Studying the timing of the seasonal movements of migratory birds, known as migration phenology, is crucial for managing and conserving migratory bird populations. To effectively protect these species during these critical periods, it is vital to employ reliable methods for assessing their migration phenology. Citizen science, which involves the participation of skilled volunteers in scientific data gathering, is a valuable resource for migration studies. It allows the collection of large amounts of data across extensive geographic areas, overcoming some limitations of other datasets and analytical methods. In this study we analysed pre- and post-breeding migration phenology of 23 game bird species in Italy, using citizen science data from the www.Ornitho.it portal. These findings highlight the value of citizen science data in obtaining migration timing and the importance of employing multiple methods to estimate it. This approach is particularly valuable for species subject to hunting, which require well-informed management.

We retrieved the data available on the Ornitho.it databank for 23 species. 

The method is based on the approach first proposed by Ambrosini et al. (2014) and refined to fit partial migratory species and citizen science data by Ambrosini et al. (2023).

The study area (the Italian peninsula and Sardinia, Sicily and the minor Italian islands) was first divided into cells of 0.5° by 0.5° degrees (latitude – longitude). Cells containing fewer than 20 records collected in fewer than 10 different days in a year (the minimum required sample size) were merged with adjacent cells (full details are reported in the Supporting information). This procedure, identical to the one used in Ambrosini et al. (2023), ultimately generated cells of different sizes, inversely proportional to data density. These cells are represented in the grid maps. Data that were spatially isolated were also identified and discarded to avoid the generation of unreasonably large cells. The number of records per day was then transformed first into the proportion of records per calendar date and then into the cumulative proportions of records at each date and cell. These were then modelled using a binomial Generalized Linear Mixed Model (GLMM) with a cloglog link function and exponential spatial covariance structure to account for spatial autocorrelation in the data. Day-of-the-year (centred to its mean value) was the only fixed predictor, while the random part of the model included cell ID as a random factor. In addition, the random part included a random slope for the day-of-the-year within cell (i.e. random intercept and slope model).

The procedure briefly described above and reported in full detail in the Supporting information thus returns the date when a given proportion of observations is estimated at each cell after considering the observations expected from birds that are stationary in the cell, if any. These values were then spatially interpolated using a grid with 0.5° × 0.5° (latitude x longitude) cells, the inverse distance weighting algorithm, with weights calculated using the Shepard method (Shepard 1968), and a leave-one-out validation routine to choose the best power function for interpolation. Finally, the resulting interpolated map was downscaled to obtain the expected values at cells of 0.1° × 0.1° of size (latitude × longitude, approximately corresponding to 10 × 8 km) using the bilinear method. Since this procedure included a predefined set of parameters with arbitrarily selected values, we selected alternative values for these parameters and re-run the analyses using all 27 possible combinations (see the Supporting information for an overview of these parameters, their meanings, and the different values used). This produced 27 final maps per species, for each predicted proportion of arrived migrants (i.e. PRED values as outlined in the Supporting information) which were then averaged. These 27 maps for all 23 game species (621 in total) can be found in the mean maps folder.