Table 1
Summary of the questions and hypothesis outlined in this work. The listed research hypotheses are holistic alternative hypotheses for working null hypotheses (i.e., obtained data distribution deviates from an “all other things being equal” scenario). Thus, here they serve as methodological predictions rather than actual ecological hypotheses.
| QUESTION | HYPOTHESIS |
|---|---|
| How people welcomed the application? | User and observation number descriptive information: no specific hypothesis (H1) |
| How were the data distributed in space? | More data from urban than rural areas because more people live on the former (H2) |
| How were the data distributed in time? | More recordings from early summer when birds establish their breeding territories (H3) More recordings during days off than working days (H4) |
| How well did recordings match bird activity? | User recordings are not assumed to match bird activity as people are often active later (H5) |
| How reliable were user annotations? | More correct (true-positives, -negatives) than incorrect annotations (false-positives, -negatives) support annotation quality (H6) |
Figure 1
The user interface contained four tabs. (a) Record: the main tab where recordings are made. (b) Observations: shows the list of all recordings made by the user, including the species identifications. (c) Identifications: shows a list of all species found from user’s recordings. (d) Map: all identifications made by all users. The list can be filtered based on species and date. The original interface was in Finnish; here, the depiction contains English translations. Information on sensitive species is not shown to all users, but the user can always see his/her own recordings.
Figure 2
Comparison between heatmap of recordings and population density in Finland. (a) The coloured dots show the number of recordings made by users and greater blue dots show the locations of six passive acoustic monitoring (PAM) sites. (b) The map of Finland coloured by the population density (Statistics Finland 2022).
Figure 3
Distributions of summary statistics of distance to user-specific centre location for users with more than 50 recordings. The panels show the (a) minimum, (b) lower quartile, (c) median, (d) upper quartile, and (e) maximum distances to the user-specific centre locations and (f) the distribution of standard deviations of distance to the centre location.
Figure 4
Number of species observations per day for application (a) and passive acoustic monitoring (PAM) recordings (b). The dashed blue lines show the number of recordings made for each day. In panel a, the vertical grey lines show dates when the application was featured on national radio (12.4.2023) and television (10.5.2023). Dashed vertical grey lines indicate Sundays.
Figure 5
User retention. The temporal distribution of recordings starting from the day of a user’s first recording grouped by the total number of recordings made.
Figure 6
The recording activity during the day. Dashed curves show the hourly relative recording effort by passive acoustic monitoring and application (App) users. Solid curves show numbers of observed species divided by the total recording effort. Dotted brown curves show the prediction for application observations, if the number of observed birds would be equal to the product of bird singing activity (deduced from passive acoustic monitoring [PAM] species observations) and user recording activity. The prediction is normalized in two alternative ways: by setting the total number of observations to match with (a) application observations (higher curve) and (b) PAM observations (lower curve).
Figure 7
Species accumulation curves for users with at least 200 recordings. Red curves highlight three example users and blue curves show similar accumulation curves for passive acoustic monitoring (PAM) recordings. Every blue curve represents one sampling site.
Figure 8
Accuracy of 2,500 user annotations. (a) y-axis shows the proportion of negative annotations that did not contain the target species, and x-axis the proportion of positive annotations that contained the target species for the selected 50 users. (b) A confusion matrix for all annotations of the selected users. Top left corner shows true positives (the species suggested by the application was correct and correctly annotated), bottom right shows true negatives (the species suggested by the application was wrong and correctly annotated), bottom left shows false-positives (the species suggested by the application was wrong but annotated as correct), and top right shows false-negatives (the species suggested by the application was correct but annotated as wrong).