A total of 3,829 mosquitoes were captured by all traps in the four field areas, corresponding to 18 collection-days/trap type/site (1,998 by BGM, 1,554 by BGM+VECT and 277 by STs). These were morphologically identified by trained entomologists: Ae. albopictus was the predominant species (90.2%) followed by Cx. pipiens (8.8%); other 36 specimens were collected (23 individuals belonging to other Aedes species, 12 Culiseta longiareolata and one not-identified specimen; Table 1). In Figure 2, box plots of Ae. albopictus and Cx. pipiens captures for each type of collection in each area are showed.
No relevant evidence of a depletion effect due to repeated captures at the sites was detected (Figure S1, Table S1).
Table 1. Total of Culicidae captured by three types of traps during thirty-six 48h-long trapping sessions in four areas across Italy and morphologically assigned to species and sex.
BGM trapping performance following VECTRACK sensor application
According to GLMM results, the total number of mosquitoes collected by BGM+VECT was on average 32.3% lower (95%CI: 2.1 – 53.1%, p-value: 0.038) than BGM (Table 2). Similar performances were estimated for the total number of Ae. albopictus adults captured (33.8%, 95%CI: 1.9 - 55.4%; p-values: 0.040) and of Ae. albopictus males (35.1%, 95% confidence interval: 2.3 - 56.8%, p-values: 0.038). However, we did not find evidence against the hypothesis of comparable trapping performance between the BGM and BGM+VECT when considering Ae. albopictus females, Cx. pipiens adults, Cx. pipiens males, and Cx. pipiens females (Table 2).
Results showed that the group of four STs/site consistently captured less mosquitoes than each BGM, irrespectively of the species or sex (Table 2). Interestingly, the BGM and BGM+VECT showed a multiplicative factor of 14.5 (95% confidence interval: 9.1 – 23.0) and 10.2 (95% confidence interval: 6.3 – 16.6) higher captures of Ae. albopictus females compared to a single ST. The capture rate of a single ST for Ae. albopictus (defined as the ratio between the number collected adult females and the number of adult females present within the flight range of this mosquito species) was previously estimated though mark-release-recapture experiments to be on average 1.838*10- 4 (Marini et al. 2010; Zardini et al. 2024). Consequently, according to our results, the capture rates of BGM and of BGM+VECT are expected to range from 1.678*10- 3 to 4.219*10- 3 and from 1.161*10- 3 to 3.045*10- 3, respectively.
Table 2. Results of the generalized linear mixed models assessing difference in the number of captured mosquitoes between trap types. The BG-Mosquitaire (BGM) trap is taken as reference and its average value is not reported. The reported parameter values and their 95% confidence intervals are exponentiated and represent the multiplicative factor by which the expected number of captured mosquitoes changes depending on the trap type.
VECTRACK algorithm performance in Aedes albopictus and Culex pipiens identification
To assess VECTRACK algorithm performance in the identification of mosquito species we focused only on the collections made by the BGM+VECT and compared the number of Ae. albopictus and Cx. pipiens morphologically identified by an expert operator with the identification provided by VECTRACK algorithm. Notably, the number of specimens belonging to other Culicidae species (not targeted by VECTRACK algorithm) was negligible (N=11).
Overall, VECTRACK algorithm counted 53 less Ae. albopictus and Cx. pipiens in the trap compared to an actual number of 1,543 morphologically identified specimens (Table 3). This difference amounts to a 3.4% of unidentified Ae. albopictus and Cx. pipiens with respect to the total captures VECTRACK algorithm counted 5% (N=69) Ae. albopictus less than the operator, while overestimated the number of Cx. pipiens by 10.4% (N=16). This amounts to a balance accuracy (as defined in González-Pérez et al. 2024) of 99.8% and 99.4%, respectively.
Table 3. Mosquitoes identified by operator-mediated morphological examination (M ID) or by VECTRACK algorithm (VT ID).
Overall, results show a very high correlation between the two identification methods applied across different sites and at different times for the total number of Ae. albopictus and Cx. pipiens (Spearman: 0.96, p-value <0.0001). Results from the regression model estimated a 1.03 (95%CI: 0.99-1.08) relationship between the overall number of adult mosquitoes identified by VECTRACK algorithm and by the expert entomologist, meaning that every 100 Ae. albopictus and Cx. pipiens identified by VECTRACK algorithm the expectation is to have captured on average 103 (SD: 8.5) Ae. albopictus and Cx. pipiens (Figure 3).
A similar good performance was obtained for Ae. albopictus with a correlation of 0.96 (p-value <0.0001) between the two identification methods. Overall, the number of morphologically identified Ae. albopictus was comparable to the number provided by VECTRACK algorithm (Chi-squared = 0.409, p-value = 0.52). Results from the linear model estimated a 1.05 (95%CI: 1.-1.09) relationship between the number of Ae. albopictus identified by VECTRACK and by the operator corresponding to an expectation of 105 (SD: 7.9) Ae. albopictus for every 100 Ae. albopictus identified by VECTRACK algorithm (Figure 3).
The number of identified Cx. pipiens was comparable between operator and VECTRACK algorithm (Chi-squared = 1.64, df = 1, p-value = 0.2). However, a less strong correlation was present (0.70, p-value <0.0001) with VECTRACK algorithm overestimating the number of Cx. pipiens and showing high residual uncertainty with respect of the number of captured Cx. pipiens. For every 100 Cx. pipiens identified by VECTRACK algorithm we expect to have captured on average 77 (SD: 5) Cx. pipiens (Figure 3).
VECTRACK algorithm performance in Aedes albopictus and Culex pipiens female and male identification
Overall, VECTRACK algorithm provides high balance accuracy for females (99.1% and 98.8% for Ae. albopictus and Cx. pipiens, respectively), while a lower value for males was observed (62.6% and 62.8%). In particular, VECTRACK algorithm underestimated the number of Ae. albopictus females (-22.9%) and overestimated the number of males (53.8%), while an opposite trend was observed for Cx. pipiens mosquitoes (23.3% overestimation of females and -6.4% underestimation of males).
In the case of Ae. albopictus, a very high correlation between estimates provided by operators and VECTRACK algorithm was found both for females (Spearman: 0.97; p-value<0.0001) and males (Spearman: 0.89 p-value<0.0001). However, a significant difference was found in the identification of sexes between the two identification methods (Chi-squared test = 66.1, p-value = <0.001). The misidentification of Ae. albopictus females and the overestimation of Ae. albopictus males was found to be systematic and consistent across sites.
The estimated relationship between Ae. albopictus females counts by VECTRACK algorithm and by operator-mediated morphological identification is 1.28 (95%CI: 1.23- 1.33) and was found to correct the estimate of VECTRACK algorithm providing an estimate of 128 (SD: 5.6) actual catches for every 100 females identified by VECTRACK algorithm (Figure 4). The estimated relationship between Ae. albopictus males counts by VECTRACK algorithm and by the operator is lower 0.66 (95%CI: 0.57- 0.76).
We did not find evidence against the hypothesis of a similar identification performance of Cx. pipiens sexes between VECTRACK algorithm and by operator-mediated morphological examination (Chi-squared = 1.17, p-value = 0.28). However, the correlation between estimates provided by the two identification methods was found to be lower for both females (Spearman: 0.67 p-value<0.0001) and males (0.37, p-value =0.025).