Evaluating experimental design to sample mosquitoes Authors: Heloise Rozier, Pierre Gloaguen, François Septier, Sébastien Boyer Published: 2025 ABSTRACT AbstractMosquito study is crucial for public health, research, and vector control. We evaluated four understudied aspects of experimental design: (i) global efficiency of trapping devices—BG-Sentinel (BG) vs. CDC light traps (LT), (ii) temporal efficiency over a three-day sampling period, (iii) the impact of sampling duration on vector presence and abundance assessments, and (iv) site visit frequency for biodiversity surveys. Both traps were deployed across 10 Cambodian provinces, with collections conducted every 24 h over three-day periods from 2019 to 2021. A total of 1992 collections yielded 181,798 mosquitoes spanning 153 identified species. Using a hurdle generalized linear mixed model to address overdispersion and zero inflation, we found that 21 of 54 studied species showed a preference for one of the trapping device: LT captured 15 species more efficiently, while six species preferred BG. Anthropophilic vectors such as Aedes aegypti and Aedes albopictus were more attracted to BG, whereas LT captured higher species richness and zoophilic vectors. For six of 13 primary vector species, we showed that the capture efficiency declines from day one to day three on a three days sampling period. However, annual trends in presence and abundance remained consistent regardless of whether sampling lasted one, two, or three days. Finally, we quantified the probability of collecting a new species during subsequent visits, showing that this probability declines with visit sampling effort at rates dependent on trap type and sampling days duration. These findings provide empirical guidelines for optimizing mosquito surveillance strategies based on study objectives.Subject terms: Population dynamics, Environmental sciences Introduction Mosquitoes transmit a wide range of pathogens, and the diseases they spread contribute substantially to global morbidity and mortality1. As of April 30, 2024, the World Health Organization (WHO) has received reports of more than 7.6 million dengue cases worldwide, of which 3.4 million have been laboratory-confirmed. These included over 16,000 cases classified as severe dengue and more than 3,000 associated deaths2. Effective mosquito surveillance enables the monitoring of the abundance of multiple species across different temporal and spatial scales, facilitating the identification of vector species3. Such data are crucial for understanding mosquito diversity and vector population dynamics, ultimately guiding the implementation of targeted vector control measures to safeguard public health4. Entomological studies predominantly rely on two standardized trapping devices to ensure reproducibility: BG-Sentinel traps (BG), which mimic human odor plumes, and CDC light traps (LT). BG and LT are considered standard tools in mosquito surveillance utilizing attractants such as light, odor and visual cues5–7. However, their efficiency varies depending on mosquito species. LT are primarily used for collecting nocturnal mosquitoes, including Anopheles and Culex species, whereas BG, which employ chemical lures, are preferred for collecting anthropophilic species, including Aedes species8,9. Despite this widespread use, a lack of standardized sampling protocols has led to inconsistencies in mosquito surveillance efforts. Many studies have compared the efficiency of different trap types, yet no consensus has been reached on an optimal sampling strategy. In temperate regions, a study conducted in Spain compared capture efficiency of BG and CDC UV traps for Aedes albopictus and Culex pipiens using 100 collections10, concluding that BG was more effective for both species. Similarly, in Zhejiang province, China, a study compared the efficiency of BG and LT across five distinct biotopes with a total of 1,300 collections [Collection refers to a mosquito sampling event carried out using a specific trap at a given time and location. Each collection represents the group of mosquitoes captured during that single trapping event]11. The results highlighted the superior efficiency of BG for capturing Aedes albopictus, while both traps performed similarly for other mosquito species across all biotopes. In subtropical climates, a study in Guangzhou, China, compared BG, LT and oviposition trapping devices, finding that BG was overall more efficient in capturing Aedes albopictus and Culex quinquefasciatus6. In tropical regions, a study in Guinea assessed the performance of five different trap types over five days, with 30 collections per trap across five sites. BG were significantly more effective in capturing Aedes aegypti in rural, semi-rural and semi-urban sites12. In contrast, a study in Peru comparing LT and passive box traps in forests and inhabited areas concluded that UV-based light traps were highly effective for sampling a diverse array of mosquito species13. Summarizing these studies across different ecotypes and climates, BG appear to be more effective at capturing Aedes albopictus and Culex pipiens in temperate areas, Aedes albopictus and Culex quinquefasciatus in subtropical areas, and Aedes aegypti in tropical areas. However, LT have demonstrated greater efficiency in capturing a wider range of mosquito species in tropical environments. Cambodia’s tropical climate provides ideal conditions for the proliferation of several mosquito species14, 43 of which are known vectors of human pathogens15. Among these, Aedes aegypti and Aedes albopictus are the primary vectors of dengue virus, which causes recurrent outbreaks with significant public health impact in Cambodia16,17. However, to the best of our knowledge, no study has yet investigated the experimental design of mosquito sampling in this area. Based on an extensive survey, this work aims to conduct such an investigation. Our survey consists in 1992 collections (996 collections for both LT and BG trapping devices), obtained every 24 h during three consecutive days on different sites at different time of the year. These collections were carried out across diverse ecotypes, ranging from low to high entropy environments. Our primary objective is to evaluate the potential particularities of each trapping device in this Cambodian context. Moreover, the richness of our study (in terms of number of collections and number of counted species) allows us to address questions about common practices in sampling design for the studies of mosquitoes (both for the monitoring of vector species and biodiversity inventories). In medical entomology, a common experimental design consists of trapping over a three-day period18,19. This standard duration is based on the average duration of mosquito gonotrophic cycle, which lasts approximatively three days20,21. After obtaining a blood meal, female mosquitoes lay their eggs within two to three days and subsequently seek another host. The three-day sampling period is intended to maximize the capture of all host-seeking mosquitoes present at a given site. However, no statistical analysis has demonstrated that three-day sampling is significantly more effective than one- or two-day sampling. Given the importance of optimizing resource allocation in vector surveillance programs, evaluating the necessity of a three-day sampling period is critical. Reducing sampling duration could help prioritize other key factors, such as increasing the number of sampling sites. The final aspect of experimental design we examined is the frequency of site visits. Many of the studies reviewed here suffer from limitations, including small sample sizes22, restricted geographic scope, and a focus on a few key vector species. In contrast, mosquito scientific projects in Cambodia have collected a large number of specimens across diverse species and geographic locations15. Our aim is to provide recommendations for mosquito monitoring protocols by studying three aspects of experimental design: trap type (BG or LT), the effect of a three-day sampling duration, and the impact of the experiment duration (number of revisit) on assessing the diversity. The findings will enhance experimental design for mosquito population dynamics studies and support the development of effective, region-specific vector control strategies. Results Trapping device effect (presence) Among the 1992 collections, 153 species were identified (SI Appendix, Table S1). Of these, 70 species (listed in SI Appendix, Table S6) were collected by only one trapping device - 53 species exclusively in LT, with 71% represented by at least two mosquitoes, and 17 species only in BG (36% having more than two mosquitoes). Additionally, 83 species were found in both traps (Fig. 1A). These results suggest that LT captures a greater number of species than BG. Mosquitoes collected from 76 sites in Cambodia between 2019 and 2021. Left: Distribution of captures and species in our dataset. (A) Number of species found on a single trapping device—BioGents (BG) or Light Traps (LT)—or both. (B) Distribution of species abundances per trap. Right: Estimated association between trapping device and species. (C) Species that were significantly more detected (presence), captured (abundance), or both by a trapping device. Trapping device effect (abundance) A total of 247,826 of mosquitoes were collected, and among them, 181,798 mosquitoes were identified to the species level (Table 1). LT captured more mosquitoes and yielded a higher rate of species identification than BG. Among the species collected in BG, 11 species were represented by only one individual, five species by two to five individuals, and one species by six to 20 individuals (Fig. 1B). In contrast, LT captured 15 species represented by a single individual, 17 species with two to five individuals, and 15 species with six to 20 individuals, as well as an additional six species with more than 21 specimens. These findings indicate that LT not only captures a greater number of species but also a higher number of individuals than BG. Table 1Adult mosquitoes collected from 76 sites in Cambodia between 2019 and 2021.TrapTotal mosquitoesIdentifiedBG104,66866,831 ( 64 %)LT143,158114,967 ( 80 %) Adult mosquitoes collected from 76 sites in Cambodia between 2019 and 2021. Species-wise effect LT was more effective for 15 species, whereas BG performed better for six species (Fig. 1C). Specifically, for five species, the trap effect was observed in both presence detection and abundance measurement; for 14 species, the effect was observed only in presence detection; and for two species, only in abundance measurement. More specifically (estimates in SI Appendix, Tables S4, S5), our study indicates that BG was more efficient for Aedes albopictus and Culex quinquefasciatus (in both presence detection and abundance measurement), Aedes aegypti, Armigeres subalbatus, and Culex brevipalpis (presence only), and Aedes gardnerii (abundance only). Conversely, LT was more efficient for Culex gelidus, Culex vishnui, and Mansonia uniformis (presence and abundance), as well as Aedes vexans, Anopheles peditaeniatus, Coquillettidia crassipes, Culex bitaeniorhynchus, Culex nigropunctatus, Heizmannia demeilloni, Mansonia annulifera, Mansonia dives, Mansonia indiana, Mimomyia luzonensis, and Uranotaenia rampae (presence only), and Anopheles nigerrimus (abundance only). Context-dependent trap effects Some trapping effects appeared to depend on external factors, as interaction terms were retained (SI Appendix, Table S4). For example, the efficiency of BG for Culex quinquefasciatus varied with the season, with BG performing better overall and particularly during the dry season. Similarly, BG was significantly better at capturing Aedes gardnerii (in terms of abundance) only during the rainy season. For Culex vishnui and Heizmannia demeilloni, the trap effect depended on the project context (SI Appendix, Table S5). In Project B (primarily conducted in protected areas), LT was significantly more efficient than BG for Heizmannia demeilloni (presence) and Culex vishnui (abundance). Effect on vector species Neither trap showed a significant advantage in detecting Japanese encephalitis virus (JEV) vectors, including Culex tritaeniorhynchus, Culex fuscocephala, and Culex sitiens. Likewise, the Culex vishnui group (gathering at least 17 species including Culex tritaeniorhynchus, Culex vishnui and Culex pseudovishnui23) did not exhibit a clear preference for either type of trapping devices (Table 2). However, BG tended to attract anthropophilic species such as Aedes aegypti, Aedes albopictus, and Culex quinquefasciatus, whereas LT was more effective for zoophilic species. Overall, LT captured a greater diversity of species. Table 2Most frequent species among the 54 species analyzed using the Generalized Linear Mixed Model (GLMM) from mosquitoes collected at the 76 sites in Cambodia between 2019 and 2021.SpeciesOccurrenceAbundanceDisease/parasite potentially hostedCulex vishnui.g969106,497Aedes albopictus6566,785DENV, ZIKAV, CHIKV, JEV, RVFV, WNV, YFVCulex quinquefasciatus34116,018ZIKAV, JEV, RVFV, WNVCulex gelidus3153,851JEVArmigeres subalbatus259965JEV, FILWBMansonia uniformis2153,491RVFV, WNVCulex bitaeniorhynchus2142,524JEV, RVFV, FILBM, FILWBCoquillettidia crassipes2011,167Culex nigropunctatus167984Mansonia indiana144720Aedes gardnerii129661Culex tritaeniorhynchus1042,818JEVCulex fuscocephala90384JEVCulex brevipalpis89229Anopheles peditaeniatus8426,647Aedes aegypti83260DENV, ZIKAV, CHIKV, RVFV, WNV, YFVCulex vishnui751,359JEVCulex sinensis691,740Mansonia annulifera53191CHIKV, FILBMArmigeres dolichocephalus39157Culex sitiens37655JEVUranotaenia rampae37132Heizmannia demeilloni36163Aedes vexans3377CHIKV, ZIKAV, JEV, RVFVThe complete table can be found in the SI Appendix (Table S7). In bold (and underline) are the species for which the BioGents trap (BG) (and Light trap - LT) is significantly better than LT (and BG) in terms of presence detection or abundance measurement. The complete tables can be found in the SI Appendix (Tables S3, S4, S5). Abbreviations for diseases/parasites include: Dengue virus (DENV), Zika virus (ZIKV), Chikungunya virus (CHIKV), Japanese Encephalitis virus (JEV), Rift Valley Fever virus (RVFV), West Nile Virus (WNV), Yellow Fever virus (YFV), Plasmodium spp. (MAL), Filariasis from Brugia malayi (FILBM), and Filariasis from Wuchereria bancrofti (FILWB). Most frequent species among the 54 species analyzed using the Generalized Linear Mixed Model (GLMM) from mosquitoes collected at the 76 sites in Cambodia between 2019 and 2021. The complete table can be found in the SI Appendix (Table S7). In bold (and underline) are the species for which the BioGents trap (BG) (and Light trap - LT) is significantly better than LT (and BG) in terms of presence detection or abundance measurement. The complete tables can be found in the SI Appendix (Tables S3, S4, S5). Abbreviations for diseases/parasites include: Dengue virus (DENV), Zika virus (ZIKV), Chikungunya virus (CHIKV), Japanese Encephalitis virus (JEV), Rift Valley Fever virus (RVFV), West Nile Virus (WNV), Yellow Fever virus (YFV), Plasmodium spp. (MAL), Filariasis from Brugia malayi (FILBM), and Filariasis from Wuchereria bancrofti (FILWB). Biodiversity indices Species richness was significantly higher in LT than in BG, whereas BG yielded a significantly higher Shannon index than LT. Notably, using both types of trapping devices together resulted in a higher species richness index but a Shannon index similar to LT (Fig. 2; estimates in SI Appendix, Table S8). LT appears to attract a higher number of species overall and can collect multiple species at once. Estimated adjusted mean (and 95% confidence interval) of each biodiversity index per trapping device based on mosquitoes collected from 76 sites in Cambodia between 2019 and 2021. Significance of differences is indicated (* for significant differences, “ns” for not significant). Day index effect As shown in Table 3 (A), the day index significantly affected the observed abundance for six of the 13 vector species, with more mosquitoes collected on the first day for both trapping devices. Table 3 (B) quantifies the distribution of first-day detections for the 13 vector species. For example, considering both BG and LT, Aedes albopictus was first detected on the first day in 79% of cases (out of 228 sessions where it was observed), 14% on the second day, and 7% on the third day. With LT alone, Aedes albopictus was missed in 34% of cases when sampling was conducted for only one day, compared to 10% with two days of sampling. With BG, the detection failure rate dropped from 25% with one-day sampling to only 1% with two days. These results suggest that trap efficiency decreases over consecutive days (additional results in SI Appendix, Table S9). Table 3For mosquitoes collected from 76 sites in Cambodia between 2019 and 2021: (A) Total abundance per day for 13 vector species. Species for which the day index effect is significant for either presence or abundance are highlighted in underline (see SI Appendix, Table S9). (B) Sampling day of first capture per species for each trapping device. For example, of the 55 collections where Aedes aegypti was found, 55% occurred on the first day.SpeciesBG+LTLTBGDay 1Day 2Day 3Day 1Day 2Day 3Day 1Day 2Day 3(A)Aedes aegypti81110694357710764Aedes albopictus327020371478597417331267316201147Aedes vexans441419391419500Armigeres subalbatus390320255846970306251185Culex bitaeniorhynchus18523703021304249186548121116Culex fuscocephala1738512614073116331210Culex gelidus18569631032140575995545120477Culex quinquefasciatus62745023472126416394601048604627Culex sitiens51013312425128118551Culex tritaeniorhynchus1115127742675311884133628913Culex vishnui59238738040625436918613311Mansonia annulifera130332885132545203Mansonia uniformis1715940836155079280216514834(B)Aedes aegypti30 (55%)19 (35%)6 (11%)3 (50%)2 (33%)1 (17%)27 (54%)17 (34%)6 (12%)Aedes albopictus179 (79%)33 (14%)16 (7%)90 (66%)32 (23%)15 (11%)155 (75%)36 (17%)15 (7%)Aedes vexans16 (59%)6 (22%)5 (19%)12 (48%)8 (32%)5 (20%)5 (100%)0 (0%)0 (0%)Armigeres subalbatus75 (60%)32 (26%)18 (14%)36 (56%)16 (25%)12 (19%)58 (59%)26 (27%)14 (14%)Culex bitaeniorhynchus71 (59%)32 (27%)17 (14%)54 (55%)30 (31%)14 (14%)44 (70%)13 (21%)6 (10%)Culex fuscocephala36 (56%)15 (23%)13 (20%)25 (53%)12 (26%)10 (21%)14 (50%)7 (25%)7 (25%)Culex gelidus105 (72%)31 (21%)10 (7%)73 (62%)33 (28%)11 (9%)60 (68%)19 (22%)9 (10%)Culex quinquefasciatus101 (72%)24 (17%)16 (11%)28 (47%)21 (35%)11 (18%)91 (75%)17 (14%)14 (11%)Culex sitiens13 (45%)10 (34%)2 (7%)10 (50%)8 (40%)2 (10%)6 (55%)4 (36%)1 (9%)Culex tritaeniorhynchus48 (69%)13 (19%)5 (7%)27 (60%)13 (29%)4 (9%)32 (74%)6 (14%)5 (12%)Culex vishnui17 (77%)1 (5%)3 (14%)15 (71%)2 (10%)3 (14%)13 (87%)2 (13%)0 (0%)Mansonia annulifera26 (60%)7 (16%)6 (14%)20 (61%)6 (18%)5 (15%)11 (73%)1 (7%)2 (13%)Mansonia uniformis77 (73%)19 (18%)7 (7%)65 (68%)21 (22%)10 (10%)32 (64%)12 (24%)6 (12%) For mosquitoes collected from 76 sites in Cambodia between 2019 and 2021: (A) Total abundance per day for 13 vector species. Species for which the day index effect is significant for either presence or abundance are highlighted in underline (see SI Appendix, Table S9). (B) Sampling day of first capture per species for each trapping device. For example, of the 55 collections where Aedes aegypti was found, 55% occurred on the first day. Sampling effort and temporal dynamics No significant differences were observed for the presence and abundance for the six vector species based on models using 1-day, 2-day, or full 3-day sampling per session (Fig. 3). The results suggest that a single day of sampling per session may be sufficient for estimating mosquito population dynamics. The month effect depends on the chosen sampling days for the main vector species among mosquitoes collected from 76 sites in Cambodia between 2019 and 2021. Estimates and 95% confidence intervals for the month covariate are shown for three models, where the dependent variable is presence/abundance from one-day sampling (blue), two-day sampling (red), and three-day sampling (green). Number of visits for biodiversity assessment In the perspective of species inventory, less number of visits is needed with LT and BG+LT than it is for BG to detect a higher number of new species (Fig. 4 top left). Most curves reached a plateau around the 300th session. Sampling for only one day instead of three would have resulted in approximately 25 species remaining undetected (regardless of trap type). BG+LT captured sooner all the 13 vector species either with 1-, 2- or 3-days sampling, compared to a single trap (Fig. 4 top right). LT seems to capture a bit earlier all the vector species compared to BG. However, using LT only one day seems to delay the detection of all vector species. At the site level, similar cumulative richness curves indicate that for 23 of 30 sites (with at least five revisits), vector species richness reached a plateau after the fourth revisit (SI Appendix Fig. S7). Sampling duration appears to affect the likelihood of reaching maximum richness. In fact, using both BG and LT with one-day sampling results in lower final richness at 11 of 30 sites. A GLMM analysis (Fig. 4 bottom) estimated the probability of detecting a new species per visit for different habitat types (forest, inhabited, rice field), using both traps over three days. Across all habitats, this probability dropped below 50% after four visits and below 20% after seven visits. Residual diagnostics confirmed the validity of the GLMM (SI Appendix, Fig. S8), supporting its utility in determining whether additional visits are necessary. As expected, the rate of species detection decay varied by type of trapping device and sampling effort (SI Appendix, Figs. S9, S10). Same study for species richness is in SI Appendix, Figs. S2 - S6. (A) Cumulative species richness and cumulative vector species richness throughout sampling sessions based on mosquitoes collected from 76 sites in Cambodia between 2019 and 2021. Cumulative richness was calculated separately for each combination of trap type (BG in red, LT in blue, BG+LT in green) and sampling days (circle for 1 day, square for 2 days, triangle for 3 days). The graph shows when each combination of trap and sampling days reaches the plateau of the 13 species discovered. For instance, the plateau is reached at the 237th session when using LT for one day. (B) Fitted probabilities of finding a new vector species throughout visits. The figures are grouped by habitat: Forest, Inhabited area, and Rice field, with different colors representing different sites. Only curves from experiments using both traps (BG+LT) over three days are displayed. All fitted probabilities for all sites, days, and traps are shown in SI Appendix Figure S11. Discussion GLMM approach For 28% of collections, no species were caught. On the 1992 collections, Aedes albopictus was found 33% of the time and Culex quinquefasciatus 17% of the time. The reason behind these few occurrences is that according to the niche theory, species have environmental preferences24. Hence, if observations are sampled through various environments, there is a high chance of not observing certain species at all in some places. Typically, if the captures are made through an ecological gradient and the conditions vary smoothly along it, the species may become rarer until they are completely absent. The specificity of zero-inflated data is not always taken into account, and some studies, to get rid of the problem, aggregate data. By aggregating data we lose information on the collection sampling day, date or site. When working at the scale of Cambodia, we account for the inflation of zero values. However, these zeros are appropriately handled by the hurdle GLMM, which explicitly models the presence of zeros separately from the nonzero values. The hurdle framework assumes that zeros arise from a distinct process, typically a binary hurdle model that determines whether an observation is zero or positive. Once past this hurdle, a separate model governs the distribution of nonzero values. The key advantage of using a hurdle GLMM in this context is its ability to differentiate between the probability of occurrence (whether the species is seen or not) and the intensity (how much when it is seen). This is particularly useful in cases where many sites have zero values due to structural reasons, such as environmental constraints or sampling issues rather than due to random chance. By incorporating mixed effects, the model allows us to assume that common patterns exist across the sites. While sites may have unique characteristics, they share underlying factors influencing both the presence of zeros and the distribution of nonzero values. The random effects capture site-specific variability, while fixed effects account for broader trends across Cambodia. The GLMM approach we used goes along with other studies on trap efficiency. A study in Zhejiang Province, China, compared BG and LT with a zero-inflated negative binomial model11, and another study conducted in la Réunion island, France, compared BG with different baits with a hurdle GLMM25. The key difference with these two zero-inflated models is how they handle zeros26. The ZI model distinguishes between “structural zeros” and “sampling zeros”. “Structural zeros” occur when the species is absent from the site, meaning positive counts are impossible. “Sampling zeros” occur when the species is present but was not observed during the sampling period. One limitation of the hurdle model is that it cannot compare collections with zero abundance to those with a strictly positive abundance. However, if a trap captures a large number of specimens in the area, we may want to say the species is present and then compare abundance values between traps from a same session. As an example, we can observe the species Culex tritaeniorhynchus, where in a session, we caught a big number of specimens only in LT (Table 4). By choosing a hurdle model instead of a zero-inflated model, we decided not to account for this specificity of the data. In fact, it is possible that these events result in the swarms of thousands of mosquitoes including multiple species27. Swarms may be more drawn to LT because swarming is a mating phase, during which females are not focused on feeding28. Then, comparing a zero to a large number may bias the explanation of the attraction of one species to one type of trap. Finally, our approach does not explore the relationships between different species. A joint species distribution modeling approach could address this gap, but the selection of species to include in such analyses would need to be carefully discussed29. Table 4Abundances of Culex tritaeniorhynchus over three specific sessions from mosquitoes collected at 76 sites in Cambodia between 2019 and 2021.TrapSession 463Session 616Session 434BG001BG000BG000LT4800LT3362460LT34700 Abundances of Culex tritaeniorhynchus over three specific sessions from mosquitoes collected at 76 sites in Cambodia between 2019 and 2021. Trapping device effect For three well studied antropophagic species, namely Aedes albopictus, Culex quinquefasciatus and Aedes aegypti, our results confirm the previous results of literature6,10–13, showing a better efficiency of BG. Additional results in SI Appendix, Tables S14, S15 confirm also the conclusions for Aedes genus to be more attracted by BG and Anopheles genus by LT (8,9). Culex genus though was not found to be significantly attracted by a specific type of trap. In 2018 in West Bengal, India, BG and LT were deployed for 12h period between 6:00 PM and 6:00 AM for night collections30. The study suggests that BG is more efficient than LT for collecting more Armigeres subalbatus, a JEV vector. On the other hand, the study conducted in Zhejiang Province11, China did not show a significant difference between BG and LT in capturing Armigeres subalbatus, with also night collections. We showed that BG is significantly better at detecting the presence of Armigeres subalbatus than LT. The study in China showed the LT caught significantly more specimens of Culex tritaeniorhynchus each time, but with a p-value = 0.0188, a p-value similar to the one we found for abundance p-value=0.0215 (SI Appendix Table S3). To our knowledge, our study is the first to provide the details of trap effect (BG against LT) on presence and abundance, all the studies cited here analysing the average density. Our study highlights differences in trap efficiency for 17 other species for which no such results exist in the literature, 15 of which are more efficiently sampled by LT. In particular, Culex bitaniorhynchus and Culex vishnui, vectors of JEV, were significantly more attracted by LT. This study then highlights the importance of such LT when the study focuses on the wider community of zoo-anthropohilic species, including some known vectors of pathogens to human. It was noted that LT led to more identified individuals than BG (Table 1). Biodiversity indices LT caught a larger number of species, resulting in a higher measure of species richness among sites, as they do not focus on anthropophagic species. Still, BG caught species not seen within LT, suggesting that both traps should be set when focusing this biodiversity index. The Shannon index was higher in BG compared to LT and also compared to LT and BG combined. A high Shannon index indicates an equilibrate repartition of the species. As LT capture rarer species, this can mechanically lowers the measure of Shannon index. For example, if we caught four species, the following proportions 0.7, 0.2, 0.0009, 0.0991 would give in a lower Shannon index than these proportions 0.7, 0.1, 0.1, 0.1. Similarly, combining both BG and LT would make us add to the equal proportion of the most abundant species, the rare species, making some species predominant and then lowering the Shannon index. These findings underline the importance of evaluating both the Shannon index and the species richness as diversity indicator31. A study on efficiency of mosquito traps (Mosquito Magnet-Pro, the Mosquito Magnet-X) showed that the trap yielding greatest species richness did not capture the greatest diversity of mosquitoes32. In a study conducted in Florida, the Shannon indices measured in BG and LT were similar33. There is plenty of indices measuring diversity but the most used being Shannon (additional information explaining our choice can be found SI Appendix Entomological indices choice explanation). Although we observed a similar Shannon index for the LT and the combined BG+LT, this does not mean the two different types of traps captured the same species communities. Nevertheless, since we had many rare species, it was still useful to look at the diversity index if one wants to identify environments that have potentially not been sampled correctly, for instance. A potentially interesting approach to assess species community similarity could involve the Morisita-Horn index34. However, this index requires non-zero species counts and thus aggregation of data. Then this index is not suitable for this context where we want to compare the two types of trap accounting the other characteristics of the data (season, location, etc.). Accounting for the circadian cycle For Culex tritaeniorhynchus, swarms formation begin at dusk and end at the beginning of the day35. Intuitively, one might assume the LT to be less effective during daylight hours. In our experiment, both traps were deployed for 24 h, so our model did not account for the moment of the day during the trapping. It raises the question of whether the LT is less efficient during the day. For 287 collections, the traps were set for 12 h at night and 12 h during the day. As a result, we have the total abundance of specimens collected by type of trap the night and the day (SI Appendix Table S13) and we found no significant interaction between trap type and time of day (day or night), nor did we observe a substantial effect of time of day on the total abundance of identified species (SI Appendix Table S12). This suggests that it is more effective to leave the traps deployed for 24 h rather than limiting them to 12 h. A lot of studies use an experimental design based on night collections11,30. Future studies could evaluate alternative trap designs, such as assessing the impact of removing dry ice, which may attract mosquitoes to the LT even during the day. Additionally, an interesting avenue for research would be to analyze the effect of LT height on species diversity. Effort effect on dynamics assessment We observed that for some species, fewer specimens were collected on the second day compared to the first. For effort effect, ie month, we did not see differences in the trend of abundance throughout the year for the main vector species. This suggests we could rather choose a one day sampling but sample every month or twice a month to study the dynamics of main vector species more deeply. Number of visits for biodiversity assessment We proposed using the outputs of a GLM to quantify the rate of decay of the probability of discovering a new vector species on a given site for a new visit. The surveillance of vector species richness is crucial in understanding vector-borne diseases36. Findings suggest that when vector species richness is sufficiently high, it is more probable that disease risk will rise rather than decline37,38. The hypothesis relies on ecological niche complementarity among species. Less competent vectors may contribute to disease transmission by occupying the spatial or temporal gaps left by dominant vectors38. Methods Data Study area Our data consists of abundance monitoring across 76 different sites throughout 10 provinces in Cambodia. Data comes from two monitoring projects, which will be named Project A (72 % of the collections) and project B in the rest of this work. Collections from Project A were realized along an ecological gradient, going from forest to urban areas, while Project B was conducted more in protected areas, limiting anthropic impacts. Both projects were conducted across different provinces of Cambodia (Fig. 5). The spatial distribution of the 76 mosquito sampling sites in Cambodia between 2019 and 2021. Two separate monitoring approaches were conducted, named Project A (72% of samples) and Project B (28% of samples). For specific information about habitat type, see Table S11 in SI Appendix. The map was generated using R version 4.3.2 (https://www.r-project.org/). The shapefile was downloaded from https://data.humdata.org/dataset/cod-ab-khm. Trapping devices In this study we consider two trapping devices, the light trap (LT) and the BG-sentinel trap (BG): the LT consisted of a light source powered by a 6V battery, and the BG consisted of a fan releasing attractants, powered by a 12V battery. LT were used with dry ice, such as BG used with both BG-Lure, a synthetic attractant that releases human skin-like odors, and dry ice. In both devices, dry ice serves as a source of carbon dioxide (CO) to attract mosquitoes. Sampling protocol At each of the 332 sessions [A session is defined as the combination of a site location and a three-days period], two trapping devices were deployed. Three collections were then performed, respectively 24, 48, 72 h after trap deployment. This then results in six collections per session for an overall of 1992 collections. Field workers changed nets, dry ice and batteries every 24 h to ensure consistency in trapping efficiency during the session. Each of the 76 sites was sampled between two and seven times on different dates, usually spaced approximately two months apart. The collections occurred between 2019 and 2021, having 888 collections during the dry season (November to March), and 1,104 collections during the rainy season (May to October). No sampling was done during April due to the hot weather. Among all these sessions, 21 involved captures every 12 h, which were used to assess a potential day-night impact on the efficiency of the traps (see the discussion section). Statistical framework Hurdle model for count data Overall, in both projects, 153 species were identified (see SI Appendix Table S1, for a complete list). For each species and each collection, the number of individuals is counted, resulting in a abundance matrix. Our main objective is to assess how our measures of abundances are linked to the characteristics of the sampling which consist in trapping device (BG or LT), season (dry or rainy), day index (first, second or third day of sampling) and the project (A or B). Days are aggregated for diversity indices. This question is explored at the species level using generalized linear models (GLM). The species studied with the GLM approach are the ones that have at least two individuals found in each modalities/combinations of modalities (these 54 species are listed in SI Appendix, Table S7). For each species, two models will be considered: A logistic regression to model the probability of presence of the species; A logistic regression to model the probability of presence of the species; A GLM on positive counts to model the observed abundance provided that at least one was captured. A GLM on positive counts to model the observed abundance provided that at least one was captured. The combination of these two models is known as a hurdle model39. Formally, for each species (in ), session (in ) and collection (in ) for a given session, we denote the number of individuals captured. It is assumed that: where is a species specific distribution on strictly positive integers, is its expectation, and is a dispersion parameter. In this article, the truncated Poisson and truncated negative binomial distributions are the possible choices for each species. Parameters and are linked to a (potentially different) subset of the measured covariates. The selection of the distributions and covariates (with potential interactions) is made using the BIC criterion (complete list of tested formulas is given in SI Appendix, Table S2). The trapping device covariate is always included as it is the main purpose of our study (and its significance will be tested after adjustement). In order to account for correlations between counts of a same session, a (species-specific) session effect was included for each model. This session effect was included as a random effect, resulting in generalized linear mixed model (GLMM). Biodiversity indices In addition to these single species count, we measured the diversity at the collection level through two different indices: The species richness: the total number of species present in a collection (denoted ). The species richness: the total number of species present in a collection (denoted ). The Shannon entropy, defined by Lande40, is given by: where is the proportion of individuals from species over all individuals found in the collection. The Shannon entropy, defined by Lande40, is given by: where is the proportion of individuals from species over all individuals found in the collection. It is worth noting that the Shannon entropy ranges between 0 and . Small values correspond to cases where a single species is predominant (being 0 if only one species is present), while the maximum value is reached when all species are equally represented. Therefore, a collection will be “highly diverse” in terms of Shannon entropy if it has a large number of species and all of them are equally represented. As for species counts, each biodiversity index is linked to covariates through a hurdle model (as for 28% of collections, no species were caught), using a count distribution for the positive part of species richness and a gamma distribution for the positive part of Shannon entropy. Trapping device effect For all models, the significance (at a level 5%) of the estimated differences between the effects of the trapping devices is tested. The test is performed over adjusted means, in order to integrate over other covariate effects. Moreover, since this test is done for every species, a Bonferroni correction for multiple tests is applied to maintain a global type I error of at most 5%. In practice, this means that a test is said to be significant if the corresponding p-value is below . Effort effect on dynamics assessment For 13 vector species (namely Aedes aegypti, Aedes albopictus, Aedes vexans, Armigeres subalbatus, Culex bitaeniorhynchus, Culex fuscocephala, Culex gelidus, Culex quinquefasciatus, Culex sitiens, Culex tritaeniorhynchus, Culex vishnui, Mansonia annulifera, Mansonia uniformis), the importance of the day index in the model is tested to investigate whether the efficiency of a given trap remains constant during a session. Because the species should be observed sufficiently within the three modalities of the day index, only 13 vector species have been selected. In addition, we evaluate the temporal dynamics of vector species in the study area. The aim here is to understand if one, two or three days sampling allowed to estimate a same temporal dynamics. To address this, we estimate the effect of the month on the presence and abundance of the main vector species in Cambodia (namely Aedes albopictus, Culex quinquefasciatus, Culex gelidus, Mansonia uniformis, Culex tritaeniorhynchus, Culex fuscocephala), and on the species richness and the vector species richness. Furthermore, we examine whether a sampling period (effort) shorter than three days would be sufficient to estimate temporal dynamics. To achieve this, a model is fitted using abundance data recorded over one, two, or three days, with the three-day sampling providing the highest accuracy but also requiring the most resources. Because the species should be observed sufficiently throughout the year, only six vector species have been selected. Number of visits for biodiversity assessment Finally, we investigate the importance of repeated visits to the same site to assess its species richness. For the sites visited between five and seven times in this study (SI Appendix, Table S10), we evaluate the relationship between the number of new species (i.e. species observed for the first time in our data for a given site) per visit and the visit number. This is done using a negative binomial GLMM, modeling the number of new species while accounting for the month of each visit and including the site as a random effect. Conclusion The BG is particularly effective at capturing well-known anthropophagous species such as Aedes aegypti, Aedes albopictus, and Culex quinquefasciatus. In contrast, the LT is better suited for capturing a greater richness of species, including rare ones. Combining both LT and BG is even better for describing species richness. Researchers interested in studying vector species, are advised to consult the detailed species list to determine if the LT might be more suitable for their study objectives. Some species showed no preference for either trap type, so using only one type of trap may help reduce costs when focusing on these species. These findings could be further refined by increasing sampling efforts in underrepresented ecosystems, which would help to improve the robustness of the results and potentially uncover new patterns in species behavior and trap preferences. The trend in the temporal variation of both the presence and the abundance of six vector species is the same either if we sample one day, two days or three days. This conclusion guides our decision to a one day collection strategy to sample vector species. The latter can help reduce costs or can eventually be performed more often throughout the year or at a larger number of sites to analyze more deeply the spatial distribution and temporal dynamics of vector species (Table 5). This decision though, can be nuanced, indeed in Table 3 (B), that for 13 vector species, we missed the species between 19% and 45% of the time by sampling only one day with BG+LT. When sampling two days, we missed the species between 7% and 19% of the time. For describing the vector species richness, we advise not visiting the site more than four times since we have less than 50% chances to discover a new vector species, regardless of the habitat. To conclude, Table 5 summarizes our proposed guidelines based on this study. Table 5Experimental design recommendation depending on the entomological target.TargetProposed solutionDENV vectorsBG on one dayJEV vectorsCombining both LT and BG on one daySpecies richnessCombining both LT and BG on one day (visit the site 5 times)Vectors richnessCombining both LT and BG on one day (visit the site 4 times)Trap recommendation is based on Table 2, sampling duration is based on Fig. 3 and visit frequency is based on Fig. 4b (for vector richness) and SI Appendix, Fig. S5 (for species richness). Additional figures show that the goodness of fit seems reasonable both for vector species and all species richness (SI Appendix, Figs. S1, S3, S4). Experimental design recommendation depending on the entomological target. Trap recommendation is based on Table 2, sampling duration is based on Fig. 3 and visit frequency is based on Fig. 4b (for vector richness) and SI Appendix, Fig. S5 (for species richness). Additional figures show that the goodness of fit seems reasonable both for vector species and all species richness (SI Appendix, Figs. S1, S3, S4). Supplementary Information Supplementary Information 1. Supplementary Information 2. Acknowledgements This article takes into account many results from the Medical and Veterinary Entomology Unit since its creation. There were several sources of funding, including the PREEMPT project HR001118S0017, financed by the Defense Advanced Research Projects Agency, the FSPI (Fonds de Solidarité pour les Projets Innovants) 2019-17 project, financed by the Ministry of Foreign Affairs, France, and ZIKAlliance no. 734548 project, financed by European Union Horizon 2020. We thank the Doctoral School ED MathSTIC Bretagne Océane for their financial support, which contributed to funding the stay of Heloise Rozier in Cambodia. No funding for Pierre Gloaguen, François Septier and Sébastien Boyer. Author contributions H.R. and P.G. performed the data analysis. H.R., P.G., and F.S. contributed to the model exploration and selection. S.B. provided entomological and field expertise. All authors reviewed and approved the final manuscript. Funding This research did not receive funding. Data availability Datasets and code are available in supplementary files. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-025-27465-0. References Associated Data Supplementary Materials Supplementary Information 1. Supplementary Information 2. Data Availability Statement Datasets and code are available in supplementary files.