Fawn birthdays: From opportunistically sampled fawn rescue data to true wildlife demographic parameters

Spring mowing in May and June is one of the main causes of mortality of roe deer fawns in agricultural regions. Knowing the exact birth distribution of fawns is important to guide farmers in their pre-mowing precautions to avoid fawn deaths. Wildlife volunteers searching fields prior to mowing can act as citizen scientists by producing data sets of rescued fawns and their approximate age at find. However, due to weather-dependent searches, the corresponding birth distributions can be highly skewed. We simulated virtual field data to examine the shortcomings of such data sources and introduced two algorithms for reconstructing reliable birth distribution parameters (mean and standard deviation) based on skewed samples. We found that weather-dependent search data biased the calculated means and standard deviations by up to 14 and 5 days, respectively. However, the use of the proposed advanced algorithms (Grid Search and Machine Learning) resulted in better estimates of the sample means and standard deviations by reducing the root-mean-square error by 65% and 80% respectively. Furthermore, the Grid Search algorithm was able to capture birth distribution parameters based on real citizen science data in Bavaria, Germany, from 2021, which are close to the results of more systematic samples of the same year. The simulation exercise highlighted the shortcomings and discrepancies of using non-probabilistic samples, for example on the occasion of mowing activities, to study demographic parameters compared to the true simulated distribution. Yet, the proposed algorithms can address these drawbacks and potentially turn citizen science data into an important data source for wildlife studies. This could ultimately help reduce wildlife losses during the mowing season by better knowing the distribution of births in a region.