Citizen science versus professional data collection: Comparison of approaches to mosquito monitoring in Germany

Mosquito monitoring data from active surveillance with BG-Sentinel traps from 2011 to 2017 and from passive surveillance - submissions to the "Muckenatlas" - between the years 2012 to 2017. The two different data sets are in one table, but separable by the table column "method". Locations of traps and submissions are not available due to data privacy regulations. The columns "bio.o" and "biocat" are only available for the passive surveillance entries as they depend on the participants notations about find spot that are not given in case of active surveillance by professional scientists.

In this context, 'passive monitoring' and 'passive surveillance' of mosquitoes refer to approaches where scientists do not actively gather the data themselves, but instead reach out to the public by setting up schemes and programmes, so that citizens can contribute data. One such programme is the citizen science project 'Mückenatlas' which includes citizens in the collection of mosquitoes to complement the active monitoring by scientists . It was launched in 2012, at a time when citizen science -the 'active public involvement in scientific research' (Irwin, 2018) -started to gain world-wide momentum.
Citizen science has since supported management decisions and actions of public authorities responding to global challenges and environmental threats (McKinley et al., 2017). Statistical methods have been developed to account for spatio-temporal biases and sampling errors in opportunistic data (Bird et al., 2014;Hochachka et al., 2012;Isaac et al., 2014;van Strien et al., 2013) and doubts on data quality have been rebutted (Danielsen et al., 2014;Lewandowski & Specht, 2015).
Public participation has proven especially useful in detecting and managing invasive species (Epps et al., 2014;Hester & Cacho, 2017;Roy et al., 2015), including arthropod vectors of disease agents (Hamer et al., 2018;Porter et al., 2019). In this context, many citizen science projects successfully focus on mosquitoes (e.g. Bazin & Williams, 2018;Jordan et al., 2017;Kampen et al., 2015;Mwangungulu et al., 2016;Palmer et al., 2017;Spence Beaulieu et al., 2019), so that mosquito-related programmes involving the public are being deployed in an increasing number of countries (e.g. Moore et al., 2019;Murindahabi et al., 2018). New projects benefit from the experience of activities already carried out and can accordingly develop tailor-made solutions or build on existing infrastructures [e.g. the adaptation of the 'Mosquito Alert' app in Hongkong (Cheung, 2017)

and the 'Mückenatlas' in New
Zealand (Museum of New Zealand, 2020) or Globe's Mosquito Mapper Tool (Muñoz et al., 2020)]. Considering the increasing number of projects, an international consortium called 'Global Mosquito Alert' seeks to keep the big picture and provides information and tools for all scales of mosquito surveillance (He & Tyson, 2017;Tyson et al., 2018).
In the case of the 'Mückenatlas', we recorded singular introductions of Aedes koreicus and the yellow fever mosquito Aedes ae gypti and were able to monitor the establishment and spread of the two major invasive species, the Asian bush mosquito Aedes japoni cus and A. albopictus across the country (Kampen & Werner, 2014;. In the event of an invasive vector species record through either active or passive monitoring from locations not considered colonized, the working groups' scientists immediately visit the place of capture to check for local reproduction, evaluate the situation and, depending on the outcome, inform public authorities to consider appropriate measures, for example, control strategies. Combining professional and citizen science data has recently been proposed for ecological research. For instance , Meentemeyer et al. (2015) predicted a future risk of the Sudden Oak Disease in California based on both data types, and Roy-Dufresne et al. (2019) showed that adding passively collected citizen science data to data generated by scientists improved distribution models of invasive rabbits in Australia. However, comparisons of the quantitative performance of each approach are rare (Goldstein et al., 2014;Palmer et al., 2017), as opposed to the qualitative performance of citizens compared to professional scientists following similar data collection protocols (Paul et al., 2014;van der Velde et al., 2017).
So far, the active and passive monitoring of the German mosquito fauna has been running hand-in-hand for more than 7 years, resulting in an extensive data collection that serve as a basis for valuable insights into the German mosquito fauna. The data have been mainly exploited regarding the detection and distribution of particularly rare species (e.g. Kampen, Schäfer, et al., 2016;Kuhlisch et al., 2019), spreading scenarios (e.g. Kerkow et al., 2019) or population genetics (e.g. Zielke et al., 2015). The data have also been used to inform authorities about the first detection and possible establishment of populations of invasive species to enable them to quickly initiate control measures. No difference has been made between the methodologies underlying the data collection, but mosquito data from both active and passive sources have been pooled. No evaluation of the two collection approaches has yet been carried out, and it has remained unclear which one contributes to which of the monitoring programme's objectives and to what extent.
Here, we quantitatively evaluate the passive and an active monitoring method within the German national mosquito surveillance programme with respect to (a) habitat coverage, (b) species recordings and (c) the ability to detect invasive species.
Specifically, we investigate the difference in the proportion and number of land use types in which the mosquitoes were caught to test for completeness of colonizable habitats. We also analyse the spectrum of species recorded by both methods to determine the respective diversity and to find possible causes for differences.
Lastly, we evaluate the capability to detect invasive species by assessing whether active or passive monitoring provided more first records of A. albopictus and A. japonicus of the affected German federal states.

| Passive mosquito monitoring by citizens
The 'Mückenatlas' developed 1 year after the official beginning of the nationwide mosquito monitoring programme, at that time rather uninfluenced by the globally emerging citizen science movement. Initially, it was not planned as a citizen science project. This idea evolved as people becoming aware of the trapping activities started to send in mosquitoes; they did so unprompted and out of curiosity. The leading scientists then decided to seize the moment and follow the idea, since then called 'Mückenatlas', by initiating a press release in April 2012, which unexpectedly received a strong response from both regional and national media. Due to the high number of submissions triggered by the news coverage, the 'Mückenatlas' workflow was gradually established as a large-scale citizen science project.
Participation in the project is very simple and requires no particular knowledge, training or protocol. People are asked to collect mosquitoes wherever and whenever they want to, with the only prerequisite that the insects remain physically intact, for example, are not smashed but caught alive, if possible, using a closeable container.
To kill a caught mosquito, it is recommended to put the sample into the freezer for 24 hr. In addition, the participants are asked to fill a submission form, which they can download from the website www. mueck enatl as.com, with information about the catch (most importantly, time and place). If internet is not available, submission forms can be sent in paper form to the participants. Hence, the project design also allows individuals to participate who are not comfortable with using digital tools, such as smartphone apps, or do not have web access at all. As a final step, the citizen scientists send the sample and the submission form at their own expense to the project's designated post office box. Only in a few cases they do not frank their packages, so that the postage costs must be paid from the project budget. For general requests about participation or other questions, a video explanation and FAQs on the website are offered; participants and other interested groups can also make contact online.
After identification, each participant receives a personal e-mail or letter from the project team, which is demonstrably one of the most important but also most time-consuming factors for the success of the 'Mückenatlas'. In this reply, the participant receives information about the species caught and also tips on how to eliminate and prevent mosquito nuisance. Even if the entry contained another taxon or the mosquito was in a condition that it could not be identified, a response is given. In addition, every participant is offered the possibility to have their name or a pseudonym marked on the website's 'collectors' map'.
Research results based on their data are regularly communicated via the website after publication. Potential participants are not specifically recruited but continuously addressed via the mass media (e.g. by issuing press releases) and, to a small extent, via social media and on the occasion of public events to draw attention to the project. Our good relations with media editors, which have developed over the years, as well as the fact that (invasive) mosquito species and the associated health risks are relevant and reportable topics help in this respect.
In general, participants submit one to five mosquitoes in a sample, most of which are in an identifiable state. In rare cases, participants operating own industrial mosquito traps send hundreds of specimens, mostly of the same species. To ensure data quality, species identification is only carried out by experienced experts of the working group.
Severely damaged specimens that cannot be identified to species level morphologically are determined genetically (Heym et al., 2018;Werner et al., 2020). By June 2020, over 25,500 citizens have participated and submitted a total of about 138,000 mosquitoes.

| Active mosquito monitoring by scientists
Active mosquito monitoring was done by trapping with BG-Sentinel traps (Biogents) equipped with gas bottles releasing CO 2 as attractant. This type of trap has proven to be more efficient and to attract a wider range of species than other trap types commonly used for collecting mosquitoes (e.g. Lühken et al., 2014). As opposed to the citizen science data, the data collected by the BG-Sentinels are standardized and allow analyses beyond the phenology and distribution of mosquitoes, such as assessing species abundances.
From 2011 to 2014, 68 traps were distributed all over Germany, placed deliberately in wetlands, urban surroundings, zoological gardens, cemeteries, airports and highway service stations. In the years 2015 to 2017, trapping followed a distribution regime: 64 traps were run annually only in the eastern half of Germany, while the western half was sampled by other groups. In that period, traps were randomly placed in a grid cell raster in near-natural, rural and urban settings, which were selected by a computer algorithm according to the percentage of these landscape structures occurring in Germany. All traps were run once per week for 24 hr from April to October, resulting in some 130,000 caught mosquito specimens.

| Datasets and statistical analyses
Trapped and submitted mosquitoes were identified morphologically under the microscope, using a determination key (Becker et al., 2010), or genetically in the case of severely damaged specimens or complex species (Heym et al., 2018;Werner et al., 2020). Information about the catches of both methods is entered into the German mosquito database CULBASE. For each species submitted to the 'Mückenatlas' or caught in one 24-hr trapping cycle, a single CULBASE entry is generated, hereafter indicated as 'observation' , regardless of the corresponding specimen count that is recorded as separate covariate. We exported datasets for active and passive monitoring for the years 2011 to 2017 and 2012 to 2017, respectively, and only used observations for comparison, disregarding the number of specimens per species and observation. All analyses were performed with the same set of covariates for both datasets (see Table S1). Mosquito groups or complexes were considered as a whole to account for impossibilities or uncertainties in differentiating females between species (see Table S2). For simplification, though, we refer to these complexes or groups as 'species' . The database automatically generates land use type based on CORINE Land Cover data level 3, which we manually re-classified to level 2 in order to improve presentation clarity. CORINE Land Cover data showed an accuracy of 82.8% for Germany in blind interpretation in 2012 (EU, 2012). Explorative and descriptive statistical analysis featuring frequency tables, (heat)maps, species accumulation curves, Fisher's exact test and Bray Curtis dissimilarity were conducted in r version 3.5.2 (R Core Team, 2018), deploying the packages summarytools (Comtois, 2019), rgdal (Bivand et al., 2019), vegan (Oksanen et al., 2019), viridis (Ganier, 2018) and ggplot2 (Wickham, 2016).

| Species recordings and composition
According to our species categorization, the 'Mückenatlas' recorded 36 mosquito species, while 38 species were trapped with BG-Sentinels.
Active monitoring needed far less municipalities than passive monitoring to collect all recorded species (Figure 2a). For the latter, it took more than 3,000 communities to reach the total number of species, although half of them (n = 18) were already detected after submissions from 57 municipalities (active: n = 19, needing 29 municipalities). The rates of the species' first records over time, as shown by the species discovery curves (Figure 2b), are comparable between the two approaches despite the earlier start and the slightly higher species richness of active monitoring. November 2012 for passive monitoring). Then it took both methods more than 5 years to collect the number of species reached by the end of 2017.
The active and passive monitoring datasets share 72.3% of the species collected (Bray-Curtis Index = 0.36), however, with a significant difference in species composition (p < 0.001, Fisher's exact test; Figure 3). The Culex pipiens complex was the most commonly recorded taxon for both methods, but it was far more often actively trapped (n = 5,847, 50.8%) than passively submitted (n = 8,382, 38.7%). Remarkably, the 'Mückenatlas' frequently registered Culiseta annulata, the only other species with a share over 10% (n = 3,790, 17.5%; see Table S2 for complete species lists and exact observation numbers). We recorded notably more Aedes geniculatus (6.2% vs. 0.6%) and A. japonicus (4.1% vs. 0.3%) in the 'Mückenatlas' than via active monitoring. Some species were found by one approach only, but both monitoring methods combined detected the currently assumed entire mosquito diversity of 52 species in Germany.

| Invasive species
Both methods detected A. japonicus, A. albopictus, Aedes petragnani and Culiseta longiareolata, but the 'Mückenatlas' additionally reported single specimens of A. koreicus and A. aegypti. First records of the most widespread A. albopictus and A. japonicus were made by the 'Mückenatlas' in more federal states (n = 10) than by active monitoring (n = 4; Figure 4).

| D ISCUSS I ON
The 'Mückenatlas' is a rather unique approach among the long- of this citizen science programme has not yet been accurately quantified in comparison to active monitoring, as done in other projects (Braz Sousa et al., 2020;Goldstein et al., 2014;Palmer et al., 2017), most of the costs of the programme are incurred by the project's staff salaries, very little by recruiting participants through public events and media relations, and almost none by the citizens' data collection.
With submissions from over 11,000 unique geo-locations and more than 3,000 municipalities, the 'Mückenatlas' achieves a broad spatial coverage, demonstrating a major benefit of citizen science (Dickinson et al., 2010;Irwin, 2018). The lower quantity of municipalities covered by active monitoring better reflects the typical proportion of land use types in Germany, and therefore, we consider the resulting active monitoring dataset representative and able to detect the entire mosquito diversity in Germany, even if most traps included in this study were placed in geographically eastern Germany. By contrast, the 'Mückenatlas' submissions indeed originated from more land use types, but display an overrepresentation of urban areas, as most of the catch locations were based in or around people's homes, in houses, apartments and gardens. An advantage is, however, that we receive data from private properties in this way that scientists normally cannot access (Dickinson et al., 2010;Epps et al., 2014). Such data are urgently needed to assess the impact of urban development on ecosystem functioning and biodiversity (Dunn & Beasley, 2016;Spear et al., 2017).
The higher species richness yielded by active monitoring is presumably caused by the selective placement of traps (e.g. in swamps, on floodplains or peatlands) in the 2011-2014 period.
The lower species richness of the passive monitoring might be due to the overrepresentation of urban areas. Considering the time needed to collect the respective number of species, both monitoring methods are comparable as shown by the species discovery curves. For both methods, the most recent 10 first records of species constituted of invasive or very rare taxa, suggesting that chances to detect either of them appear to be equally low for active and passive monitoring.
To simplify the comparison of species recordings, we have assigned the mosquito species to the corresponding groups and complexes. However, when analysing the data in an entomological-medical context, it is essential to consider differences in the ecological traits of the individual taxa, such as within the Culex pipiens complex. Both monitoring methods differed significantly in species composition, and surprisingly also in the most frequently recorded species. Although C. annulata and A. geniculatus are geographically widespread species, they were considerably less frequently collected by active monitoring than by passive monitoring.
Reasons for the high submission numbers to the 'Mückenatlas' compared to active trapping are probably the morphological appearances of both species. The ringed legs of C. annulata and the black-and-white habitus of A. geniculatus match the characteristics described by the media when featuring the invasive A. albopictus or A. japonicus. In addition, C. annulata and A. geniculatus are fairly large-sized mosquitoes, and communication with submitters to the 'Mückenatlas' has shown that invasive species are generally thought to be extraordinarily big, not least owing to the name affix 'tiger'. This substantiates our suspicion that participants actively look out for, or only become active when they think to have recognized invasive species, creating a recording bias known from other studies (Roy et al., 2015;. The same effect probably causes the higher number of registrations of the actually invasive A. japonicus in the 'Mückenatlas', although experience shows that this species is not readily collected by the BG-Sentinel trap (pers obs.).
In the case of C. annulata, the difference in seasonality of both monitoring methods affects the number of observations as well. Culiseta annulata often overwinters in basements of, or fire wood stacks near, houses and is continuously submitted to the 'Mückenatlas' during the winter months and early spring, whereas the BG-Sentinel traps were solely operated from April through October, missing the chance to catch overwintering specimens.
Among the six species not shared between the two approaches, two invasive mosquitoes were only detected by the 'Mückenatlas',  (Kampen, Jansen, et al., 2016). Species only found by active monitoring (Anopheles algeriensis, Aedes diantaeus, Culex martinii and Uranotaenia unguiculata) are either rare, bound to specific habitats outside urban areas, exophilic or do not feed on humans (Becker et al., 2010). The rediscovery of A. algeriensis and C. martinii (Kuhlisch et al., 2018b;Tippelt et al., 2018)  This citizen science project has thus become an invaluable tool for surveying invasive mosquitoes, corroborating recent findings of the usefulness of passive surveillance for dealing with biological invasions (Hester & Cacho, 2017;Sladonja & Poljuha, 2018). As a practical example of management implications and the interplay of both monitoring methods, the city of Erding in Bavaria initiated eradication measurements in a cemetery after sampling provided evidence of local reproduction. In another case, the 'Mückenatlas' submission of the first A. albopictus from Thuringia (Jena) led to 3 years of active monitoring tracking established populations in different cemeteries (Kuhlisch et al., 2018a).

| CON CLUS IONS
In this study, we compared active (via BG-Sentinel traps) with passive (via a citizen science project) mosquito monitoring efforts over a time period of 7 years. Our analyses revealed that passive monitoring is an efficient way to collect species data in direct proximity to humans and their surrounding environments, reducing volunteer management and equipment costs, and empowering citizens to provide important information that benefits both society and science. Passive monitoring performed better in detecting invasive species, because citizen scientists predominantly sampled in urban areas where most invaders arrive with introduction vehicles, but also due to increased alertness towards the perils of A. japonicus and A. albopictus resulting from massive German media coverage. This sampling bias of citizen scientists is mitigated by active monitoring, which performs notably better in capturing the entire mosquito diversity through selective placement of traps. In addition, trapping appears to be especially useful to validate first detections as well as estimate infestations with subsequent, methodically conducted surveillance.
With these project-specific advantages, the 'Mückenatlas' proved to be a valuable tool to obtain an increasingly accurate picture of the occurrence and distribution of mosquitoes over a long period of time, including the spread and detection of invasive species. Its project design could serve as an example for other citizen science programmes to complement or substitute active approaches aiming at (a) large-scale, long-term surveillance, (b) detecting invasive or rare species and (c) a comprehensive recording of (mosquito) biodiversity in urban settings. As opposed to that, we think that the 'Mückenatlas' approach is less suitable for studying specific species over a short time period, for spatially limited regions or selected habitat types (except for indoor diversity) and for investigations bound to certain times, that is, when randomness and loss of control is not acceptable. In these cases, apps like 'Mosquito Alert' , traps run by citizens (Johnson et al., 2018) or a strict protocol followed by a designated stakeholder group (Tarter et al., 2019) might be more appropriate.
While the citizen science programme has been running successfully in Germany since 2012, its design might face difficulties in other countries due to cultural, economic and social differences. People might not be willing or able to cover postal costs, especially in socioeconomically weak countries, which are particularly threatened by mosquito-borne diseases. Moreover, attitudes towards science might not be positive enough, the health concern or the interest in the living environment not strong enough to justify sufficient time investment. Therefore, we recommend prior proof-of-concept studies to test a project's design, workflow and acceptance, as carried out by Braz Sousa et al. (2020), also to create a solid basis for grant applications.
Momentum is there to encourage local and national authorities to trust the solid evidence that formal surveillance programmes could benefit from a citizen science component. Especially to achieve the goals of Integrated Vector Management as defined by the WHO -such as cost-efficiency, sustainability, precise knowledge on distribution and empowerment of communities -the involvement of citizen science can play an increasingly important role in the future (Fernandes et al., 2018;Fouet & Kamdem, 2019). However, it must be clear that citizen science cannot be the one-fits-all solution, but only one tool in the toolbox of mosquito surveillance.

ACK N OWLED G EM ENTS
First and foremost, we thank all 'Mückenatlas' project participants for their submissions and the volunteers taking care of the mosquito traps. We also thank the laboratory staff and the PhD students of

DATA AVA I L A B I L I T Y S TAT E M E N T
An anonymised version of the data without geo-coordinates of the trap and catch locations is available via the Open Research Data repository at the Leibniz-Centre for Agricultural Landscape Research (ZALF), Germany, https://www.doi.org/10.4228/ZALF.DK.151 (Pernat et al., 2020). Sharing the raw data including the exact geo-coordinates would violate the personal privacy of the citizen scientists.