A road mortality record submitted to the Faune France platform is the result of a chain of events. Five steps separate the collision from the data entry, making it possible to estimate the true number of collisions by correcting for each one systematically.
Collisions in which the animal moves away after impact, remains lodged in the vehicle, or is thrown off the road cannot be quantified due to a lack of available information. The correction method therefore likely underestimates the true number of collisions, as it only estimates cases where the dead animal is present on the road surface.
see Román, J. et al. (2024) for the first estimates of this biasThe animal disappears from the road within 1–15 days depending on the species, through the action of scavengers, vehicles, or even passers-by who move carcasses.
Transportation Research Part D (2024)Passage frequency is predicted by a machine learning algorithm trained on the actual movements of a sample of participants, as well as national road usage statistics.
in preparationDetection rate varies with animal size. Experiments with Faune France contributors quantified detectability from a moving vehicle, and the proportion of records actually spotted from a car was estimated through a participant survey.
bioRxivThe main source of uncertainty — both very difficult to measure and highly variable across participants and species. The median reporting rate is estimated at around one in every two animals detected. The correction model assumes this rate equals 100% for charismatic species (lynx, wolf, ...).
HAL, chapter 4For many wildlife species, records from Faune France and other citizen science platforms are the only available window onto road mortality at a national scale. Without this imperfect citizen data, we would have no information on collisions for less frequently encountered species and for large portions of the road network.
But estimating precisely how many animals die on roads remains one of the most difficult challenges in quantitative ecology — and here we are doing it from the least standardised data source imaginable. The corrections presented here compensate for measurable biases, but some mechanisms remain beyond our reach: the influence of vehicle speed on carcass detection, the reporting habits specific to each participant, or cases where the animal simply never ends up on the road surface. These uncertainties accumulate at every step and can produce very wide estimation ranges.
These estimates make visible a form of mortality that was not: that of inconspicuous species, hard to spot on the road, and therefore often overshadowed in collision statistics dominated by large game such as roe deer and wild boar. They also help identify species for which mortality is alarming, and for which dedicated surveys and protective measures are needed.