     year = {2019},
#>     note = {R package version 0.0.1},
#>     url = {https://github.com/annakrystalli/seabirddietDB},
#>   }

About

Seabird Diet (stomach content) Database collected around the British Isles (1933 - 2017)

Resources

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE
MIT
LICENSE.md

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published