Fig. 2: Overview of the DT modelling strategy.

We parameterized a prior model by combining long-term bird observations with spatial and temporal predictors. a, Continuous recordings provide prior information about when birds vocalize, conditional on their presence. b, Long-term citizen science observations provide prior information about the timing of migration. c, Systematic transect line counts, as combined with data on land cover, forest structure and climatic predictors, provide prior information about the spatial distributions of birds. d–f, The continuously accumulating MK app data are used to update the detection model (d), the migration model (e) and the spatial distribution model (f) and, hence, knowledge of bird spatiotemporal distributions and singing activity. g, Probabilistic predictions by the three model components yield the probability that a given bird species is detected in a given MK app recording, as for this to happen (1) the bird should have returned from migration (or be resident), (2) the location should be part of the birds spatial distribution and (3) the bird should vocalize in a manner that leads to detection in the MK app.