"Flora of Russia" project on iNaturalist

As of 13 Sep 2020, 1,736 members of iNaturalist have joined the project (see also Fig. 3). The core of the project team is formed by 129 people, who are listed simultaneously amongst the top 200 identifiers and top 500 observers of the project, including 15 project members affiliated with the Lomonosov Moscow State University. Of the 129 members, 112 confirmed their formal contribution to this data paper (see the "Author contributions" section and the "Community coverage" section for additional information). Dr. Alexey P. Seregin is the founder and an administrator of the project.

The project covers the territory of the Russian Federation as defined by the national legislation, i.e. including the Republic of Crimea and the City of Sevastopol claimed by Ukraine.

Main features of iNaturalist as a data collection platform

Any user can register as an "observer" on iNaturalist. Users may upload observations of organisms through their account using the website https://www.inaturalist.org/ or the free mobile applications "iNaturalist" and "Seek". A total of 1.28M observers are involved in the work of the platform, including 14.3K observers with at least one observation from Russia (Table 6).

In order to meet the minimum requirements for further scientific use, an observation needs to have: (1) a date; (2) a georeference; (3) a photograph/series of photographs or (for animals) an audio recording(s) of the object's sounds, created by the observer; (4) the organism needs to be recorded in the wild. Provided that these requirements are fulfilled, the observation is marked as "needs ID", regardless of whether the author identified the organism or not. Once an observation receives identical identifications by more than two thirds of the iNaturalist users at the level of species (in some cases of genus), it becomes "research-grade", a category for verified observations. A supporting identification by a second user makes an observation "research-grade" while identification by a single user is not enough. Disagreeing identifications may once again exclude an observation from this category. Low-quality photos or photos of plants accurate identification of which requires a study of some micromorphological, anatomical or genetic traits usually do not reach the "research grade" or, in the latter cases, remain identified and verified only at the level of genus.

The observers may choose a licence allowing further re-use of the data. Observations licensed with three Creative Commons Licences (CC0, CC-BY, CC-BY-NC) and of "research-grade" quality are automatically exported to the GBIF. As of 5 Sep 2020, the iNaturalist database contained 48.6M observations that have met the minimum quality requirements (Table 1), of which 29.2M have achieved "research-grade" (Table 3). Unfortunately, only 19.7M observations have been exported to the GBIF due to copyright restrictions (Table 5).

The implementation of artificial intelligence (AI) for identification is a key feature of iNaturalist (Van Horn et al. 2018, Cui et al. 2018). It gives the users a suggestion about the most similar species after analysing the photos ("visually similar") and taking into account the geographical distribution of other records ("seen nearby"). Initially, the portal's AI compared the newly-uploaded photos with a basic set of images, which, in 2017, comprised 859,000 photos of more than 5,000 species. The images of varying quality had been collected using different types of cameras, but their identifications have been double-checked. Primary results showed that modern AI methods, at that time, gave an accurate identification for 67% of observations, which well illustrates the complexity of the dataset (Van Horn et al. 2018). In 2018, most images of plants and animals from any part of the world were likely to receive from the system an identification of species inhabiting North America. Over the course of 2019 and 2020, AI has almost stopped suggesting incorrect identifications for plants within European Russia (Seregin et al. 2020). It still works somewhat worse with photos from Asian Russia and the Caucasus. Millions of new photos reviewed by the expert community and constantly added to the library of standard images allow AI to improve the performance. Its capabilities are, however, still inferior to expert assessments with regard to certain groups of organisms or certain geographic territories. Nevertheless, the system's general awareness of the world flora is many times larger than that of an individual botanist. In many ways, this particular feature of iNaturalist attracts both amateurs and professionals. The success of iNaturalist has made possible the further use of AI for species recognition by photograph for millions of images in the GBIF database (Robertson et al. 2019).

Portal on the flora of Russia

To collect data on the plant distribution in the City of Moscow, the Moscow University team initially organised the "Flora of Moscow" project on iNaturalist on 29 Dec 2018. An immediate positive feedback from users and a surge of interest forced us to create 85 more regional projects with a uniform ideology in early January 2019 and organise them as part of the "Flora of Russia" umbrella project. Each regional portal automatically includes observations of vascular plants uploaded on iNaturalist which have achieved "research-grade" and found within the administrative boundaries of a specific federal territory. The home page of each regional portal displays its statistics and basic information.

The "Flora of Russia" homepage (Fig. 4) includes a "scoreboard" with a ranking of regional projects (ordered by the number of observations, species and observers), basic statistics, a list of the latest observations, news from the project journal and a general map of all data. There are links leading to the project description, the project journal, the rankings of the top observers (ordered by the number of observations and species), top identifiers, most often recorded species and detailed statistical reports. Thus, both regional projects and the all-Russian portal are organised in the form of ranking tables, stimulating both individual and team activity of observers in accordance with the gamification paradigm (Bowser et al. 2013).

Figure 4.  

The homepage of the "Flora of Russia" project on iNaturalist (Russian-language interface, statistics as of 15 Sep 2020).

The experts (most of whom are the authors of this paper) review the unverified and unnamed observations to suggest the correct name which may either confirm or disprove the opinion of the observer. Typically, most clear photographs from European Russia and the Russian Far East are identified within a couple days after uploading.

The project is functioning on a voluntary basis. Although being created in the Lomonosov Moscow State University, it does not have formal institutional funding. Members of the project search for their own budget for field trips and online activity. Some grants of the co-authors are acknowledged in this paper.

The standard procedure of sampling is described on iNaturalist in the form of 17 paragraphs in the "Observations" section of the help page (last revised 8 Sep 2020 by Sam Kieschnick).

Data quality control is necessary for maintaining a high quality of records within a dataset. In the "Flora of Russia" project description, there is a well-structured, detailed and constantly improving section with recommendations for users in Russian. Apart from the general information (including short videos about iNaturalist and a description of available research tools of the portal), there are two particularly important sections, i.e. "Recommendations for new users" and "Recommendations for event curators". Both sections provide detailed instructions for the user on what, how and where to create a good-quality observation on iNaturalist. However, many users are not familiar with these guidelines. This imposes a certain responsibility on the identifiers and the project curators, who act as data stewards. The most important and/or frequently occurring issues are listed below.

For each project on iNauralist, at least one or two curators should be assigned to review the uploaded observations and make comments, if necessary. The most frequent mistakes are:

1. low-quality or wrong-angle photos,
2. observations of cultivated plants without a relevant indication,
3. either unintentional or intentional duplication of the same observation,
4. unintentional merging of numerous observations into a single one,
5. lack of date or location of an observation,
6. lack of any original identification (at least a coarse one),
7. upload of copyright media.

In some cases, an inaccurate location for an observation shows up automatically, caused by specific GPS settings on the smartphone or camera. We report these issues to observers for further manual correction or mark such observations as "location is not accurate". We highly recommend georeferencing using the "GPS only" mode instead of either "GPS plus mobile networks" and "mobile networks only". The latter two options may shift the observation's georeference to the nearest mobile tower instead of the actual observer's location. Additionally, all records with positional accuracy exceeding 50,000 m were marked as having inaccurate location on 25 Sep 2020 and reported to users in the project journal post. Suspicious positional accuracy of 0, 1 or 2 metres recorded in thousands of observations is an artifact set up automatically during the uploading of observations by the devices.

Another difficult and common problem is the separation of cultivated plants from garden escapes (naturalised or casual). Cultivated plants may be well recognisable and could reach "research grade" rapidly. We ask experts and project curators to double-check "research-grade" observations to detect plants growing only in cultivation.

A well-designed and useful feature in iNaturalist is the possibility to call for attention of a specific user using the "@" prefix (for example, @krestov). This is very important for maintaining the appropriate quality as experts may respond and help in identification.

Undoubtedly, the data quality depends on the quality of the uploaded photographs and field experience of the users. We ask project curators to post links to regional checklists, field guides and illustrated atlases for interested naturalists in the project description.

Constant quality control is especially important during various events such as bioblitzes or mandatory student practices. As their numerous participants mostly lack experience in collecting biodiversity data through iNaturalist, the work of curators and teachers should be constant during the whole period of these events.

Russia is a large country with an area of over 17 million km² and an unevenly distributed human population. For instance, in Chukotka, the population density is only 0.07 people per 1 km², whereas in the City of Moscow, it is 4,925 people per 1 km² (Table 7). The geographic coverage of the dataset is characterised by significant spatial disparities in the presented data for all indexes, including the number of observations, species and observers (Fig. 5).

Figure 5.  

A map of 750K observations from the "Flora of Russia" project showing an extreme disproportion in data coverage (source: iNaturalist.org).

Number of observations. The key index of the "Flora of Russia" project is the number of uploaded observations (Fig. 1). The project reached 750,000 observations of "research-grade" quality on 7 Sep 2020, whereas ca. 135,000 unverified observations make the project's backlog, which is not included in the dataset. The stable snapshot of the dataset produced on 8 Sep 2020 contains 750,143 records (see "Data resources" section).

The City of Moscow topped the project by the number of observations from 18 Aug 2019 to 15 Jun 2020, when Moscow Oblast, the region with the largest community of observers, took the lead (Table 8). Other regions of Central Russia - Bryansk, Tula, Nizhny Novgorod and Kursk Oblasts - hold the third to sixth places in the ranking.

The top 10 regional projects contribute 45.4% of observations of the entire project and this proportion is constantly decreasing due to the growth of the communities in other regions. For instance, the proportion of observations made in the top ten regions was 55.5% on 9 Jan 2020. However, the disproportion in the spatial coverage is obvious even within the leading regions (Fig. 6).

a

b

c

d

e

f

Figure 6.

Progress of the data coverage for the top three most sampled regions within the "Flora of Russia" - 9 Jan 2019 vs. 7 Sep 2020 (source: iNaturalist.org).

a: Moscow Oblast, 9 Jan 2019 - 1,286 observations from 180 users  
b: Moscow Oblast, 7 Sep 2020 - 72,861 observations from 1,906 users  
c: City of Moscow, 9 Jan 2019 - 1,115 observations from 181 users  
d: City of Moscow, 7 Sep 2020 - 66,124 observations from 1,596 users  
e: Bryansk Oblast, 9 Jan 2019 - 67 observations from 4 users  
f: Bryansk Oblast, 7 Sep 2020 - 34,830 observations from 175 users  

Less than 500 observations have been made within each of the four regions of the Caucasian biodiversity hotspot (Kabardino-Balkaria, Chechen Republic, North Ossetia, Ingushetia), desert regions (Astrakhan Oblast, Kalmykia), as well as in Magadan Oblast, Tyva Republic, Jewish Autonomous Oblast and Nenets Autonomous Okrug.

Observations per capita. If we normalise the number of observations per 1,000 inhabitants, it turns out that the two most active communities are in the City of Sevastopol and Kamchatka, followed by Bryansk Oblast, Kursk Oblast, Mordovia, Kostroma Oblast and, unexpectedly, Chukotka, a vast region with a very small population. In general, this index best reflects both the involvement of the local residents in the "Flora of Russia" project and the activity of this particular region's community.

Observation density. Spatial sampling is best characterised by the density of observations per a standard area (for example, per 1,000 km²). The three federal cities are far ahead: Moscow (26K observations), Sevastopol (25K) and St. Petersburg (7K), here, a large urban population is concentrated on a small area. Federal cities are followed by four regions of Central Russia with a relatively small area and active local iNaturalist communities, i.e. Moscow Oblast, Tula Oblast, Chuvashia and Bryansk Oblast.

Observations per recorded species. The number of observations per recorded species is the integrated index which best characterises both the data density and species representation. The gradual accumulation of observations leads to consequent revealing of all known species or, at least, of regularly observed plants. When recording a new species becomes a rare event and an active community still posts many new photos, the average number of observations per species begins to grow rapidly. According to this index, the leaders are Moscow Oblast (65), City of Moscow (61), Bryansk Oblast (31), Tula Oblast (29), Nizhny Novgorod Oblast (27) and Omsk Oblast (25). Regions with rich floras (for example, montainous areas) outperform the relatively-poor plains because more observations need to be made there to record numerous rare species.

41 and 82 Latitude; 19.5 and -169 Longitude.

As of 7 Sep 2020, the "Flora of Russia" project included observations of 6,857 species of vascular plants (Fig. 2). Plants of the World Online (POWO) serves as a taxonomic backbone for tracheophytes on iNaturalist. There are some tools used for automatic, semi-automatic and manual addition of new taxa and modification of the taxonomic information. Reasonable deviations from POWO could be accepted on iNaturalist by the curators after community discussions. The taxonomic opinion of an observer, if necessary, may be recorded in the description section of an individual observation.

Unfortunately, Russia lacks both a modern checklist of vascular plants and a standard flora. Therefore, we could assume that the project covers ca. 55% of the Russian plant diversity out of 12,500 species estimated by Kamelin (2007). That is quite a satisfactory figure since the Russian flora includes many species which require collection and proper identification of herbarium specimens (Hieracium, Alchemilla, Crataegus, some Poaceae and Cyperaceae etc.). There are also many rare endemics in hardly accessible mountain areas and quite a few insufficiently-known species recorded from scattered localities.

The list of the most recorded species of the project includes species which are widespread, easily recognisable and identifiable during all seasons (Table 9). These are mostly perennial herbs tolerant to intensive human activity, but also some common trees. Since the observations are concentrated in European Russia (Fig. 5), top-observed species of the project perfectly match the most common plants of temperate Europe, based on frequency of occurrences in the national grid mapping projects (Seregin 2011). The five most observed species have more than 5,000 observations each.

There are 1215 unique observations in the project. They include:

• 994 species,
• 19 nothospecies (hybrids),
• 133 heterotypic infraspecific taxa,
• 36 homotypic infraspecific taxa,
• 33 genera lacking chance for finer identification.

The following users have created the greatest number of unique species and nothospecies records: R.A. Murtazaliev (195 observations), V.S. Volkotrub (81), M.M. Mallaliev (40), S.A. Nesterova (32), A.I. Pyak (32), S.R. Mayorov (31), S.A. Svirin (27), I.N. Pospelov (25), Aleksandr L. Ebel (23), D.A. Bochkov (23), D.G. Chimitov (19), M.S. Knyazev (17), A.P. Seregin (17), E.S. Kashirina (16), A.V. Popov (13), O.A. Chernyagina (13), E.A. Razina (12) and N.S. Liksakova (12). Altogether, 224 observers were lucky to contribute at least one unique observation of a species or a hybrid.

To assess the regional representation of our data, we have compiled a table on the regional diversity of the Russian flora with necessary references (Table 10). The numbers of known species across the regions are not always perfectly comparable, since the authors of regional floras, guides and checklists used various species concepts which were either "splitters" or "lumpers". The overestimate for Volgograd Oblast (Sagalaev 2008) is especially notable.

Table 10 gives a good overview of plant diversity across the country with notable hotspots revealed earlier by Malyschev for standard areas (Malyshev 1975, Malyshev 1992):

1. Caucasus (Dagestan, Krasnodar Krai, Kabardino-Balkaria, North Ossetia, Chechen Republic, Stavropol Krai etc.);
2. Russian Far East (Primorsky Krai, Khabarovsk Krai);
3. Southern Siberia (Irkutsk Oblast, Altai Krai, part of Krasnoyarsk Krai etc.);
4. Crimea.

Additionally, the Caucasus is listed as the only biodiversity - and especially tracheophyte diversity - hotspot of global importance in Russia (Myers et al. 2000, Barthlott et al. 2007).

We present data on the taxonomic diversity of vascular plants within the regions (Table 11) using two indexes, i.e. (1) the number of the recorded species and (2) the number of the lowest-rank taxa of the taxonomic tree with "research grade". The second index includes varieties, subspecies, species and those genera which cannot be accurately identified to species rank by uploaded photos (for instance, Alchemilla, Pilosella, Hieracium, Euphrasia and some genera requiring specific features not always captured by observers, like Melilotus and Epilobium without flowers etc.). The number of the recorded species is more suitable for the further taxonomic analysis.

The community has observed the highest number of plant species in Dagestan (1,927 species), which is the richest region of Russia in terms of the number of known species (Table 10). However, the rich flora of Dagestan is still represented here by only 57.0% of known species (Murtazaliev 2016). It is followed by other six territories of the Russian Federation from the above mentioned plant diversity hotspots — the Caucasus (represented by Krasnodar Krai), the Crimean Peninsula (the Republic of Crimea and the City of Sevastopol), the southern part of the Russian Far East (Primorsky Krai) and the mountains of Southern Siberia (Altai Krai). Four well-represented regions of Central Russia (Kursk, Voronezh, Moscow and Bryansk Oblasts) form the next group with a good proportion of flora detection ranging from 56.3% to 81.8%.

If we consider the number of species known from each region, Kursk Oblast is the leader in terms of the proportion of observed species (81.8% out of 1,409 known species). Nizhny Novgorod Oblast, Bryansk Oblast, Kostroma Oblast and the City of Sevastopol also have over 70% of known species already represented on iNaturalist, although Nizhny Novgorod Oblast lacks a modern flora checklist, since the number of taxa for the region published by Bakka and Kiseleva (2008) is out of date.

In regional lists, 53 species are counted as the leaders of the scoreboards (Table 12). This list includes some notable invasive species like Acer negundo (a leader in five regions), Heracleum sosnowskyi (two regions), Ambrosia artemisiifolia, Erigeron annuus, Hordeum jubatum, Impatiens glandulifera, and Lupinus polyphyllus (one region each). Usually, this high performance of invasive alien species is a result of intentional recording in line with a regional assessment of aliens performed by the project members.

Conifers, which could be observed during the whole year, provide another example of the most observed species across regions. For instance, Pinus sylvestris is a top species in ten regions as well as Larix gmelinii (two regions), Picea obovata, Pinus sibirica and Juniperus deltoides (one region).

Other examples include common plants of meadows and urban lawns (e.g. Achillea millefolium, Taraxacum aggr. officinale, Tanacetum vulgare, Artemisia vulgaris, Cirsium arvense, Cichorium intybus), common tundra plants in Arctic regions (Cornus suecica, Papaver pulvinatum, Rubus chamaemorus), species abundant in the taiga zone (Vaccinium vitis-idaea, Chamaenerion angustifolium) or common plants of dry grasslands (Tulipa suaveolens, Fragaria viridis, Cichorium intybus). There are exceptional cases of over-recording of orchids (Cypripedium macranthos, Dactylorhiza euxina).