A total of 1,531 moth samples were collected in summers of 2017 (Chang, 2018; Jiang, 2018), 2018 (Jia, 2019), and 2020 from the Dongling Mountain (∼39°51′07″N, 115°29′13″ E) in Beijing, Zhoushan Island (∼30°5′22″N, 122°3′15″E) in Zhejiang Province, and Weihui city (∼35°39′18″ N, 113°59′2″E) in Henan Province, China, respectively(Figure 1A). In the following, samples collected from Beijing, Zhejiang, and Henan are represented by BJDLS, ZSDXG, and HNYHG, respectively. All adult moths were captured at night by setting light traps between 8 and 12 p.m. (Yang et al., 2012).

SUN-FL tethered flight mills (Beijing Taichi Pengcheng Electronics Company, China) were used to perform hanging flight experiments. This device records: total distance flown (TDF), total flight duration (T), the average speed during flight (VTDF,VTDF=TDFT), longest single flight (LSF), the time corresponding to the longest single flight (T_LSF), and average speed corresponding to the longest single flight (VLSF,VLSF=LSFTLSF) (Ávalos et al., 2014). The captured moths were subjected to a hanging flight experiment in a closed dark room. In the 2017 and 2018 experiments, each moth was included in a continuous 24-h hanging flight experiment without energy supplementation, while in the 2020 experiment, each experiment lasted under 10 h (moths generally became inactive before the end of the 10-h experimental period in all cases). We chose four parameters; V_TDF, LSF, T_LSF, and V_LSF, which may be less affected by hanging flight times (though in the subsequent analysis since we found most moths become inactive in less than 10-h hanging flight, thus experiment durations over this do not impact results).

All specimens collected from BJDLS (2017) and ZSDXG (2018) were transported at 0°C to the laboratory in Capital Normal University for DNA extraction and PCR amplification for subsequent DNA barcode identification (Hao et al., 2020). Experimental samples from HNYHG were identified based on morphology by Associate Professor Jianxin Cui of the School of Resources and Environment, Henan Institute of Science and Technology.

For all samples collected from BJDLS and ZSDXG, total genomic DNA was extracted individually from 2 to 3 legs, using QIAamp DNA Mini kit (Qiagen, Venlo, the Netherlands) following the protocol of manufacturer. PCR amplification was performed using standard DNA barcoding primers LCO1490 (GGTCA ACAAA TCATAA AGATA TTGG) and HCO2198 (TAAAC TTCAG GGTGA CCAAA AAATCA) (Folmer et al., 1994). The PCR products were bidirectionally sequenced at Beijing Zhongkexilin Biotechnology Co., Ltd., using an ABI 3730XL analyzer. All sequences of samples ∼658 bp were submitted to the BOLD system¹ for taxonomic identification (Ratnasingham and Hebert, 2007).

We removed the total flight distance parameter below the lower quartile. One of the reasons for these data could be that the wings of some moths are glued in the process of experiment, reducing their willingness or ability to fly. If these data were included, we would underestimate the fly ability of moths (Jantzen and Eisner, 2008; Combes et al., 2010). All calculations and statistical tests involved in this article are implemented by R (v4.0.4) (R Core Team, 2013). For the flight parameters of species level, family level, and superfamily level, the shapiro.test function in R was used to check for the normal distribution of the data, and the Wilcox.test function in the stats package to compare the differences among species, families, and superfamilies. The lm and predict functions in R were used for fitting and prediction of unary linear regression, respectively.

After obtaining the flight data of 36 (n ≥ 10) moth species from 10 different families, we simulated the migration of different species between mainland and island (see below) in R. In the simulation, we approximate Zhoushan Island and its adjacent mainland (Ningbo) into two circular regions, and the minimum distance between the two circles (e.g., the shortest distance between Ningbo and Zhoushan Island) is about 8.3 km measured based on Map World on National Platform for Common Geospatial Information Services².

We assume that the ability of moths to cross the strait to complete the spread between the mainland (island) and the island depends on the LSF. Regardless of airflow, human factors, and more stochastic natural factors, species can complete migration only when the LSF is greater than the shortest distance between the islands (but the stepping stones that exist between the islands reduce the difficulty of crossing the islands.

According to the LSF of the wild populations of the moth species that we measured, we simulated the spread of each species from the center of the area through computer simulations and observed the spread of offspring individuals across islands within 500 generations (two generations per year (Hao et al., 2020), it will take about 250 years). In our data, there are a total of 36 species with at least 10 individuals (Supplementary Table 1), of which the LSF data of 34 species (94.4%) had a log-normal distribution, thus the LSF data of all moths are assumed to correspond with log-normal distribution in the simulation. We simplified Ningbo (with an area of about 9,816 km²) and Zhoushan Island (with an area of about 502 km²) into two circular areas with radius of 56 and 12 km, and the closest distance (CD) between the two circles is 8.3 km. Here we defined a new concept P_(LSF > CD) to explain the dispersal ability of species, and its specific definition is that the proportion of the number of individual that LSF is longer than the closest distance (CD) between mainland and island to the population size.

A total of 1,142 (∼74.4%) experimental flight data samples were analyzed, i.e., including 222 samples from BJDLS (the minimum TDF is 18.84 m), 409 samples from ZSDXG (the minimum TDF is 67.23 m), and 511 samples from HNYHG (the minimum TDF is 33.91 m). Filtered data were collected from seven Lepidoptera superfamilies (Noctuoidea, Bombycoidea, Zygaenoidea, Geometroidea, Drepanoidea, Lasiocampoidea, and Pyraloidea), 12 families (Erebidae, Noctuidae, Notodontidae, Saturniidae, Sphingidae, Zygaenidae, Limacodidae, Geometridae, Drepanidae, Lasiocampidae, Crambidae, and Pyralidae), and 186 species, and among them, 81 species with at least 3 individuals and 36 species with at least 10 individuals. A total of 631 individuals from BJDLS and ZSDXG were identified by comparison with the BOLD system database. Among them, 547 individuals had best hits with similarity scores of at least 98 (part 1), and the remaining individuals had similarity scores between 90.43 and 97.99% (part 2). The remaining 511 individuals from HNYHG were identified using morphology (part 3). Part 1 contains 134 species, part 2 contains 40 species, and part 3 contains 39 species. There are 13 common species in parts 1 and 3, but only one common species, Catocala kaki (Noctuidae) in parts 2 and 3.

We compared the difference in flight abilities (V_TDF, LSF, T_LSF,and V_LSF) at different taxon levels in all filtered data. First, we checked whether the flight data at different taxon levels conformed to a normal distribution and found that only a small proportion of the data conformed (at the superfamily level, only 0–42.9% data of the above four parameters conform to a normal distribution; at the family level, only 8.3–41.7% data of the above four parameters conform to a normal distribution; only 12.5–55.6% of the flight capability data at family level captured from different sampling sites show a normal distribution. Therefore, the Wilcoxon rank-sum test of the one-tailed hypothesis is selected to compare the sample differences between different classification levels and different sampling sites.

The lowest V_TDF, V_LSF, LSF, and T_LSF were found in moths from superfamily Geometroidea, which were lower than those in the other five superfamilies except for no significant difference with superfamily Lasiocampoidea (Figure 1C). Superfamily Drepanoidea is significantly higher than the other five superfamilies (p < 0.05) on the flight parameter V_TDF except that it does not show any significant difference from superfamily Zygaenoidea. In V_LSF. Bombycoidea is significantly higher than the other five superfamilies (p < 0.05) with the exception of Drepanoidea. While superfamily Pyraloidea has the longest LSF and V_LSF among seven superfamilies in our study (p < 0.05).

Correlation analysis between flight parameters and moth morphological data showed that V_TDF and V_LSF had a significant positive correlation with the increase in moth body size [body length (BL), total wing area (TWA), forewing area (FA), and hind-wing area (HA)], while T_LSF had a weak negative correlation with moth TWA and HA (Supplementary Figure 2), highlighting the importance of individual wing-traits in determining flight performance and ability.

Previous studies on the flight ability of insects using a flight mill system mainly focused on agricultural pests (Hashiyama et al., 2013; Sappington and Burks, 2014; Hoddle and Hoddle, 2016; Fu et al., 2017a,b; Lopez et al., 2017; Babu et al., 2020). This study is the first to use a unified method to study the flight ability in Lepidoptera families and superfamilies to explore traits related to possible invasion and colonization ability. In this experiment, 1,142 specimens of seven Lepidoptera superfamilies were collected. In total 186 species had their flight ability measured in the experiment. Superfamily Geometroidea showed no significant difference with superfamily Lasiocampoidea but was significantly weaker than the other five superfamilies (Figures 1Ca–d). A total of 197 samples of superfamily Geometroidea were used for data analysis, accounting for 17.3% of the total samples, and only family Geometridae species are included in this superfamily. Geometrid moths are frequently flattened when in a static state, and the color of body and wing is often very similar to the surrounding environment. One of the best known Geometrid moths is the peppered moth (Biston betularia). The initial phenotype of this moth was speckled black and white, but industrialization in Britain and other regions in the nineteenth century increased the proportion of melanic individuals (Cook et al., 2012; Cook and Saccheri, 2013). This change in the main color-form as a consequence of selective predation is still considered as an example of the rapid evolution as a consequence of natural selection in a changing environment, which is driven by selective predation (Cook et al., 2012; Cook and Saccheri, 2013). Individuals of American peppered moth (Biston betularia) larvae are selected to match background color depending on their different host plants, and the slender shape makes them look more like branches, thereby using a similar approach to reduce predation (Noor et al., 2008; Eacock et al., 2017). Thus, the impact of selection on key traits is clear, but at least in this group, it does not require an ability to fly to escape predators, and conversely, the general body form of Geometrids may be expected to have a weaker flight performance.

Superfamily Drepanoidea had the highest in V_TDF parameters (Figure 1Ca), which was faster than the other five superfamilies (with the exception of Lasiocampoidea). But at present, we know very little about this superfamily, with only a few omics studies about the social caterpillar Drepana arcuata from this superfamily (Yack et al., 2001; Yadav et al., 2020a,b), which provide no insights into flight performance. Superfamily Pyraloidea (family Pyralidae) showed the strongest LSF and T_LSF in comparison to other superfamilies (Figures 1Cc,d), suggesting that it has high flight endurance. Superfamily Pyraloidea has the smallest body size in all samples collected in this study, such as the smallest BL, weight, and wing area. At present, there are too few comparative studies on the high taxonomic rank flight ability of Lepidoptera, but species-specific morphology is likely to be responsible for the differences found here.

Among 187 individuals belonging to superfamily Bombycoidea in this study, 171 individuals (∼91.4%) are from family Sphingidae and the remaining individuals are from family Saturniidae. The results show that the of Bombycoidea are significantly higher than those of the other five superfamilies (Figure 1Cb), possibly because superfamily Bombycoidea includes the family Sphingidae with the highest (Supplementary Figure 3). Our results prove that moths can have extensive niche differentiation, and two families with distant phylogenetic relationships (Figure 1B) may have similar flight abilities. Traits to escape predation are important, as Lepidoptera can account for more than 50% of the food of some insectivorous bats (Galan et al., 2018), and within this, Noctuidae can account for 50% of Lepidoptera species consumed (Zeale et al., 2011). Hawk moths (Lepidoptera, Bombycoidea, Sphingidae), Notodontidae, and Noctuidae (Lepidoptera and Noctuoidea) are more active and fly faster than most other moths. Large species with fat bodies may be particularly vulnerable to bats, such as many species in Sphingidae and Noctuidae, and need to be able to either fly fast or have other adaptations to reduce the risk of predation (Ntelezos et al., 2017; Rubin et al., 2018; Aiello et al., 2021). Hawk moths have the most aerodynamic body form of most moths, based on both wings and body morphology (Willmott and Ellington, 1997), which enables very high-speed short-distance flight.

Affectd by the environment of the habitats, different geographical populations may have different flight abilities, although LSF and T_LSF have no statistically significant differences, the mean value of V_TDF and V_LSF in island are much higher than mainland [ratio span (island/mainland): LSF (1.90–4.03), T_LSF (1.03–3.20), V_TDF (0.71–1.45), V_LSF (0.85−1.66)] (Figures 2A-D). We found that the same group (at family level) between mainland and island showed difference in flight ability. We speculate that the difference may be due to different species composition. Because most of the flight data of the same family at different sites are from different species. To test this hypothesis, we compared the data of common species [B. panterinaria (Geometridae), T. juno (Erebidae), and U. meticulodina (Notodontidae)] from the island (ZSDXG) and the mainland (HNYHG) (Figures 3A–D). U. metaculodina showed that the mean value of island individuals is lower than that of mainland individuals (ratio of mean value are 0.62 and 0.76 in LSF and T_LSF, respectively). But the average values of LSF and T_LSF in the samples of B. panterinaria and T. juno collected from islands are larger than those from mainland samples, although the difference was not statistically significant (ratios of mean value are 23.30 and 19.84 between parameters LSF and T_LSF in species B. panterinaria and ratios are 2.00 and 2.30 between LSF and T_LSF in T. juno). Based on the above results, we infer that islands may recruit more species with strong flight capacity (in parameter LSF and T_LSF) than the mainland, which may expand differences in average of LSF and T_LSF between mainland and island; moreover, prevalent wind current speed and direction are likely to influence this. The species found on both islands and continents show that island individuals have a weaker flight speed (V_TDF&andV_LSF) than continental individuals, which indicates that flight distance (LSF) may be more important for species ability to spread than flight speed.

Zhoushan Island is only separated from the mainland about 9,000 years ago due to the rising sea level caused by climate change (Wang, 1980). Zhoushan Island is the largest island in the Zhoushan Archipelago and is close (the shortest distance is only about 8.3 km) to the mainland. In such a short period of time and extremely similar natural environment, the same species may not have time to differentiate their flight abilities or adapt to local conditions, such as differing wind speeds. However, although 9,000 years is not enough for species differentiation, the level of gene flow in addition to environmental pressures depends on wing-traits that relate to flight ability. It is generally difficult to prove this through actual observation data for these small insects, but we have measured the flight parameters of a considerable number of species, so we can verify the possibility through the simulation.

Our results show that L. muscosalis, the species with the largest LSF, has the highest number of cross-island generations and the highest number of cross-island individuals per generation in 500 generations. With the decrease of species dispersal ability (P_(LSF>CD)), the number of cross-island generations and the number of cross-island individuals in each generation decreased rapidly. However, it is interesting to note that the decline rate of the cross-island generations is slower than that of the number of cross-island individuals per generation. When the average number of cross-island individuals decreased from 166.6 to 6.6 (corresponding to the first four species with P_(LSF>CD) value declined from 0.151 to 0.009 in Figure 4), the number of cross-island generations remained at a high level (495–487). When P_(LSF>CD) decreased from 0.005 to 0.0012 (From T. juno to A. major in Figure 4), the number of cross-island individuals was very low (from 1 to 2), and the number of cross-island generations decreased rapidly from 375 to 24. For the remaining species, there was almost no cross-island migration within 500 generations, and the P_(LSF>CD) dropped to less than 0.0006.

Regression analysis also shows that when P_(LSF>CD) drops below 0.004 (obtained from the simulation of spreading from the mainland to the island) and 0.003 (obtained from the simulation of spreading from the island to the mainland), the species lose the ability to migrate over the sea (the predicted value of the number of individuals that successfully spread is less than 0) (Figure 6 and Supplementary Figure 6). Regardless of whether the species is spreading from the mainland to the island or from the island to the mainland, the requirements for P_(LSF>CD) are relatively consistent. As long as a species can complete the spread from the mainland to the island (or from the island to the mainland), it may also be able to complete the reverse proliferation activity with the help of wind conditions. Our results showed that the flight ability of the individuals that successfully cross to or from the island is significantly stronger than that of the individuals that remain on the mainland (Figure 5). Similar results are also found in the simulation of species spreading from a small circle (island) to large circle (mainland) (Supplementary Figures 4, 5). We did not simulate whether individuals with strong flight ability would produce a larger proportion of offspring with strong flight ability (though it is likely to be heritable), but a study on Episyrphus balteatus (Diptera and Syrphidae) with migration behavior showed that the offspring of migrants in this species had higher flight activity than those of resident individuals (Dällenbach et al., 2018). The reality is far more complicated than the simulation. At present, we do not know how many of the species we collected from the island existed before Zhoushan Island separated from the mainland. Although the individuals successfully landing on the island from the mainland are all with stronger flying capability, the individuals from the islands that mate with them may be not strong enough, so they may could not produce offspring with strong flying capabilities. Molecular approaches would be needed to determine gene flow between mainland and island to determine how these populations interact and how many still show genetic connectivity to mainland populations. Furthermore, mainland individuals generally come from more diverse communities, so may be more competitive than species adapted to more depauperate island communities.

The known factors that may affect the flight ability of Lepidoptera and other insects include sex and days post-eclosion. The effect of sex on flight ability varies between species. In Helicoverpa armigera (Lepidopera and Noctuidae), males have better flight abilities than females (Jyothi et al., 2021), while in Amyelois transitella (Lepidopera and Pyralidae) and Rhynchophorus ferrugineus (Coleoptera and Dryophthoridae), there is no sexual difference (Ávalos et al., 2014; Sappington and Burks, 2014). The flight capacity of newly emerged adults is weak in many groups, reaching the peak at 3 or 4 days post-eclosion (Tanaka and Yamanaka, 2009; Liu et al., 2011; Fu et al., 2017b).

In addition, how Lepidopterans determine the direction of flight is a longstanding question for researchers. Species, such as monarch butterflies (Reppert et al., 2016) and Bogong moths (Dreyer et al., 2018) are able to navigate using geomagnetic fields, time compensated sun compass, and identifying landmarks to determine migration routes. But the processes guiding short-distance dispersal are less well understood. Lepidoptera individuals need to determine the destination if they travel between islands and other landmasses, and many sensory cues are not available over water, and loss of energy and poor weather are more likely to cause death, though studies show that some species can detect polarized light, and this may impact navigation over extended distances. Recently, MacDonald et al. (2019) partly proved that butterflies rely on vision to detect patches of suitable habitats. Their research shows that individuals with normal vision are 30.1 times more likely to navigate to the target island after being released than those with induced flash blindness.

Although we collected a large number of Lepidopteran samples from the field, it is almost impossible to identify the physiological state of each sample, and its performance on the flight mill depends greatly on the physiological state. This may be the main reason for the high variability for some measures within species in our study. According to the activity of insects during sampling, the low TDF may be caused by the discomfort of the tested insects caused by the flight mill or the weakening of the flight ability caused by human operation error, but these factors should not be responsible for consistent differences observed between samples from different species and localities. However, whether it is reasonable to filter data in this way remains to be verified. Whilst we changed the experimental duration in 2020, the four flight parameters used in our analysis are unlikely to have been impacted. In addition, the dispersal direction of individuals in our simulation is completely random, and MacDonald et al. (2019) also pointed out that this random flight null assumption tends to overestimate the proportion of successful dispersal of butterflies toward adjacent habitat patches (MacDonald et al., 2019).