ForestEyes

Current Results

The first ForestEyes' workflow covered areas in Rondônia state (year 2011) and Awá-Guajá Indigenous Reservation (early 2014). 104 tasks were made, being 72 to Rondônia and 32 to Awá-Guajá. The images were obtained by NASA's MODIS sensor. The other two workflows used Landsat-8 images of an area of Rondônia State in the years of 2013 (1028 tasks) and 2016 (1022 tasks). These Landsat-8 images are freely available at USGS EarthExplorer.

Table General Information shows the amount of answers received and the number of volunteers that have participated in these workflows.

Processing

For each task, the first 15 answers (excluding the ones sent by the same user) were considered and the classification of the task was given by majority vote. The Table shows the classifications given to MODIS workflow and the images show the classifications given to LANDSAT-8 segments workflows.

The segments classified as undefined or where happened a draw received a new segmentation and the new segments are available for volunteers' analysis and classification at LANDSAT-8 Segments 2013_2016 workflow.

With the amount of votes given for each class (Forest, Non-forest or Undefined), it is possible to calculate the level of difficulty of each task by using Shannon's Entropy (equation below) [1].

By considering the class given by majority vote as the correct one, it is possible to calculate the volunteers' score. It takes into account the number of answers given by the volunteer and the number of hits, which means the number of tasks that the volunteer answered the same as the majority [1]. The hit rate and score are calculated by the equations below.

The volunteers' ranking with the 20 first positions can be seen below.

References

[1] J. S. Arcanjo, E. F. P. Luz, A. L.Fazenda, and F. M.Ramos, “Methods for evaluating volunteers contributions in a deforestation detection citizen science project,” Future Generation Computer Systems, vol. 56, pp. 550557, 2016.