In order to minimize the negative impacts of roads on wildlife mortality and fragmentation, ecologists and road managers have been working together on assessing spatial patterns of roadkills, taxa most sensible, and road and landscape characteristics that influence roadkill numbers. Of special concern when analyzing these spatial patterns, is the location of roadkill hostspots, ie. segments of roads with clusters of wildlife mortality. The accuracy in the spatial definition of hotspots is of prime importance not only to conservation biologists, but also to road agencies and planners, as mitigation of roadways is usually expensive. Recently it has been shown that lower frequencies of road monitoring (longer intervals between samplings) may be responsible for losses of more than 50 % of roadkill numbers registered for several taxonomic groups, when compared with a daily sampling. This result highlights the need to account for other possible sources of inaccuracies when monitoring roadkills with varying sampling frequencies. Particularly important is the evaluation of the spatial accuracy of roadkill hotspot locations when different sampling efforts are implemented because inaccurate results may fail to detect “real” roadkill hotspots or can direct highly-cost mitigation measures to the inappropriate road sections. In the present study, we aim to assess the spatial discrepancy of hotspots location using four sampling frequencies (scenarios), and determine for which taxonomic groups is this spatial discrepancy most severe. We used a dataset of a one-year long roadkill daily survey, including 4453 individual records of vertebrate carcasses, for which survival time on the road is known. This dataset was arranged in five data matrices concerning different sampling frequencies: daily sampling (the baseline data), and four scenarios, 2-day interval, weekly, bi-weekly, and monthly sampling. We considered the global species data (all taxonomic groups together) and each of the 13 taxonomic groups considered for the analyses. For analyses, the road was divided in 500-m sections and hotspots were calculated according to Malo's method (using a Poisson distribution). We considered a threshold of 95 % and a corresponding minimum of two observations (roadkilled animals) in order to proceed with the analyses. In order to evaluate spatial discrepancy in hotspot location at road sections (presence/absence of hotspot) between daily and each of the four sampling scenarios, we used the Phi correlation. For global data, spatial discrepancy of hotspots increased most from weekly scenario onwards (phi weekly = 0.66, phi bi-weekly =0.61, phi monthly =0.58), while the 2-day scenario had the lowest discrepancy (phi 2-day =0.89). None of the four scenarios produced a hotspot map identical to the one obtained through daily survey, neither with global data nor with separated taxa. Even for the highest correlated scenario (2-day sampling), a different hotspot map was obtained for all studied taxa. Taxa with higher discrepancy in hotspot maps were bats, toads, salamanders, snakes and small mammals. Birds of prey, hedgehogs, carnivores, and lagomorphs had the lowest spatial discrepancy in hotspot maps. These results must be taken into account when planning roadkill monitoring programs, specially if we are dealing with small species.
