Introduction
Contributory citizen science platforms are open for people who want to be engaged in scientific activities. In the case of iNaturalist, every registered person can upload opportunistically collected observation records of biodiversity, which later can be used by researchers in various studies, for example, in the fields of biogeography (Gaier and Resasco 2022), natural history (Unger et al. 2023), and evolution (Jansen Prujin and Mayer 2025). A major motivation behind the establishment of such platforms is to contribute to biodiversity or nature conservation (Aguirre et al. 2024). Such platforms can efficiently increase the coverage and decrease the costs of biodiversity monitoring as they complement the activity of professionals and government agencies (Levrel et al. 2010; Theobald et al. 2015) and can provide country-wide and long-term reliable information for conservation decision-making (Aguirre et al. 2024, Schmeller et al. 2008). For example, the volunteer-based Common Bird Monitoring Scheme (Mindennapi Madaraink Monitoringja, MMM) coordinated by the non-governmental organization Birdlife Hungary is the longest running country-wide monitoring in Hungary, providing the national bird census data. Their results are also used by governmental entities, for example, to report for the European Union’s Natura 2000 directives (Szép et al. 2012).
However, such initiatives are often biased in various ways. Carlen et al. (2024) summarized these sampling biases and identified four, non-exclusive filters that alter our knowledge on the actual species pool of an area and can lead to a reported, incomplete species list. The participation filter reflects who is reporting the data. This necessarily includes where these people are located and to which areas they have access (Di Cecco et al. 2021). The socioeconomic background of the observers might also influence their willingness to participate in such activities, and this often leads to reporting bias towards more urbanised areas with higher income residents (Estien, Carlen, and Schell 2024). The sampling filter implies that people upload observations under certain circumstances, for instance, close to their home or where they perform recreational activities, such as in green spaces and protected areas. For example, Dimson and Gillespie (2023) found that the majority of people using iNaturalist in Hawaii are visitors spending their vacation on the islands—not residents. The detectability filter is based mostly on traits of the species such as their abundance, daily activity period, habitat preference, or external morphology, which determines the frequency of their encounters with observers (Troudet et al. 2017). Lastly, the preference filter reflects what species fascinate people enough to upload them. In general, visually more appealing creatures, such as large-bodied birds (Callaghan et al. 2021) or more vivid butterflies (Goldstein et al. 2024) are more likely to be reported on citizen science platforms.
In this study, we aimed to identify the sampling biases, especially spatial biases in the iNaturalist database of records from Hungary, to inform future volunteer-based mapping projects for the country. Based on previous knowledge examining spatial patterns of opportunistically collected data (Geurts, Reynolds, and Starzomski 2023; Tiago et al. 2017) and a visualisation of the uneven distribution of records in Hungary, we were interested what factors could explain these patterns. From a conservation point of view, such study is also relevant because as part of the European Union’s biodiversity strategy, Hungary must expand the area of the Natura 2000 network of protected areas to 30% of the country’s territory by 2030 (European Union 2020), which requires up-to-date information on biodiversity and biodiversity data collection.
Similar to global trends, engagement in citizen science activities are popular in Hungary as well. In a recent analysis, Vásárhelyi et al. 2026 examined different socioeconomic factors and the effect of protected areas on the spatial distribution of records gathered in 17 different citizen science projects related to biodiversity from Hungary, and revealed interesting insights about the society and the data provided. In another study related to Hungary, Aguirre et al. 2024 examined the applicability of citizen science data from various Hungarian databases in scientific research and nature conservation.
Focusing on taxonomic, temporal, and spatial biases of citizen science observations, we further deepened our understanding about public engagement in data collection. We analysed the records uploaded to iNaturalist, a database that is both widely used and freely accessible for scientific research. In particular, we explored the socioeconomic and land use variables explaining spatial bias. This study can serve as a baseline for the design of future targeted biodiversity mapping projects.
Methods
Citizen science observations
We downloaded every verifiable observation record directly from iNaturalist via the export tool on July 05, 2024 from within the borders of Hungary (Supplemental File 1). Verifiable observations always contain photos or audio files. These observations belong to two groups: first, the “Needs ID” group, where at least one observer, usually the uploader, added an identification, but there is no further verification; or if there are verifications, there is no consensus about the identity. Second is the “Research Grade” group, where more than two-third of the identifiers agreed on the observation identity. Ecological studies usually use Research Grade records because of their generally more accurate species identification (White et al. 2023; Mesaglio et al. 2025), however, as we were interested in observer activity rather than the ecology of species, we downloaded all verifiable records, not just the Research Grade observations.
Data preparation
We calculated the number of observations, the number of Research Grade observations, and the percentage of Research Grade observations within the so-called iconic taxon groups, in iNaturalist terminology. These are larger, recognisable groups used by iNaturalist regardless of their taxonomic level, for example “Amphibians”, “Reptiles”, “Molusca”, or “Plants”. We took the estimated number of species in Hungary within the iconic taxa from published assessments where this was available (animals: Páll-Gergely et al. 2024; plants: Király et al. 2009; fungi: Rimóczi 1997), and compared it with the number of species in the iNaturalist dataset using only the Research Grade observations to ensure accuracy in species identification. As the species lists from Páll-Gergely et al. 2024 and Rimóczi 1997 are not publicly available, and the list from Király et al. 2009 is available only in printed book format, it was not possible for us to cross-reference every species from these lists with those on the iNaturalist species list, so there could be species uploaded to iNaturalist not included in the cited references because they are newly discovered in Hungary.
For the temporal analysis, we used the date of observation and not the date of upload. We assessed the number of observations in months, days, and hours (Central European Time). We excluded 2024 from these analyses, which was not yet complete, to compare only years for which records were complete. For the engagement analysis, we calculated the number of records uploaded to iNaturalist per individual user.
For the spatial analyses, we excluded records with downgraded accuracy (“obscured coordinates” in iNaturalist terminology) and records with coordinate accuracy higher than 10,000 m. This means that we involuntarily excluded threatened taxa (species or subspecies) as iNaturalist auto-obscures the records of them (208 taxa, 2.42% of the uploaded taxa). We aggregated the observation records into a 10×10 km resolution grid and into polygons of the protected areas (hereafter both will be referred to as spatial units). We obtained the spatial grid from the European Environmental Agency (https://www.eea.europa.eu/data-and-maps/data/eea-reference-grids-2/gis-files/hungary-shapefile), which is frequently used in biodiversity studies in Europe. Spatial data on protected areas were acquired from the Ministry of Agriculture, Hungary upon request; however identical versions are available at the World Database of Protected Areas (https://www.protectedplanet.net/country/HUN). Protected areas encompass both the nationally declared protected areas and the Natura 2000 network that consists of Special Protection Areas (SPAs) and Special Areas of Conservation (SCIs). As almost every nationally protected area in Hungary is covered by SPAs, SCIs, or even both, we merged overlapping areas by dissolving their borders.
We studied the effect of environmental factors that may explain observer activity on two dependent variables: the number of observations, and the Shannon diversity index of iconic taxa. The first dependent variable simply describes how many records observers uploaded from a cell, while the second shows the potential taxonomic bias in space. For explanatory variables, we calculated the extent (area) and Shannon diversity index of land use types (SDIL), as well as the mean altitude; the proportion of protected areas; the density of roads; the distance from the Hungarian capital, Budapest; the distance from the nearest city with county rights; and the mean gross income of people. Data on land use types were obtained from the Ecosystem Map of Hungary (EMH, Ministry of Agriculture, Hungary 2019; Tanács et al. 2019). The EMH is a raster map with a 20×20 m resolution and three hierarchical levels of classification of land use types containing 6, 22, and 56 categories, respectively. We used the land use types in Level 1, which included urban areas, croplands, grasslands, forests, wetlands, and rivers and lakes. The SDIL was calculated based on Level 3 classification, which enabled us to calculate a more precise estimation of Shannon diversity. The mean altitude of spatial units was calculated based on the Shuttle Radar Topography Mission (SRTM) (Farr and Kobrick 2000). The proportion of protected areas was calculated based on the dissolved protected area polygons within the grid cells. Road density was calculated as the length of roads within grid cells from the GRIP global roads database (https://globio.info, Meijer et al., 2018) with the inclusion of Levels 1–4 (Highways, Primary roads, Secondary roads, Tertiary roads) but with the exclusion of Level 5 (Local roads) to avoid spatial autocorrelation with the urban areas land use type. The distance from Budapest was calculated as the distance of the centre of each spatial unit from the centre of Budapest, while the distance from a city with county rights was calculated as the distance of the centre of each spatial unit from the centre of the closest of the 26 cities. The latest estimation (2022) of the mean gross income of people was downloaded from the Hungarian Central Statistical Office (HCSO) (2024) in district level and were averaged to grid cell units. In the analysis restricted to protected areas only, we excluded road density and mean gross income but included the combined area of protected areas. The former socioeconomic variables were excluded because protected areas are usually outside urban settlement where income is realised, and most parts of them are roadless. The combined area of protected areas was log-transformed for analyses.
Statistical analyses
We compared the number of observations to an equal distribution using χ2 tests for months, days of the week, and hours of the day, respectively. We assessed the effects of spatial drivers on the number of observations and Shannon diversity index of iconic taxa as dependent variables both in the grid cells and in the protected areas by constructing generalized linear models (GLM) using the ‘lme4’ package of R (Bates et al. 2015). We then applied a model selection approach in an information-theoretic framework (Burnham and Anderson 2002) to identify models with substantial empirical support based on Akaike differences (Δi = AICi – AICmin < 2.0) and to find the best set of explanatory variables. Following model selection, we performed model-averaging of the ΔAIC < 2.0 models using the ‘MuMIn’ package of R (Bartoń 2019) to estimate parameter coefficients. The above listed analyses were conducted in R (R Core Team 2021).
Results
We downloaded a total of 316,751 verifiable observation records, of which 195,205 (61.63%) were Research Grade observations. The majority of records from both types belonged to Insecta and Plantae iconic taxa, while Actinopterygii and Protozoa were the least numerous (Table 1). The proportion of Research Grade observations from the overall observations differed among taxonomic groups because Amphibia, Reptilia, and Aves had the best ratio while Other animals (several invertebrate groups, e.g., Platyhelminthes, Annelida, etc.) and Protozoa the worst. Overall, the number of species observed from Hungary on iNaturalist is equivalent to about 21% of the species known to occur in the country. The representativeness differed among taxonomic groups; some iconic taxon groups were well represented, for example, Amphibia, Reptilia and Aves, while others such as Arachnida, Insecta, or Other animals were poorly documented (Table 1).
Table 1
Number of observation records on iNaturalist compared with Research Grade observations and the estimated number of species in Hungary.
| ICONIC TAXON GROUP | NUMBER OF OBSERVATIONS ON INATURALIST | NUMBER OF RESEARCH GRADE OBSERVATIONS | PERCENT OF RESEARCH GRADE OBSERVATIONS | ESTIMATED NUMBER OF SPECIES IN HUNGARY | NUMBER OF SPECIES FROM HUNGARY ON INATURALIST | PERCENT OF HUNGARIAN SPECIES ON INATURALIST |
|---|---|---|---|---|---|---|
| Insecta | 146,579 | 81,547 | 55.63 | 29,356 | 4,756 | 16.20 |
| Plantae | 101,777 | 71,559 | 70.31 | 2,721 | 1,991 | 73.17 |
| Fungi | 19,874 | 9,007 | 45.32 | 3,000 | 873 | 29.10 |
| Aves | 20,854 | 19,211 | 92.12 | 426 | 382 | 89.67 |
| Arachnida | 11,400 | 4042 | 35.46 | 2,694 | 273 | 10.13 |
| Mollusca | 4,171 | 2,404 | 57.64 | 278 | 89 | 32.01 |
| Other animals | 3,668 | 714 | 19.46 | 2,939 | 91 | 3.09 |
| Reptilia | 2,840 | 2,791 | 98.27 | 16 | 16 | 100 |
| Mammalia | 2,625 | 1,971 | 75.09 | 83 | 70 | 84.34 |
| Amphibia | 2,038 | 1,484 | 72.82 | 17 | 17 | 100 |
| Actinopterygii | 589 | 410 | 69.61 | 96 | 52 | 54.16 |
| Protozoa | 336 | 65 | 19.35 | ? | 16 | ? |
| Total | 316,751 | 195,205 | 61.63 | 41,626 | 8,607 | 20.67 |
The earliest observation record was from 1907, and between 0 and 56 were recorded from later years until 2004 (Figure 1a; records shown only from the year 2000). The 4,977 archive records (prior to 2008, the launch year of iNaturalist) comprised 3.2% of the total records. From 2005, records started to appear every year with a small peak between 2006 and 2011 and a rise from a couple hundred records to a maximum of 55,550 in 2023. There were more records from year to year since 2018. Temporal patterns significantly differed from an equal distribution in every case (months: χ2 = 57053.03, df = 11, p < 0.0001, days: χ2 = 20354.5, df = 7, p < 0.0001, hours: χ2 = 179878.76, df = 23, p < 0.0001).
Figure 1
Number of observation records by (a) year of observation, (b) aggregated in months, (c) days, and (d) hours. Records are shown only from 2000 onwards in panel a to make the trend more comprehensible.
The number of observations was low during the winter months, and peaked in May and June (Figure 1b). There were more records on weekend days than weekdays (Figure 1c). Hourly activity peaked in the middle of the days, around 1300 (1:00 PM); however, there was a second, smaller peak around 2000 (8:00 PM) (Figure 1d).
Observations were uploaded by 5,427 people, and the number of records per person showed great variability. The majority of observers uploaded only 1 or 2–10 records (Figure 2), and 45 observers (0.9%) uploaded more than 1,000 observations, from which three of them uploaded more than 10,000 observations.
Figure 2
Number of observation records by the number of observers.
We used 285,988 records (90.3% of total records) for spatial analyses after the exclusion of downgraded and not-precise coordinates. The number of observations in grid cells showed spatial aggregations in certain areas, most notably around Budapest, but also in the Sopron Mountains, Lake Balaton, Kiskunság, the North Hungarian Mountains and the vicinity of the cities of Pécs, Szeged, Miskolc, and Debrecen, among others (Figure 3a). Notable examples of protected areas are shown in Figure 3b, where Duna-Ipoly National Park (NP), Balaton-Felvidék NP, or Zemplén Protected Landscape represent the western and northern, more mountainous part, while Kiskunság NP, Hortobágy NP, or Debreceni Nagyerdő Protected Area represent the lowland middle part of the country. The Shannon diversity of iconic taxa showed a more even pattern throughout the country (Figure 3c). There were slight differences between the number of observations (Figure 3b) and Shannon diversity of iconic taxa (Figure 3d) in protected areas, with rather even distribution throughout the country, with more records and higher diversity from larger patches of protected areas.
Figure 3
Number of observation records in (a) 10 × 10 km grid cells and (b) protected areas, and Shannon diversity index of iconic taxa in (c) 10 × 10 km grid cells and (d) protected areas in Hungary. The capital city of Budapest, selected cities with county rights, and Lake Balaton are labelled in panel a, while selected protected areas are labelled in panel b. NP: national park, PL: protected landscape, PA: protected area.
From the explanatory variables, distance from Budapest, and distance from city with county rights negatively influenced the number of observations in all candidate models (Table 2; Table 3). In contrast, the SDIL, urban areas, forests, mean altitude, proportion of protected areas, and mean gross income positively influenced the number of observations (Figure 3a; Table 2). The Shannon diversity of iconic taxa (Figure 3c) was positively influenced by the SDIL, urban areas, croplands, grasslands, rivers and lakes, mean altitude, and proportion of protected areas (Table 2). When we used protected areas as spatial units, the number of observations (Figure 3b) was positively influenced by the SDIL, mean altitude, and area (Table 3). Finally, the Shannon diversity of iconic taxa in protected areas was positively influenced by wetlands, mean altitude, and area (Figure 3d; Table 3).
Table 2
Results of Generalised Linear Model analyses with spatial variables in 10 × 10 km grid cells. Significant variables are highlighted in bold.
| NUMBER OF OBSERVATIONS | SHANNON DIVERSITY INDEX OF ICONIC TAXA | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ESTIMATE | SE | ADJUSTED SE | z–VALUE | p–VALUE | ESTIMATE | SE | ADJUSTED SE | z–VALUE | p–VALUE | |
| (Intercept) | 0.395 | 0.004 | 0.004 | 105.719 | < 0.00001 | 0.693 | 0.008 | 0.008 | 89.665 | < 0.00001 |
| Shannon diversity of land use types | 0.03 | 0.005 | 0.005 | 6.215 | < 0.00001 | 0.076 | 0.01 | 0.01 | 7.525 | < 0.00001 |
| Urban areas | 0.05 | 0.008 | 0.008 | 5.961 | < 0.00001 | 0.04 | 0.011 | 0.011 | 3.777 | < 0.00001 |
| Croplands | – | – | – | – | – | 0.046 | 0.011 | 0.011 | 4.386 | < 0.00001 |
| Grasslands | 0.007 | 0.005 | 0.005 | 1.479 | 0.139 | 0.035 | 0.01 | 0.01 | 3.417 | 0.001 |
| Forests | 0.019 | 0.006 | 0.006 | 3.408 | 0.001 | 0.016 | 0.012 | 0.012 | 1.367 | 0.171 |
| Wetlands | 0.007 | 0.005 | 0.005 | 1.541 | 0.123 | 0.014 | 0.01 | 0.01 | 1.504 | 0.133 |
| Rivers and lakes | 0.005 | 0.004 | 0.004 | 1.09 | 0.276 | 0.028 | 0.009 | 0.009 | 3.178 | 0.001 |
| Mean altitude | 0.019 | 0.005 | 0.005 | 3.568 | < 0.00001 | 0.061 | 0.012 | 0.012 | 5.122 | < 0.00001 |
| Proportion of protected areas | 0.04 | 0.005 | 0.005 | 7.512 | < 0.00001 | 0.033 | 0.011 | 0.011 | 3.034 | 0.002 |
| Distance from Budapest | –0.017 | 0.005 | 0.005 | 3.278 | 0.001 | –0.024 | 0.011 | 0.011 | 2.17 | 0.03 |
| Distance from city with county rights | –0.022 | 0.004 | 0.004 | 4.913 | < 0.00001 | –0.039 | 0.009 | 0.009 | 4.147 | < 0.00001 |
| Road density | –0.013 | 0.008 | 0.008 | 1.645 | 0.1 | 0.011 | 0.016 | 0.016 | 0.675 | 0.5 |
| Mean gross income | 0.022 | 0.006 | 0.006 | 3.882 | < 0.00001 | 0.022 | 0.012 | 0.012 | 1.764 | 0.078 |
Table 3
Results of Generalised Linear Models analyses with spatial variables in protected areas. Significant variables are highlighted in bold.
| NUMBER OF OBSERVATIONS | SHANNON DIVERSITY INDEX OF ICONIC TAXA | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ESTIMATE | SE | ADJUSTED SE | z–VALUE | p–VALUE | ESTIMATE | SE | ADJUSTED SE | z–VALUE | p–VALUE | |
| (Intercept) | –0.059 | 0.017 | 0.017 | 3.572 | < 0.00001 | –0.02 | 0.031 | 0.031 | 0.659 | 0.51 |
| Shannon diversity of land use types | 0.024 | 0.01 | 0.01 | 2.483 | 0.013 | 0.022 | 0.015 | 0.015 | 1.492 | 0.136 |
| Urban areas | – | – | – | – | – | < 0.00001 | < 0.00001 | < 0.00001 | 0.952 | 0.341 |
| Croplands | 0.001 | 0.007 | 0.007 | 0.172 | 0.864 | < 0.00001 | < 0.00001 | < 0.00001 | 0.352 | 0.725 |
| Grasslands | 0.003 | 0.007 | 0.007 | 0.413 | 0.679 | < 0.00001 | < 0.00001 | < 0.00001 | 0.384 | 0.701 |
| Forests | –0.003 | 0.007 | 0.007 | 0.361 | 0.718 | < 0.00001 | < 0.00001 | < 0.00001 | 1.002 | 0.316 |
| Wetlands | 0.006 | 0.007 | 0.007 | 0.793 | 0.428 | < 0.00001 | < 0.00001 | < 0.00001 | 2.228 | 0.026 |
| Rivers and lakes | –0.003 | 0.007 | 0.007 | 0.464 | 0.643 | < 0.00001 | < 0.00001 | < 0.00001 | 1.257 | 0.209 |
| Mean altitude | 0.028 | 0.007 | 0.007 | 3.776 | < 0.00001 | 0.001 | < 0.00001 | < 0.00001 | 5.07 | < 0.00001 |
| Area | 0.091 | 0.004 | 0.004 | 20.514 | < 0.00001 | 0.08 | 0.004 | 0.004 | 18.313 | < 0.00001 |
| Distance from Budapest | –0.055 | 0.007 | 0.007 | 7.462 | < 0.00001 | –0.001 | < 0.00001 | < 0.00001 | 6.331 | < 0.00001 |
| Distance from city with county rights | –0.029 | 0.007 | 0.007 | 3.907 | < 0.00001 | –0.002 | 0.001 | 0.001 | 4.228 | < 0.00001 |
Discussion
Our results show that the number of observation records on iNaturalist from Hungary has been sharply increasing (Figure 1a), but also that spatial and other sampling biases are prevalent. As Hungary lies in the temperate climatic zone and has four distinct seasons, we expected that most records would be uploaded from the vegetation period (from spring to autumn) when both biota and people are most active. The May and June maxima of records agree with this but also coincide with the end of the school year in Hungary. We also found that people were more likely to make observations in the weekend days than during weekdays, as found in other studies (Di Cecco et al. 2021; Knape et al. 2022). Hourly activity peaked during midday, however there were enthusiastic observers who recorded observations at night as well. The second, smaller peak in the hours (Figure 1d) is a result solely of an increase in Insecta observations, suggesting the usage of iNaturalist by amateur or professional entomologists. Observer engagement was highly fluctuating, with most people using iNaturalist to submit only one or a few records. This could be because some of the observers were from other countries and spent a short time in Hungary, or because they try the application only once or a couple of times and later abandon using iNaturalist. However, there are more than 1,000 observers in Hungary who regularly engaged in species recording (Figure 2). The effect of such observers can be enormous. For example, there is a small peak in the observations per year (Figure 1a) between 2006 and 2011 where about 70% of the observations were made by a single observer who retrospectively uploaded records after joining iNaturalist. These are similar to worldwide trends. For example, a global analysis by Di Cecco et al. (2021) found that the top 10% of observers provide 87%, and the top 1% provide 62% of the observations.
As expected, observations were more frequent from popular taxonomic groups than from less known groups (the detectability and preference filters of Carlen et al. 2024; Table 1). Insecta and Plantae were by far the most uploaded iconic taxa from Hungary, and indeed members of these groups are usually present in every landscape, including gardens and city parks. It is consistent with previous studies that people usually upload observations close to their home and from more developed areas (Di Cecco et al. 2021; Carlen et al. 2024; Vásárhelyi et al. 2026). Our spatial analyses clearly supported this pattern as the number of observations increased with the proportion of urban areas and decreased with the distance from Budapest or major cities (Table 2). Large bodied and more conspicuous groups such as Aves, Amphibia, and Reptilia were also more frequently observed despite their smaller number of species (Callaghan et al. 2021). Other animals and Protozoa, which often contain unpopular, simple looking, small bodied, even microscopic organisms were the least uploaded to iNaturalist both by number of observations and number of species. The patterns are similar in Research Grade observations as certain groups are less likely to be identified by other observers. Our results are in accordance with those of a recent study on the so-called identification crisis in Hungary and globally. Páll-Gergely et al. (2024) used Hungary as a case study to reveal the disappearance of taxonomists in zoology and consequently the erosion of knowledge to identify animals. They found that currently about 45% of the animal species occurring in Hungary have no expert to identify them; all belong to less visible, less popular invertebrate taxa. This trend, and generally that taxonomy is likely in a decline, interested the scientific community in the recent past (Tancoigne and Dubois 2013; Cortés-Fossati et al. 2025). Platforms such as iNaturalist might counter this with the involvement of a larger crowd of motivated, non-professional people in identification (Löbl et al. 2023); however, its effect is yet to be investigated, and research on preference bias is necessary (Deacon, Govender, and Samways 2023).
Observations showed an aggregated and spatially biased pattern (the participation and sampling filters of Carlen et al. 2024; Figure 3a). Most observations were from and around Budapest, and the number of observations decreased with increasing distance (Table 2). Data aggregations also appeared similarly around major cities such as Pécs, Szeged, and Debrecen. The number of observations increased with income and the proportion of urban areas, which overall suggests that most observers record observations from higher-income, urbanised areas, as found elsewhere (Carlen et al. 2024; Di Cecco et al. 2021; Estien, Carlen, and Schell 2024; but see Dimson and Gillespie 2023). These patterns are in congruence with the results of Vásárhelyi et al. (2026), where they found that spatial differences in certain socioeconomic factors, such as average tax base or the proportion of adults with a diploma, affect spatial biases across multiple citizen science projects in Hungary. At the same time, variables associated with the naturalness of the landscape, such as the SDIL, forests, mean altitude, and proportion of protected areas also had positive influence. This supports that areas where people spend their free time and subsequently upload records to iNaturalist are mostly montane, forested, protected areas that serve as destinations for outdoor recreational activities. Trends were not as straightforward for the Shannon diversity of iconic taxa because values were more evenly distributed throughout Hungary (Figure 3c). Generally, taxonomic bias in space was not as prevalent as the bias in the number of observations. Because many cells had only a few observations, we assumed that they were visited by only one or a handful of observers. This implies that observers on average are less specialised on certain taxon groups (e.g., only upload plants) but record different taxa during field visits. Similarly to the number of observations, the Shannon diversity of iconic taxa was positively influenced more by the proportion of urban and naturalness variables (Table 2). Interestingly, the effect of croplands was also positive, which is either an effect of their large proportion in the lowland landscapes of Hungary or an effect of a handful of observers willing to search even in biologically less attractive areas.
In the case of protected areas, those that are closer to cities with county rights and Budapest also had more observations and higher Shannon diversity of iconic taxa (Figure 3b, d; Table 2). It is not surprising that area was the most important predictor with larger protected areas holding more observations and diversity. Higher mean altitude, that is, mountainous areas, also positively influenced both the number of observations and diversity. Notable mountainous protected areas with high numbers of observations are Duna-Ipoly National Park just in the outskirts of Budapest, őrség National Park, and Zemplén Protected Landscape among others. Some of the lowland protected areas also had high number of records, such as the northern part of Kiskunság National Park and Hortobágy National Park, both of which are famous for their distinct “puszta” habitats—a specifically Hungarian cultural and natural grassland landscape. Debreceni Nagyerdő Protected Area clearly shows how an urban forest attracts observers. It is located directly adjacent to Debrecen, the second largest city after Budapest, and has many more records than the similarly sized nearby protected areas (Figure 3b) Among the other variables, only the SDIL for observations and wetlands for diversity were positively significant, implying that the types of habitats were less important for observers when they chose to visit protected areas and uploaded records from there.
From another viewpoint, our analyses are suitable not just to reveal where people go and what they upload but also what they miss, and consequently where there is room for improvement in biodiversity data collection and research. As we already summarised, simple-looking and unpopular invertebrate groups were less likely to be uploaded to iNaturalist. A possible solution to overcome the taxonomic bias in reporting could be the training of enthusiastic volunteers to identify certain groups in dedicated projects. For example, Theobald et al. (2015) found in their quantitative review that those citizen science projects in which volunteers were trained in species identification methods were more likely to be published than those in which they were not. To overcome spatial bias, projects first should reveal sampling coldspots, or in other words, develop maps of ignorance (Rocchini et al. 2011), and then motivate volunteers to explore undersampled areas (Tulloch et al. 2013). Up-to-date and precise maps should be part of such projects, where volunteers could be informed about areas best suited for them to visit. The organisation of bioblitz events to focal areas might be useful to counter sampling bias. The purpose of these events is to collect as much biodiversity data (usually species presence, for practical reasons) as possible in an area in a given time, sometimes in a competitive way. It can significantly increase the amount of data compared with previous stages (Fontúrbel et al. 2022), and the majority of participants find these events very engaging (Rokop et al. 2022).
In Hungary, most of the data collection is skewed towards urban areas and mountainous protected areas, while the landscape between them often remains unexplored. It is understandable as most of the diverse lowland habitats, especially seminatural lowland oak forests, steppe grasslands, and marshlands were converted to arable land in the previous century (Biró, Bölöni, and Molnár 2017), and people usually prefer mountains over lowlands when they go for a hike (Molokáč et al. 2022). To challenge this, future biodiversity mapping projects should encourage observers to visit less popular areas, especially lowlands, rather than legally protected landscapes far from urban settlements. A possible opportunity to overcome data gaps is the involvement of lower-income and/or more rural social groups in data collection with local initiatives. This will support more balanced and less biased biodiversity maps, which can better serve as a basis for scientific studies and conservation decision making.
Supplementary File
Supplemental File 1
Observations uploaded to iNaturalist from Hungary (requested on July 05, 2024). DOI: https://doi.org/10.5334/cstp.861.s1
Data Accessibility Statement
The data analysed is available from https://inaturalist.org and as a supplementary material. The sources of environmental and socioeconomic variables are referenced in the text in detail.
Ethics and Consent
All data are drawn from existing publicly available databases.
Author Contributions
Márton Szabolcs conceptualized the research and designed the methodology. Márton Szabolcs, Bálint Wenner, and Edvárd Mizsei conducted the formal analysis of the data. Márton Szabolcs and Szabolcs Lengyel acquired the research funding. All authors contributed critically to the drafts and gave final approval for publication.