Wildlife road mortality data from SE surveys were collected each week from September 2019 through June 2021 for a total of 92 surveys. During most weeks in this time range, two observers in a Dodge Ram 1500 pickup truck (the passenger was seated in the front passenger seat) drove down 15-km transects on the roads state highway (SH) 48, SH 100, farm to market (FM) 106, and FM 510 in Cameron County, Texas USA ( Figure 1 ). Each individual road mortality survey encompassed all transects surveyed on a given day, surveying 60 km in total. All road mortality surveys were conducted weekly, with observers driving at 64 km/hr along transects. SH 48 and SH 100 were four-lane, divided highways with maximum speed limits of 121 km/hr. They were driven both in easterly and westerly directions during each survey as mortalities could not be seen in all lanes going only one way due to the presence of concrete traffic barriers. FM 106 and FM 510 were two-lane undivided roads with maximum speed limits of 97 km/hr and 89 km/hr respectively. They were driven in just one direction during each survey as mortalities could be seen in both lanes while going in either direction. The direction FM 106 and FM 510 were driven and the order all roads were driven were alternated weekly. The alternation lessened the chance of missing persistent mortalities with greater visibility while driving in one direction than the other. Surveys were conducted between 08:00 and 13:00.

Before participating in road mortality surveys and data collection, all observers were required to review photos and videos from a database of wildlife previously observed in the area on a computer. All trainees began road mortality surveys by conducting a practice survey with review from an experienced surveyor before any data was collected. Before beginning a survey, the number and identity of observers and road survey order of each week’s survey were recorded. While conducting surveys, the truck’s hazard lights and an additional lightbar (Code3 21TR, Code3, St. Louis, MO) mounted on a BackRack (BackRack, Oakville, Canada) rack behind and above the cab were used to enhance public and observer safety. When a carcass within 10 m of the road was observed, the driver stopped the vehicle on the road shoulder and the passenger (or driver, if solo) checked the ArcGIS collector (Esri Inc., Redlands, CA, USA) on a tablet computer (2019 Samsung Galaxy Tab A, Samsung Electronics America, Inc., Ridgefield Park, NJ, USA) to see if the mortality event was new or if it had been previously recorded. Carcasses that had been recorded in a previous week had their continued presence recorded. Analyses in this study included only the first records of mortalities. If new, the passenger (or driver, if solo) exited and used ArcGIS collector and tablet to take a picture of the carcass and record information. Data recorded included species (or most precise taxon) identification, latitude and longitude, location of carcass on the road (e.g., left or right lane), which road was being surveyed, time and date of collection. Helmets and reflective safety vests were worn while outside the vehicle and the data collector waited for a pause in traffic to collect data if the carcass was not on or past the right shoulder. While the passenger collected data, the driver kept watch for approaching traffic, to warn the data collector of oncoming traffic if necessary. For 2-lane roads, location on the road was recorded in terms of “north” and “south” as opposed to “left” and “right”. Several times during surveys precipitation noticeably hindered observation. In such cases, the driver pulled over until conditions became acceptable. On 4-lane roads, driving only in the right lane (unless necessary to switch to the left lane due to construction or another such issue) was crucial to obtaining consistent data. Given surveyors typically drive at a lower speed than the speed limit, slow vehicles ahead of the survey vehicle were typically not an issue. If a safety issue presented itself and passing a vehicle would mitigate the safety issue (such as a car driving slowly with hazards on), passing was performed. Otherwise, slowing down or even pulling over and waiting for a slow vehicle to move out of the area was preferred. On 2-lane roads, when driving under the speed limit, the survey vehicle was parked on the shoulder to let other vehicles pass to maintain community goodwill and for safety purposes.

Near the beginning of the COVID-19 pandemic, to impede its spread, the University of Texas Rio Grande Valley (UTRGV) issued restrictions on vehicle travel with more than one person until vaccinations became available. From March 2020 through May 2021, 58 SE surveys were performed. For 46 surveys in this period, only one observer performed the survey while another person drove behind them and monitored road safety. For the other 12 surveys during this period, two observers performed the survey; one observer was a trainee in these instances.

Daily two-way traffic count data on SH48 for weeks 4–27 were retrieved for station S236 from the Texas Department of Transportation (TxDOT) Statewide Traffic Analysis and Reporting System (https://www.txdot.gov/data-maps/traffic-count-maps/stars.html) and averaged by week. Weekly data for Pre, During and Post lockdown showed a substantial decrease occurring during the lockdown period as compared to Pre and Post lockdown periods ( Supplementary Figure 1 ). No traffic data were available for the other three survey roads.

Proportions of mortalities from each individual road did not all meet the assumption of normality (Shapiro-Wilk test, FM 510, P<0.01), so were normalized by applying the natural logarithmic function. Using ANOVA, no difference was found in the mean proportion of mortalities coming from SH 48 between observation periods (P=0.113). Differences were found for SH 100 (P<0.05) and FM 510 (P<0.05). However, post-hoc Tukey test showed differences only between DL and PostL on SH 100 (P<0.05, 95% CI [0.385, 2.3685]) and PreL and PostL on FM 510 (P<0.05, 95% CI [0.2461, 1.9752]). A Kruskal-Wallis H test (the data were not normal, Shapiro-Wilk, P<0.05 for all observation periods for FM 510 and PreL and PostL for SH 100 and transformation could not normalize the data) was also performed on the mortality counts per survey by road. This showed that mortalities per survey were not different across roads (P=0.186). Therefore, only observation period was included as an independent variable in analyses of whether a COVID-19 lockdown lowered the number of local wildlife road mortalities. During lockdown period, data were normal (Shapiro-Wilk, P>0.05). The assumption of homogeneity of variances was violated (Levene’s test, P<0.05), so a one-way Welch’s ANOVA was utilized. One-way Welch’s ANOVA showed that mean number of mortalities per survey were not different between the three observation periods (Welch’s F _{2, 13} = 2.542, P=0.116).

The resemblance matrix data had homogeneous dispersion (PERMDISP, F _{2, 21} = 0.85491, P >= 0.472) and PERMANOVA was run. No significant differences were found for each of the three combinations of observation periods; DL and PreL (P = 0.499); DL and PostL (P = 0.346); and PreL and PostL (P = 0.099). The SIMPER procedure ( Table 3 ) showed that both interactions involving the observation period itself (between DL and PreL and between DL and PostL) were more similar than the PreL and PostL interaction. For that interaction, lagomorphs contributed more (23.43% versus 18.23% for DL and PreL, and 18.73% for DL and PostL) and snakes contributed less to the total dissimilarity versus the other interactions (12.89% versus 20.80% for DL and PreL, and 20.51% for DL and PostL).

Pairwise PERMDISP was run for each dataset (“all,” “total,” and “size”) and each showed homogeneous dispersion between 1 and 2 observers for all species group counts (F _1,90 = 4.0155, P = 0.062) and for total mortality counts (F _{1, 90} = 0.24672, P = 0.621) but not for the “size” dataset (F _{1, 90} = 5.2896, P<.05). PERMANOVA was therefore only performed on the “all” and “total” datasets. There were differences in the centroids between 1 and 2 observers for both the “all” dataset (t = 1.6735, P<0.05) and the “total” dataset (t = 4.4155, P<0.005). SIMPER analysis of the transformed “all” dataset revealed that differences in numbers of mortalities observed with 1 observer versus 2 included contributions from both large (41.41%) and small (40.76%) animal mortalities ( Table 4 ).

Subsets of large, small, and total animal mortalities observed with 1 and 2 observers all failed tests of normality (Shapiro-Wilk, P<0.05), as did various transformations of the datasets. Therefore, independent-samples median tests were used to compare median mortality counts and Mann-Whitney U tests were used to compare mortality count distributions.

A difference (P ≤ 0.001) in the median number of mortalities recorded per survey with 1 observer versus 2 was found using an independent-samples median test ( Figure 2 ). The same test found such difference among both only large animals (P<0.01) and only small animals (P ≤ 0.001). Mann-Whitney U tests revealed significant differences in the distribution of the 1-observer data versus 2-observer data for all animals (U = 1617, z = 4.382, P ≤ 0.001), for just large animals (U = 1424, z = 2.879, P<0.01), and for just small animals (U = 1566, z = 4.020, P ≤ 0.001) ( Table 4 ).

The ANOVA on weekly survey mortality counts showed no significant differences in month (p=0.340) and year (p=0.346) indicating no strong seasonality variation in the number of mortalities observed.

An important finding of this study is that the COVID-19 lockdown mandated by Cameron County did not lower wildlife road mortalities as compared to before or after the lockdown, so the hypothesis that it did was not supported. Mortality counts did not differ between observation periods and were closest to differing between PreL and PostL. A reduction in average weekly traffic was evident on State Highway 48, similar to other studies of the COVID lockdowns (Bíl et al., 2021; Shilling et al., 2021). While our results contrast with another study that found reduced wildlife road mortalities across four USA states (California, Idaho, Maine, and Washington) with reduced traffic during COVID-19 lockdowns (Shilling et al., 2021). Based on monthly automobile insurance claims, Abraham and Mumma (2021) reported reduced traffic volumes during the COVID pandemic nationwide but despite this, traffic collisions were unchanged and wildlife vehicle collisions increased as the pandemic went on. Moreover, they found that rural areas away from city centers saw no change in wildlife vehicle collisions during the lockdown period, similar to the result of the present study. Indeed, traffic reduction has been shown to potentially lead wildlife to be less wary of traffic and attempt to cross roadways more often (Seiler and Helldin, 2006). In urban areas there was a trend of wildlife being detected closer to roads, based on GPS tracking data during the COVID-19 lockdowns (Tucker et al., 2023). This also could have happened with scavengers removing more roadkill on our study roads due to decreased traffic, however it could have increased their mortalities as well. However, iNaturalist observations around North American urban centers of bobcats and coyotes did not increase during the pandemic whereas puma (Puma concolor) sightings increased (Vardi et al., 2021). This finding of no change in sightings of smaller mammals such as bobcats and coyotes is similar to our findings for road mortalities in the present study. Reduced traffic during the lockdown on our study roads likely resulted in faster driving (Gargoum et al., 2016; Yasin et al., 2021; Abraham and Mumma, 2021), leaving wildlife less time to avoid oncoming vehicles. In addition, seasonal variation and movements of some species may have masked any changes in mortalities due to the lockdown, such as lagomorphs and snakes (Canova and Balestrieri, 2018; Mata et al., 2009). There were two types of roads surveyed in this study, four-lane divided highways and two-lane undivided highways. The four-lane highways had wider rights-of-way that could have resulted in missed mortality counts due to reduced mowing and visibility in these areas during the pandemic. A limitation of our study was the lack of traffic data for three of the roads, but this was unavoidable given the pandemic. Due to the lack of traffic counts on the smaller two-lane highways, traffic may have been higher than predicted on these roads leading to fewer changes in road mortalities.

There was a significant difference in observed mortalities between number of observers for large animals, small animals, and overall, supporting hypothesis 2. The difference was stronger for small animals than for large animals. As large animals are easier to see, they may be easier for a solo driver to spot, especially on the edge of their field of view (FOV) at any given moment. Foot surveys of birds and bats near wind turbines showed smaller species to have lower detection rates (Morrison, 2002). Surveys of road mortalities in Brazil both on foot and via SE surveys showed SE surveys involve lower detection rates than walking surveys, especially for smaller animals (Santos et al., 2016). Vehicle observers in a 3-year study of wildlife road mortalities on 5 major Tasmanian road networks failed to detect any frogs or small lizards despite their likely presence and despite over 15,000 km of total survey effort (Hobday and Minstrell, 2008). There was no difference in the number of felid mortalities, the most important target taxa for road mortality research in south Texas. If such species are the main aim of a project, choosing 1 observer over 2, safety considerations notwithstanding, may be preferred. Canids and artiodactyls, other taxa that are common conservation targets, contributed relatively little to the difference as well. The 1-observer and 2-observer datasets differed in months covered, with the 1-observer data being biased toward earlier in the year than the 2-observer data. Collectively, these findings suggest that seasonality may have played a role in the observed differences, so a longer study period would have been preferable, but not possible due to the unique circumstances of the COVID pandemic

The finding of a significant effect on observer number difference still does not explain our initial finding of no effect on road mortalities due to the reduced traffic during the lockdown period. One and two observer surveys occurred throughout the pre- and during-lockdown periods and were only strictly required two weeks into the lockdown period and post lockdown period. Due to the unequal distribution of one- and two-observer surveys during the COVID lockdown period, this was a confounding factor that could not be tested. However, if an effect of one observer surveys occurred during the lockdown period, this would have reduced the observed mortalities, an effect that we did not detect in our analyses.