Elephant expedition
This dataset contains images captured by 40 motion- and heat-triggered camera traps (Bushnell Aggressor No Glow) at forest-savanna transitions in the forest-savanna mosaic of Lopé National Park, Gabon. Cameras were attached to trees at 1-1.5 m above ground, focusing perpendicularly towards the eleph...
| Main Author: | |
|---|---|
| Format: | Dataset |
| Language: | English |
| Published: |
University of Oxford
2018
|
| _version_ | 1797056433141841920 |
|---|---|
| author | Cardoso, A |
| author2 | Cardoso, A |
| author_facet | Cardoso, A Cardoso, A |
| author_sort | Cardoso, A |
| collection | OXFORD |
| description | This dataset contains images captured by 40 motion- and heat-triggered camera traps (Bushnell Aggressor No Glow) at forest-savanna transitions in the forest-savanna mosaic of Lopé National Park, Gabon. Cameras were
attached to trees at 1-1.5 m above ground, focusing perpendicularly towards the elephant path, so as to be the most likely to capture a passing elephant, whilst also not being directly in the elephant’s way and inciting them to destroy the camera. The camera traps could be triggered up to once per second and ran from May 2016 to June 2017.
This resulted in 1.2 million images, making manual analysis by a small research team unfeasible. Images were thus classified by >10 000 interested members of the public (citizen scientists) via an interactive web interface (“Elephant Expedition”, created in collaboration with Zooniverse.org, available at https://www.zooniverse.org /projects/anabellecardoso/elephant-expedition). Images containing animals were retired after ten independent classifications, and the count of forest elephants recorded.
Each image was associated with up to ten potentially conflicting classifications, so we applied a plurality algorithm that assigned each image a single aggregated classification (Swanson et al. 2015, script available at https://github.com/zooniverse/Data-digging/tree/master/example_scripts/R_code/
survey-tasks/generalized). The assigned animal was the one with the highest number of “votes”, and the assigned number of elephants in the image was the median count.
Considering that elephants are readily distinguishable from other animals in the study site, that images with animals were classified by ten volunteers, and that the high trigger frequency resulted in multiple images usually being captured during the same elephant visit, we have a high level of confidence in the data obtained from citizen science classifications.
This data set contains the following:
1) Images from the camera traps, each with a unique label indicating at which site and at what time they were captured. Images are organised into folders according to which field season they were captured (field seasons 1, between 1 and 2, 2, and 3).
2) A map of the location of each camera trap. The site numbers on this map correspond to the location labels of the captured images (e.g. images labelled T1 are from transect site 1 on the map).
3) A metadata file containing information on which time periods each camera trap was functional and non-functional for, and where date stamps were captured incorrectly and how they were corrected.
4) The annotations (classifications) from citizen scientists associated with each image. These include:
i) "FINAL_elephant-expedition-classifications" and "FINAL_elephant-expedition-classifications.csv-flattened" which is the raw classification data from the Zooniverse project and the flattened version of this data respectively.
ii) "FINAL_elephant-expedition-subjects" which is a file linking the image names used in the classification file to the original image labels.
iii) "EE-aggregate" which is the aggregated classification dataset.
iv) "EE-aggregate-elephants" which is a subset of (iii) containing only elephants.
v) "EEdatecorrected_T1" is a subset of (iv) containing only data from transect site 1, where the dates have been corrected if there were any mistakes. Similarly for all other "EEdatecorrected_TX" files.
vi) "EE_merge" is the merged version of all the files in (v), so the final version of the aggregated, date corrected, citizen science classifications.
This data is currently access controlled. If you would like to make use of this data, please contact Anabelle Williamson Cardoso (anabellecardoso@gmail.com) to arrange the embargo be lifted.
These data is the intellectual property of Anabelle Williamson Cardoso. Rights to the data are also held by Agence Nationale des Parcs Nationaux (ANPN) du Gabon. |
| first_indexed | 2024-03-06T19:22:57Z |
| format | Dataset |
| id | oxford-uuid:1aba94dc-c49d-4e2a-8c41-424ed716bea6 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-06T19:22:57Z |
| publishDate | 2018 |
| publisher | University of Oxford |
| record_format | dspace |
| spelling | oxford-uuid:1aba94dc-c49d-4e2a-8c41-424ed716bea62022-03-26T10:56:27ZElephant expeditionDatasethttp://purl.org/coar/resource_type/c_ddb1uuid:1aba94dc-c49d-4e2a-8c41-424ed716bea6EnglishORA DepositUniversity of Oxford2018Cardoso, ACardoso, ACardoso, ACardoso, ACardoso, ACardoso, ACardoso, ACardoso, ACardoso, AMalhi, YOliveras, IAbernethy, KBond, WThis dataset contains images captured by 40 motion- and heat-triggered camera traps (Bushnell Aggressor No Glow) at forest-savanna transitions in the forest-savanna mosaic of Lopé National Park, Gabon. Cameras were attached to trees at 1-1.5 m above ground, focusing perpendicularly towards the elephant path, so as to be the most likely to capture a passing elephant, whilst also not being directly in the elephant’s way and inciting them to destroy the camera. The camera traps could be triggered up to once per second and ran from May 2016 to June 2017. This resulted in 1.2 million images, making manual analysis by a small research team unfeasible. Images were thus classified by >10 000 interested members of the public (citizen scientists) via an interactive web interface (“Elephant Expedition”, created in collaboration with Zooniverse.org, available at https://www.zooniverse.org /projects/anabellecardoso/elephant-expedition). Images containing animals were retired after ten independent classifications, and the count of forest elephants recorded. Each image was associated with up to ten potentially conflicting classifications, so we applied a plurality algorithm that assigned each image a single aggregated classification (Swanson et al. 2015, script available at https://github.com/zooniverse/Data-digging/tree/master/example_scripts/R_code/ survey-tasks/generalized). The assigned animal was the one with the highest number of “votes”, and the assigned number of elephants in the image was the median count. Considering that elephants are readily distinguishable from other animals in the study site, that images with animals were classified by ten volunteers, and that the high trigger frequency resulted in multiple images usually being captured during the same elephant visit, we have a high level of confidence in the data obtained from citizen science classifications. This data set contains the following: 1) Images from the camera traps, each with a unique label indicating at which site and at what time they were captured. Images are organised into folders according to which field season they were captured (field seasons 1, between 1 and 2, 2, and 3). 2) A map of the location of each camera trap. The site numbers on this map correspond to the location labels of the captured images (e.g. images labelled T1 are from transect site 1 on the map). 3) A metadata file containing information on which time periods each camera trap was functional and non-functional for, and where date stamps were captured incorrectly and how they were corrected. 4) The annotations (classifications) from citizen scientists associated with each image. These include: i) "FINAL_elephant-expedition-classifications" and "FINAL_elephant-expedition-classifications.csv-flattened" which is the raw classification data from the Zooniverse project and the flattened version of this data respectively. ii) "FINAL_elephant-expedition-subjects" which is a file linking the image names used in the classification file to the original image labels. iii) "EE-aggregate" which is the aggregated classification dataset. iv) "EE-aggregate-elephants" which is a subset of (iii) containing only elephants. v) "EEdatecorrected_T1" is a subset of (iv) containing only data from transect site 1, where the dates have been corrected if there were any mistakes. Similarly for all other "EEdatecorrected_TX" files. vi) "EE_merge" is the merged version of all the files in (v), so the final version of the aggregated, date corrected, citizen science classifications. This data is currently access controlled. If you would like to make use of this data, please contact Anabelle Williamson Cardoso (anabellecardoso@gmail.com) to arrange the embargo be lifted. These data is the intellectual property of Anabelle Williamson Cardoso. Rights to the data are also held by Agence Nationale des Parcs Nationaux (ANPN) du Gabon. |
| spellingShingle | Cardoso, A Elephant expedition |
| title | Elephant expedition |
| title_full | Elephant expedition |
| title_fullStr | Elephant expedition |
| title_full_unstemmed | Elephant expedition |
| title_short | Elephant expedition |
| title_sort | elephant expedition |
| work_keys_str_mv | AT cardosoa elephantexpedition |