Methods |
|
Corresponding author: Christiaan P. Gildenhuys ( christiaan.gildenhuys@gmail.com ) Academic editor: Anibal Pauchard
© 2025 Christiaan P. Gildenhuys, Luke J. Potgieter, Cang Hui, David M. Richardson.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Gildenhuys CP, Potgieter LJ, Hui C, Richardson DM (2025) Deriving inventories of non-native plant species from iNaturalist: Insights from urban centres of the Western Cape, South Africa. NeoBiota 104: 27-58. https://doi.org/10.3897/neobiota.104.155832
|
Accurate, up-to-date inventories of non-native species are important to document and improve our understanding of biological invasions globally and inform management decisions. Traditional methods for the collation of inventories are time- and resource intensive, and lists become outdated if not regularly updated. The community science platform iNaturalist can contribute to the collation of regularly updatable (“living”) inventories of non-native species. However, robust and transparent workflows are needed to optimise data quality to take full advantage of iNaturalist. We present the semi-Automated Non-Native Inventory Compilation (sANNIC) workflow for the collation and completeness assessment of non-native vascular plant inventories from iNaturalist. The workflow is informed by the World Checklist of Vascular Plants (WCVP) and is used to compare native ranges to a reference area. The utility of the workflow is demonstrated by compiling non-native species inventories of 100 urban centres in the Western Cape province, South Africa. A total of 947 taxa of wild-growing, i.e. casual, naturalised and invasive plants were observed in these urban centres which showed varying levels of sample completeness. Most small towns had too few records for a completeness assessment. Larger urban centres and those near the coast were typically better sampled. This work highlights the potential for iNaturalist to construct non-native species inventories given sufficient coverage and thorough curation.
Alien plant species, biological invasions, citizen science, community science, inventory, plant invasions, urban ecology
Biological invasions are a major threat to biodiversity and human well-being globally and their impacts are projected to increase (
Despite the utility and importance of inventories and species lists, their collation and presentation are often haphazard. A wide range of methods are employed to compile inventories, including systematic surveys, literature searches, and expert opinion (
The community science (also known as citizen science) platform iNaturalist (www.inaturalist.org) can help to produce more consistent, easily updatable (or “living”) inventories, and allow for the quantification of uncertainty and sampling effort. iNaturalist is an unstructured community science initiative which allows voluntary participants (observers) to add occurrence records of any taxon by uploading one or more images (or sound files) of a single species with the associated time, location, and other information onto the online platform (
iNaturalist has been widely applied in many parts of the world to study and monitor non-native invasive species. For example,
Although community science projects provide useful data, limitations can cause misleading results if biases and inaccuracies are not properly addressed (
Workflows that can be regularly and cost-effectively repeated offer the opportunity to streamline the collation of up-to-date inventories, and can standardise and improve reproducibility, transparency, and data quality. For example,
Given the inconsistent sampling effort of opportunistic data, reporting on the likely completeness of inventories generated from iNaturalist is vital. Tools for assessing sample completeness and diversity estimation of incomplete samples could address this need (
Urban centres provide a good focus for the development of such a workflow. Although urban centres are not the primary focus for many iNaturalist users (
We aim to develop a semi-automated workflow, the “semi-Automated Non-Native Inventory Compilation” (sANNIC), for deriving non-native plant inventories and assessing their completeness from iNaturalist data. To this end we: a) develop an automated R-based script “Automated Non-Native Inventory Compilation” (ANNIC) to automatically build preliminary non-native species lists; b) integrate these lists into iNaturalist projects to thoroughly curate each species and ensure the highest possible data quality and optimal use of iNaturalist; and c) propose a protocol for the evaluation of species list completeness using principles of Chao completeness assessment. We then apply this workflow in 100 urban centres in the Western Cape province, South Africa, to produce an openly available and constantly updating living inventory of non-native species in urban centres. Finally, we discuss the implications, limitations and transferability of the proposed workflow.
Here we describe the proposed semi-Automated Non-Native Inventory Compilation (sANNIC) workflow (Fig.
Representation of the semi-Automated Non-Native Inventory Compilation (sANNIC) workflow for the derivation of inventories of non-native plant taxa from iNaturalist. Steps outlined in a black square represent those conducted in R. Steps highlighted in light orange represent those steps that require working in iNaturalist. Diamond shapes represent decision points. Rounded shapes represent the outcome of the workflow. Asterisks (*) indicate optional steps.
The area and taxon of interest first need to be defined on iNaturalist. The simplest option is to create a project to capture the relevant information. Areas of interest can be uploaded (if not already present) as ‘places’ which can be used to refine a search query or used as a geographic filter in a project. The taxonomic group of interest can be specified using a search query in the project settings (see
There are several options for exporting data from iNaturalist depending on the amount of data needed. Data can be accessed using the export function on the iNaturalist website which allows for up to 200 000 records per query. Various Application Programming Interfaces (APIs) are available, for example R packages such as ‘rinat’, which can export up to 10 000 observations per day (
Although our workflow is tailored to work with observations exported directly through the iNaturalist website, it can be modified to function with different data sources. For our workflow the following columns need to be included in the iNaturalist export: id (a unique identifier for each observation), observedon (the standardised date the observation was made), user_id (a unique identifier for each user), quality_grade (including “Research Grade” and “needs id”), captive_cultivated (whether the observation is flagged as captive or cultivated), longitude and latitude (the coordinate location associated with each observation), taxon_id (a unique taxonomic identifier), taxon_family_name (family name of the taxon as recorded on iNaturalist), taxon_subfamily_name (subfamily name of the taxon as recorded on iNaturalist), taxon_genus_name (genus name of the taxon), taxon_species_name (species name of the taxon if identified to species level), taxon_hybrid_name (hybrid name of the taxon if identified as a hybrid), taxon_subspecies_name (subspecies name of the taxon if identified as a subspecies). Additional columns like the url (a link to the web address of each observation), positional_accuracy (precision of the coordinates) and others are not required but can be useful to include in any export. Any additional columns can be added to the input as needed.
The Automated Non-Native Inventory Compilation (ANNIC) R script automatically classifies species as native or non-native based on the World Checklist of Vascular Plants (WCVP) by using iNaturalist occurrence records as input and assisting users in manually reviewing species that cannot be assigned automatically. Written in R version 4.4.3 (R Core Team 2025), it is available as an R script on Zenodo (https://zenodo.org/records/15210704) and the latest version is available on GitHub (https://github.com/christiaan-g/sANNIC_workflow). It requires the ‘tidyverse’ packages to conduct data manipulation (
Some taxa, regardless of image quality of the record, cannot be consistently identified to species level from iNaturalist records. This is mainly because of unclear distinctions between species within a genus or other taxonomic group, difficulty in distinguishing similar species, absence of sufficiently high-resolution images showing key traits required for species-level identification, hybridisation, difficulty in identifying taxa at the juvenile stage, and a lack of available taxonomic knowledge to identify records to species level. Such taxa are considered “problematic taxa”; the R script allows for these taxa to be collapsed to a higher taxonomic rank, including genus, family, subfamily, or any manual specification. These taxa will be included if present in the dataset regardless of their quality grade; special attention is therefore needed during the curation process (see section 2d. “Manual checking”). This step is optional but when working with iNaturalist data we recommend assessing the presence of problematic taxa. All problematic taxa in this workflow are assumed to be non-native.
The World Checklist of Vascular Plants (WCVP) is a continuously updated database of described vascular plant species curated by the Royal Botanic Gardens, Kew (
Names are matched to the WCVP to ensure congruity between the different data sources, using exact and fuzzy matching algorithms in the ‘rWCVP’ package (
Venn diagram visualising the output of the ANNIC-R script. Proportions are not to scale. The large green circle represents taxa listed in the World Checklist of Vascular Plants (WCVP) and for which native distribution information is available. The orange circle represents taxa listed in our region of interest in iNaturalist. Species above the black line are native, those below the line are non-native. The small dashed green circle represents species which were listed in the WCVP with incorrect native distribution information. a , b. Represent native and non-native species for which good matches were not found in the WCVP or for which distribution information is not available; c, b. Represent species listed in our project for which good matches and distribution information was available in the WCVP; e, f. Represent taxa for which the distribution information in the WCVP was incorrect.
The WCVP stores distribution data according to the World Geographic Scheme for Recording Plant Distributions (WGSRPD;
All taxa previously flagged for manual checking because of unresolved matching with the WCVP or missing native range information in the WCVP, can be manually assessed. While this step is optional, omitting it will result in the exclusion of certain taxa (Fig.
Some species listed in the WCVP have erroneous distribution data and can therefore be falsely flagged as native or non-native (Fig.
After a non-native species list is produced, a new project can be created on iNaturalist and set to collate all taxa in the list. This is done in the project settings by manually entering each species name. Due to limitations on the number of taxa that can be added to an iNaturalist project, some taxa can be listed at genus level where all species in that genus are non-native to the reference area. For example, in South Africa, we included the genera Acacia, Eucalyptus, and Pinus for which all species are non-native. If the number of taxa is still too large, one solution is to create multiple co-occurring projects each targeting a number of species in the allowed limit. For example, these can be grouped by alphabetical names of families.
Once an iNaturalist project has been created, records need to be systematically reviewed to ensure, and improve, data quality. This should be done within the iNaturalist platform because improving the records will ultimately contribute to improving the overall data quality on the platform. We also recommend engaging with users on the platform to aid in the collection and identification of relevant records. Many users on iNaturalist may also be relevant stakeholders in the region of interest whose activities, livelihoods, or well-being could be directly or indirectly affected by biological invasions. The workload of this manual curation step varies depending on the region and the taxon of interest. Some taxa are already well curated on iNaturalist (
We recommend investing effort in evaluating whether iNaturalist records represent captive/cultivated records. This workflow is intended for wild-growing plants, but iNaturalist records are often not correctly tagged as captive/cultivated. For plants, records can be tagged as cultivated based on considerations including, signs of cultivation or care (including trimming, weeding, support, planting pattern, etc), life stage (e.g., seedlings are more likely to be wild-growing/self-sown than large adults which may have been planted long ago), and environmental setting (e.g. plants in a garden are more likely to be cultivated than plants in a riparian zone or semi-natural vegetation). Other information such as prior knowledge of the species, location, observer notes, and communication with observers can be useful in separating wild-growing from cultivated plants. For discussion of considerations, see
Sample coverage is a concept first developed by Alan Turing (1912–1954) and later published by
We suggest calculating sample completeness measures, using either abundance or incidence-based methods, at orders q = 0,1,2 using the Completeness function in the ‘iNext.4steps’ package in R (
The major assumption of completeness estimation is the independence of samples, which is not always the case with opportunistically collected data. Sometimes observers will record the same species multiple times in quick succession, ignoring all other species, thereby violating the assumption of independence. This can severely bias sample completeness estimation, especially when there are few observations. We recommend data coarsening when this is the case so that observations by the same observers, of the same species, in the same area on the same day are assumed to be duplicates and then removing the additional records. This effectively produces a more conservative sample coverage estimate but accounts for cases where extreme biases are likely. Other stricter or more lenient corrections can be made depending on the needs of the user. Note that this will not remove the general bias generated from opportunistic data but will at least mediate some of the more severe cases. If there is inflation in the number of singletons and doubletons compared to a ‘true’ sample or if the heterogeneity of observation probability of unseen species is highly skewed compared to the ‘true’ heterogeneity, the sample completeness can still be underestimated or overestimated. Nonetheless, the proposed completeness assessment can be a highly useful tool to assess how complete the iNaturalist data are with appropriate interpretation.
We showcased the application of the sANNIC workflow to derive non-native species inventories from iNaturalist by applying it to 100 urban centres in the Western Cape, South Africa. We created and uploaded several urban shapefiles (Fig.
Urban centres of the Western Cape, South Africa, included in the study to develop a workflow to produce inventories of non-native plant taxa. Boundaries of urban centres were based on the built-up area from the South African National Land Cover database (available at: https://egis.environment.gov.za/sa_national_land_cover_datasets). Numbers refer to the urban centres listed in Table
In this project, we are interested in all non-native species which have wild-growing (i.e. non-cultivated) records. This includes all species in category C1 or higher according to the
We exported over 300 000 records of vascular plants in the project directly from the iNaturalist website recorded after 2017, with at least one photo, and with an accuracy unknown or less than 1000 m using a series of queries on the export page. Most records (>95%) contributed to iNaturalist in the Western Cape urban centres were observed after 2017 and any records before this time may not reflect the current non-native flora. Observations with an accuracy of greater than 1000 m were excluded as there is a high likelihood that these records fall outside the urban delineations. Approximately 2% of records with accuracy values exceeded 1000 m. We assume, thus, that records with no accuracy recorded fell within the urban delineations, seeing this as an acceptable omission we included all records with no accuracy recorded. We included all the variables listed in Step 1 for export and added the following variables to allow for easier quality checking only: updated_at (the datetime that changes were last made to the observation), license (license or waiver chosen by observer for the observation), url (a link to the web address of each observation), num_identification_agreements (number of identifications that agree with the observer’s identification), num_identification_disagreements (number of identifications that disagree with the observer’s identification), positional_accuracy (precision of the coordinates), and scientific_name (the scientific name for each taxon).
All subspecies were collapsed into species, and we considered several problematic taxa which could not be easily and reliably identified and were collapsed to the specified taxonomic rank in Appendix
Note that our list of “problematic taxa” could change in the future as more information becomes available about these taxa or as we become aware of other difficult taxa. As we could not attain Research Grade quality for these records due to the requirement for species-level identification, each record was carefully checked to ensure it was correctly identified to at least the rank to which it was aggregated.
All taxon names were matched with the latest snapshot of the WCVP (updated May 2024) using the ‘rWCVP’ package. Any taxa which achieved poor matches were manually checked to see whether the suggested matches were correct. Most of these taxa with poor matches were either taxa not yet listed in the WCVP, obscure or artificial hybrids, typographic differences (e.g. “Opuntia monacantha” vs. “O. monacanthos”) and non-standard names (e.g. “Hermania sp12”). We checked the native/non-native status of all taxa that needed manual checking by conducting a search for the species name on Google Scholar or referring to local authorities such as the South African National Biodiversity Institute (SANBI) website (https://pza.sanbi.org/; last accessed 25 March 2025). Hybrids were included as non-native if at least one of the parent taxa were non-native to our study region (e.g., Rubus × fultus which is a hybrid of the native R. pinnatus and non-native R. affinis).
The WGSRPD splits South Africa into four botanical countries: Cape Provinces (CPP; comprising the modern provinces of the Eastern, Northern, and Western Cape), Free State (OFS), KwaZulu-Natal (NAT), and Northern Provinces (TVL; comprising the modern provinces of Gauteng, Limpopo, Mpumalanga, and Northwest). For the purposes of this case study, CPP was used as the region of interest and only taxa recorded in the WCVP as native to the CPP were considered native. Taxa not recorded in the CPP or recorded as introduced in the CPP were considered non-native. We manually uploaded all taxon names compiled in this non-native list to a project in iNaturalist titled “Non-native plants of Western Cape urban areas” (https://www.inaturalist.org/projects/non-native-plants-of-western-cape-urban-areas; last accessed 3 March 2025). We manually curated the project by assessing the cultivation status and correct identification of records as discussed in Step 3. Any native species which were found to be incorrectly included in the project were removed and were recorded as manual native species for future iterations of the workflow (see section 2e). Any non-native species which were incorrectly excluded from the project (based on our knowledge of local invaders) were added and recorded as manual non-native species. The completeness assessment was conducted for the Western Cape urban centres in its entirety and for each of the urban centres which had adequate records (Appendix
In our Western Cape case study, we aimed to identify the most prevalent taxa within our region of interest. As a proxy of prevalence, we ranked each taxon according to the number of observations in an urban centre because species that are more prevalent are also likely to be observed more frequently on iNaturalist. This consideration alone, however, biases the list in favour of the best-sampled urban centres. One approach to address this is to count the number of urban centres in which each taxon was found; however, this does not account for the prevalence of species within each urban centre. To account for sampling variations among urban centres, we thus derived a prevalence index that is calculated as the sum of a taxon’s inverse rank squared for each urban centre weighed by the urban centre’s sample coverage (Sample Completeness, q = 1). Let PIj be the prevalence index of species j; n the number of urban centres; SCi the sample coverage at urban centre i (i.e. sample completeness of order q = 1; Appendix
We reported some of the patterns observed in the non-native flora across the Western Cape urban centres by determining the most represented families and the most common life form descriptions extracted from the WCVP. These were determined by number of species, number of total observations, and number of species weighted by their prevalence index scores. We also compared the taxon list to the GloNAF list of known naturalised or invasive species for both the region and globally (version 2.01 available at: 10.5281/zenodo.13235357). We collapsed all listed problematic species (Appendix
Applying our workflow to the Western Cape urban centres yielded a total of 4467 vascular plant taxa, excluding the collapsed problematic taxa. Matches in the WCVP were found for 4405 taxa, of which 4364 were exact matches, 34 were edit distance fuzzy matches and seven were phonetic matches. Multiple matches were found for 496 taxa, and no matches were found for the remaining 62 taxa. Most multiple matches could be resolved algorithmically. All taxa with edit distance fuzzy matches, unresolved multiple matches, no matches, and for which no distribution information was available on the WCVP, were flagged for manual checking. A total of 126 matches of 108 taxa had to be manually checked (Fig.
Taxa erroneously flagged in the WCVP as non-native in our region of interest were removed from the list, namely Pteridium aquilinum, Tarchonanthus littoralis, Tetragonia fruticosa, Scabiosa columbaria, Solanum rubetorum, Senecio pinifolius, and Salicornia pilansii. We also added non-native taxa which were erroneously excluded from the inventory, namely Cotula australis and Rumex acetosella. All these taxa had erroneous native range information in the WCVP, highlighting the importance of this reviewing step.
We consulted with many local and international taxonomic experts through iNaturalist. Many identifications were made by the authors DMR, CPG, and LJP, reflecting our own expertise and personal consultations and reference to the taxonomic literature. Remaining uncertainties in species identifications, for various reasons, were accounted for by including them in our list of problematic taxa (Appendix
After curation of the data, we improved the percentage of Research Grade records in the project (excluding problematic taxa) from 86.4% to 94%. Given the variation in image quality on iNaturalist, we expect that it will be difficult to increase this number further. The nomenclature between iNaturalist and the WCVP are largely congruent as iNaturalist adopts the WCVP nomenclature to a large extent (
A total of 947 wild-growing non-native taxa with 61 259 corresponding records were observed in iNaturalist. The completeness profile of the 100 urban centres, as a whole, indicates a high sample completeness across all measures. The rare species sensitive sample completeness q = 0, i.e. the proportion of estimated species which were observed, was 0.7823 (95% CI: 0.7314–0.8331); Sample coverage, q = 1, was 0.9958 (95% CI: 0.9953–0.9962); and the abundant species sensitive measure, q = 2, was close to 1. This indicates that almost all relatively abundant species were recorded and as many as 78% of all non-native species, including very rare species, were recorded. A total of 219 taxa were only observed once and 91 observed only twice. The top 100 species ranked by the Prevalence Index are shown in Table
Top 100 most prevalent non-native plant taxa in urban centres of the Western Cape, South Africa derived from iNaturalist using a semi-automated workflow. Taxa marked with an asterisk (*) represent mainly invaders of patches of natural or semi-natural vegetation at the edge or embedded within the boundaries of urban centres.
| Taxon | Prevalence index | Number of urban centres found | Total number of observations |
|---|---|---|---|
| Acacia cyclops* | 11.665 | 37 | 722 |
| Acacia saligna* | 11.465 | 45 | 1794 |
| Plantago lanceolata | 10.271 | 58 | 1584 |
| Echium plantagineum | 9.052 | 41 | 1551 |
| Acacia mearnsii* | 8.362 | 33 | 617 |
| Lantana camara | 7.263 | 29 | 583 |
| Tropaeolum majus | 6.234 | 29 | 1139 |
| Pinus* | 6.200 | 30 | 943 |
| Malva arborea | 5.755 | 25 | 256 |
| Lagurus ovatus | 5.431 | 21 | 590 |
| Briza maxima | 5.280 | 28 | 755 |
| Hypochaeris radicata | 5.125 | 29 | 1015 |
| Solanum mauritianum | 5.084 | 22 | 472 |
| Ricinus communis | 4.897 | 35 | 435 |
| Schinus terebinthifolia | 4.836 | 29 | 646 |
| Myoporum insulare | 4.792 | 28 | 186 |
| Trifolium repens | 4.726 | 25 | 495 |
| Physalis peruviana | 4.560 | 25 | 380 |
| Cenchrus clandestinus | 4.548 | 29 | 791 |
| Oenothera sect. Oenothera | 4.517 | 25 | 339 |
| Silene gallica | 4.465 | 24 | 333 |
| Sonchus oleraceus | 4.358 | 33 | 743 |
| Nephrolepis cordifolia | 4.349 | 23 | 362 |
| Melia azedarach | 4.292 | 36 | 543 |
| Medicago polymorpha | 4.207 | 31 | 427 |
| Spartium junceum | 3.884 | 24 | 400 |
| Gaudium laevigatum* | 3.860 | 19 | 235 |
| Eucalyptus* | 3.851 | 23 | 699 |
| Erodium moschatum | 3.751 | 25 | 500 |
| Oenothera lindheimeri | 3.743 | 28 | 237 |
| Catharanthus roseus | 3.731 | 24 | 363 |
| Cenchrus setaceus | 3.717 | 22 | 467 |
| Cirsium vulgare | 3.703 | 26 | 381 |
| Lysimachia loeflingii | 3.695 | 29 | 262 |
| Cortaderia selloana | 3.621 | 17 | 384 |
| Anredera cordifolia | 3.560 | 21 | 264 |
| Trifolium angustifolium | 3.426 | 18 | 498 |
| Stenotaphrum secundatum | 3.424 | 20 | 426 |
| Helminthotheca echioides | 3.416 | 25 | 489 |
| Opuntia ficus-indica* | 3.315 | 26 | 170 |
| Acacia melanoxylon* | 3.297 | 18 | 346 |
| Verbena bonariensis | 3.256 | 22 | 322 |
| Avena | 3.154 | 26 | 497 |
| Hedera | 2.966 | 19 | 668 |
| Arundo donax | 2.959 | 27 | 321 |
| Lotus subbiflorus | 2.940 | 21 | 269 |
| Solanum nigrum | 2.882 | 27 | 378 |
| Fumaria muralis | 2.839 | 23 | 436 |
| Coleus barbatus | 2.799 | 22 | 426 |
| Thunbergia alata | 2.789 | 19 | 389 |
| Euphorbia peplus | 2.763 | 20 | 381 |
| Limonium sinuatum | 2.742 | 17 | 56 |
| Taraxacum officinale | 2.727 | 22 | 438 |
| Cyrtomium falcatum | 2.671 | 16 | 123 |
| Melilotus indicus | 2.659 | 23 | 201 |
| Lysimachia arvensis | 2.629 | 20 | 145 |
| Oxalis corniculata | 2.602 | 21 | 407 |
| Chenopodiastrum murale | 2.592 | 23 | 222 |
| Datura stramonium | 2.587 | 22 | 197 |
| Atriplex semibaccata | 2.516 | 23 | 81 |
| Melaleuca viminalis | 2.414 | 17 | 445 |
| Myoporum montanum | 2.409 | 23 | 133 |
| Paspalum urvillei | 2.377 | 19 | 248 |
| Tecoma stans | 2.367 | 19 | 78 |
| Brassica tournefortii | 2.306 | 20 | 187 |
| Euphorbia terracina | 2.292 | 15 | 334 |
| Opuntia monacantha | 2.288 | 16 | 126 |
| Erigeron karvinskianus | 2.267 | 16 | 282 |
| Phytolacca octandra | 2.258 | 20 | 164 |
| Bidens pilosa | 2.231 | 16 | 173 |
| Hibiscus trionum | 2.196 | 16 | 135 |
| Vinca major | 2.182 | 20 | 351 |
| Pittosporum undulatum | 2.177 | 17 | 192 |
| Cestrum laevigatum | 2.176 | 17 | 246 |
| Ipomoea indica | 2.065 | 18 | 411 |
| Acacia longifolia | 2.055 | 15 | 464 |
| Canna × hybrida | 2.040 | 20 | 153 |
| Verbascum virgatum | 2.024 | 20 | 172 |
| Erigeron bonariensis | 2.016 | 19 | 412 |
| Vicia benghalensis | 1.984 | 15 | 202 |
| Hakea sericea* | 1.956 | 16 | 62 |
| Paraserianthes lophantha | 1.928 | 15 | 406 |
| Raphanus raphanistrum | 1.925 | 19 | 374 |
| Cerastium glomeratum | 1.920 | 19 | 172 |
| Rumex crispus | 1.907 | 18 | 191 |
| Foeniculum vulgare | 1.899 | 19 | 264 |
| Limonium perezii | 1.799 | 14 | 145 |
| Kalanchoe delagoensis | 1.795 | 15 | 50 |
| Vicia sativa | 1.792 | 18 | 182 |
| Sesbania punicea | 1.788 | 20 | 282 |
| Myoporum laetum | 1.786 | 18 | 103 |
| Erigeron sumatrensis | 1.784 | 17 | 250 |
| Nothoscordum gracile | 1.769 | 17 | 269 |
| Amaranthus | 1.766 | 21 | 278 |
| Acacia podalyriifolia | 1.756 | 16 | 87 |
| Geranium molle | 1.744 | 17 | 201 |
| Portulaca oleracea | 1.692 | 17 | 241 |
| Nasturtium officinale | 1.679 | 10 | 194 |
| Solanum lycopersicum | 1.652 | 15 | 121 |
| Diplotaxis muralis | 1.636 | 14 | 67 |
The most represented families by number of species were Fabaceae, Asteraceae, Poaceae, Solanaceae, and Myrtaceae (Suppl. material
The distribution of records across the urban centres was highly uneven, with Cape Town having the highest number of records (31 034), followed by Stellenbosch (13 689), Somerset West (3 156), George (2 721), and Mossel Bay (1191). The observed species richness was also highest in Cape Town (686), followed by Stellenbosch (410), Somerset West (383), George (376), and Hermanus (253). According to sample coverage q = 1, the best sampled urban centres were Cape Town (0.994), followed by Stellenbosch (0.991), Somerset West (0.963), George (0.949), and Mossel Bay (0.941). Completeness estimates, number of records and species richness for all urban centres are shown in Table
Completeness profiles for four urban centres in the Western Cape ranging from most complete (top left) to least complete (bottom right). Panels (a–d) represent urban centres with the following profiles. a. Highly complete sample coverage (q = 1, > 0.8); b. Lower completeness, but with a high proportion of highly abundant species sampled (q = 2, > 0.8); c. Poor sampling, but a rising completeness profile indicates that sample completeness can at least be estimated; d. Poor sampling and a falling completeness profile indicates that estimates of sample completeness may be unreliable.
One hundred urban centres in the Western Cape, South Africa, included in the study to derive an inventory of non-native plant taxa from iNaturalist using a semi-automated workflow. SC.LCL is the lower confidence limit (95%) and SC.UCL is the upper confidence limit of the sample coverage (sample completeness for q = 1).
| Nr | Urban centre | Area (km2) | Number records | Number species | Sample Coverage | SC.LCL | SC.UCL |
|---|---|---|---|---|---|---|---|
| 1 | Abbotsdale | 2.32 | 0 | 0 | NA | NA | NA |
| 2 | Albertinia | 3.48 | 3 | 3 | NA | NA | NA |
| 3 | Amalienstein & Zoar | 3.31 | 1 | 1 | NA | NA | NA |
| 4 | Ashton | 4.93 | 7 | 5 | NA | NA | NA |
| 5 | Atlantis | 16.44 | 20 | 15 | NA | NA | NA |
| 6 | Barrydale | 2.91 | 37 | 29 | 0.281 | 0.139 | 0.423 |
| 7 | Beaufort West | 14.56 | 45 | 18 | NA | NA | NA |
| 8 | Bella Vista | 2.55 | 0 | 0 | NA | NA | NA |
| 9 | Betty’s Bay | 8.48 | 753 | 161 | 0.900 | 0.881 | 0.918 |
| 10 | Bonnievale | 7.24 | 0 | 0 | NA | NA | NA |
| 11 | Bot River | 1.74 | 47 | 30 | 0.374 | 0.175 | 0.573 |
| 12 | Brakrivier | 24.96 | 781 | 195 | 0.868 | 0.845 | 0.891 |
| 13 | Bredasdorp | 7.50 | 46 | 36 | 0.323 | 0.154 | 0.492 |
| 14 | Brenton | 2.51 | 175 | 84 | 0.743 | 0.688 | 0.797 |
| 15 | Caledon | 6.23 | 55 | 31 | 0.569 | 0.456 | 0.682 |
| 16 | Calitzdorp | 3.86 | 25 | 17 | NA | NA | NA |
| 17 | Cape Town | 846.78 | 31034 | 686 | 0.994 | 0.994 | 0.995 |
| 18 | Ceres | 6.65 | 8 | 8 | NA | NA | NA |
| 19 | Chatsworth | 2.95 | 0 | 0 | NA | NA | NA |
| 20 | Citrusdal | 4.20 | 0 | 0 | NA | NA | NA |
| 21 | Clanwilliam | 4.79 | 14 | 7 | 0.601 | 0.353 | 0.850 |
| 22 | Dana Bay | 4.84 | 213 | 81 | 0.746 | 0.703 | 0.788 |
| 23 | Darling | 4.16 | 37 | 19 | 0.731 | 0.585 | 0.878 |
| 24 | De Doorns | 9.05 | 0 | 0 | NA | NA | NA |
| 25 | Dysseldorp | 3.01 | 1 | 1 | NA | NA | NA |
| 26 | Fisantekraal | 6.39 | 2 | 1 | NA | NA | NA |
| 27 | Fisherhaven | 1.94 | 564 | 75 | 0.926 | 0.908 | 0.945 |
| 28 | Franschhoek | 4.28 | 35 | 25 | 0.479 | 0.338 | 0.621 |
| 29 | Gansbaai | 12.14 | 106 | 53 | 0.684 | 0.594 | 0.774 |
| 30 | Genadendal | 2.35 | 10 | 8 | NA | NA | NA |
| 31 | George | 74.79 | 2721 | 376 | 0.949 | 0.941 | 0.957 |
| 32 | Grabouw | 9.92 | 73 | 53 | 0.425 | 0.298 | 0.553 |
| 33 | Greyton | 2.67 | 175 | 102 | 0.581 | 0.502 | 0.659 |
| 34 | Haarlem | 2.58 | 0 | 0 | NA | NA | NA |
| 35 | Hawston | 2.41 | 30 | 15 | NA | NA | NA |
| 36 | Heidelberg | 3.80 | 7 | 6 | NA | NA | NA |
| 37 | Hermanus | 23.32 | 1110 | 253 | 0.904 | 0.888 | 0.919 |
| 38 | Herolds Bay | 3.63 | 54 | 33 | 0.461 | 0.355 | 0.568 |
| 39 | Hoekwil | 3.07 | 97 | 59 | 0.529 | 0.409 | 0.650 |
| 40 | Hopefield | 5.86 | 10 | 8 | NA | NA | NA |
| 41 | Klapmuts | 6.81 | 11 | 10 | NA | NA | NA |
| 42 | Klawer | 2.54 | 4 | 4 | NA | NA | NA |
| 43 | Kleinmond | 5.16 | 763 | 215 | 0.837 | 0.819 | 0.855 |
| 44 | Knysna | 35.30 | 575 | 202 | 0.818 | 0.787 | 0.849 |
| 45 | Kranshoek | 8.62 | 88 | 43 | 0.635 | 0.519 | 0.750 |
| 46 | Ladismith | 3.98 | 18 | 16 | NA | NA | NA |
| 47 | Laingsburg | 2.77 | 2 | 2 | NA | NA | NA |
| 48 | Lambert’s Bay | 3.52 | 0 | 0 | NA | NA | NA |
| 49 | Langebaan | 19.68 | 54 | 39 | NA | NA | NA |
| 50 | Leeu Gamka | 1.98 | 3 | 2 | NA | NA | NA |
| 51 | Lutzville | 4.19 | 1 | 1 | NA | NA | NA |
| 52 | Malmesbury | 12.19 | 30 | 25 | 0.246 | 0.017 | 0.475 |
| 53 | Mamre | 2.15 | 6 | 6 | NA | NA | NA |
| 54 | Melkbosstrand | 8.17 | 68 | 33 | 0.661 | 0.537 | 0.785 |
| 55 | Montagu | 6.34 | 35 | 27 | 0.358 | 0.167 | 0.550 |
| 56 | Moorreesburg | 6.08 | 0 | 0 | NA | NA | NA |
| 57 | Mossel Bay | 39.68 | 1192 | 182 | 0.941 | 0.930 | 0.952 |
| 58 | Murraysburg | 2.05 | 1 | 1 | NA | NA | NA |
| 59 | Napier | 2.92 | 111 | 48 | 0.749 | 0.666 | 0.833 |
| 60 | Outshoorn | 23.63 | 64 | 37 | NA | NA | NA |
| 61 | Paarl & Wellington | 96.54 | 451 | 172 | 0.789 | 0.754 | 0.823 |
| 62 | Paternoster | 2.15 | 51 | 30 | 0.558 | 0.426 | 0.689 |
| 63 | Pearly Beach | 2.35 | 133 | 29 | 0.595 | 0.448 | 0.742 |
| 64 | Piketberg | 4.67 | 19 | 9 | 0.563 | 0.313 | 0.812 |
| 65 | Plettenberg Bay | 24.59 | 627 | 136 | 0.840 | 0.812 | 0.868 |
| 66 | Pniel & Kylemore | 4.01 | 151 | 60 | 0.660 | 0.581 | 0.739 |
| 67 | Porterville | 3.70 | 16 | 9 | 0.293 | 0.000 | 0.692 |
| 68 | Prince Albert | 4.33 | 19 | 13 | 0.457 | 0.180 | 0.734 |
| 69 | Prince Alfred Hamlet | 2.93 | 3 | 3 | NA | NA | NA |
| 70 | Pringle Bay | 2.93 | 127 | 62 | 0.673 | 0.591 | 0.734 |
| 71 | Riebeek West | 2.00 | 42 | 31 | 0.406 | 0.230 | 0.583 |
| 72 | Riebeek-Kasteel | 2.45 | 67 | 39 | 0.606 | 0.454 | 0.758 |
| 73 | Riversdale | 7.63 | 61 | 39 | 0.564 | 0.436 | 0.692 |
| 74 | Riviersonderend | 2.21 | 14 | 12 | NA | NA | NA |
| 75 | Robertson | 10.04 | 7 | 7 | NA | NA | NA |
| 76 | Saldanha Bay | 16.47 | 12 | 9 | NA | NA | NA |
| 77 | Saron | 2.52 | 0 | 0 | NA | NA | NA |
| 78 | Sedgefield | 6.66 | 153 | 73 | 0.718 | 0.656 | 0.779 |
| 79 | Slangrivier | 2.31 | 1 | 1 | NA | NA | NA |
| 80 | Somerset West | 78.96 | 3156 | 383 | 0.963 | 0.958 | 0.969 |
| 81 | St Helena Bay | 13.46 | 20 | 15 | NA | NA | NA |
| 82 | Stanford | 2.76 | 63 | 49 | NA | NA | NA |
| 83 | Stellenbosch | 33.57 | 13689 | 410 | 0.991 | 0.990 | 0.993 |
| 84 | Still Bay | 8.77 | 244 | 110 | 0.776 | 0.730 | 0.824 |
| 85 | Struisbaai & L’Agulhas | 7.23 | 28 | 16 | 0.689 | 0.526 | 0.852 |
| 86 | Swellendam | 10.48 | 165 | 86 | 0.689 | 0.631 | 0.747 |
| 87 | Touws River | 3.22 | 3 | 3 | NA | NA | NA |
| 88 | Tulbagh | 3.72 | 20 | 18 | NA | NA | NA |
| 89 | Uniondale | 3.09 | 24 | 16 | 0.463 | 0.210 | 0.716 |
| 90 | Vanrhynsdorp | 3.69 | 17 | 13 | 0.426 | 0.143 | 0.709 |
| 91 | Velddrif | 8.56 | 37 | 25 | 0.448 | 0.226 | 0.671 |
| 92 | Villiersdorp | 3.25 | 5 | 5 | NA | NA | NA |
| 93 | Vredenburg | 11.44 | 11 | 9 | NA | NA | NA |
| 94 | Vredendal | 10.85 | 6 | 6 | NA | NA | NA |
| 95 | Wilderness | 8.16 | 332 | 110 | 0.804 | 0.766 | 0.842 |
| 96 | Windmeul | 2.98 | 16 | 13 | NA | NA | NA |
| 97 | Witsand | 2.22 | 43 | 22 | 0.675 | 0.545 | 0.804 |
| 98 | Wolseley | 4.86 | 14 | 10 | 0.521 | 0.223 | 0.819 |
| 99 | Worcester | 34.95 | 45 | 32 | 0.422 | 0.262 | 0.582 |
| 100 | Yzerfontein | 3.46 | 46 | 29 | 0.527 | 0.379 | 0.674 |
Our study confirms the utility of iNaturalist data for compiling species inventories. The inventory generated from the workflow ultimately contributes to the iNaturalist platform by encouraging species identification and data curation which improves the data for future use (e.g., for monitoring and detection). The project derived from the workflow (https://www.inaturalist.org/projects/non-native-plants-of-western-cape-urban-areas) is freely accessible. As additional observations are posted in our region of interest, the project will automatically be updated. All data and R code used in this paper are freely available on Zenodo (https://zenodo.org/records/15210704) and GitHub (https://github.com/christiaan-g/sANNIC_workflow). The sample completeness estimation flags both areas that are adequately sampled and those that are under-sampled and can guide the allocation of resources for adaptive biodiversity monitoring (
Our workflow further quantifies some of the uncertainty inherent in the compilation of non-native species inventories. We used a sample completeness profile to assess the inventory completeness of urban centres (Fig.
Although our data were sourced from iNaturalist and our code is designed for this platform, the workflow can easily be adapted to incorporate data from other sources. For example, data can be sourced from other popular community science platforms such as Pl@ntNet (https://plantnet.org/en/) and eBird (https://ebird.org/home) or from traditional databases of herbarium and museum specimens. We recommend using the proposed workflow alongside other sources of data to improve the comprehensiveness of the inventory. The data from many of these sources are already available on GBIF. Each data source will have their own data quality requirements which should be carefully evaluated and used to guide appropriate adaptations of the workflow. We used the WCVP to extract native range information for each species because it is the most extensive database available for vascular plants. Other sources of information (e.g. IUCN native range shapefiles) could be used for other taxonomic groups such as birds and mammals. However, for many taxonomic groups, no compiled databases are available, and knowledge remains limited.
The proposed sANNIC workflow functions fundamentally differently from workflows such as DASCO (
The level of curation required for the application of the sANNIC workflow in different regions is likely to vary, given the differences in identification and observation efforts across regions (
We observed a highly uneven sampling effort across the urban centres of the Western Cape as most towns had insufficient records for a completeness assessment. Sampling effort is skewed toward larger urban centres closer to the coast, though there were some exceptions. This spatial skewness is typical for iNaturalist and other unstructured community science platforms as users are more likely to reside in these areas or visit them for recreational purposes. The skewness provides a challenge for species monitoring across the area of interest, as many areas remain under sampled. We accounted for the uneven sampling by using the prevalence index which considers the rank of a species at each urban centre where it was sampled, weighed by the Sample Coverage of that centre. Using this approach, we produced a list of the most prevalent species in urban centres of the Western Cape while minimising potential biases introduced by spatial heterogeneity. The prevalence index can also be adapted for use in a grid- or observer-based framework to account for spatial heterogeneity or potential observer bias. However, this index may still be biased by general observer preferences and/or the inherent recognisability of species (
The number of records in some areas could be increased by City Nature Challenges (
Although estimating sample completeness is essential for interpreting the data, several limitations need to be considered. Uneven spatial or seasonal sampling within urban centres can affect sample coverage and influence how accurately the number of species can be estimated. For example, if observations are strongly concentrated during a particular time of year, the sample coverage will be overestimated since only species visible during that time of year will be observed. Strong spatial autocorrelation of records may result in an overestimation of sample coverage and underestimation of species richness if species occurrence is highly heterogeneous. We encourage caution when using sample completeness measures, especially in areas with few records and strongly encourage the reporting of uncertainty of sample completeness measures using bootstrapping, as is available in the ‘iNEXT’ packages (
The list of species generated in this case study has many potential uses and implications for the regional understanding of plant invasions in urban areas of the Western Cape of South Africa. Many taxa in our list are among the most common urban plant invaders globally (
The best represented families, (Fabaceae, Asteraceae, and Poaceae) in our case study are not surprising; these are generally the best represented plant families in non-native floras globally (Suppl. material
We believe our workflow provides the foundation for further research into the non-native species of any region with sufficient iNaturalist coverage. If an inventory with a high coverage for a particular area is available, research can focus on clarifying the introduction status of each species, introduction pathways, possible impacts, and patterns of establishment and invasion within the region of interest. We highlight the necessity of thorough curation of the records directly on iNaturalist and believe that our workflow could be used alongside other sources of data such as systematic surveys to produce high quality inventories of non-native species.
This research was supported by the Centre for Invasion Biology and the Department of Botany and Zoology, Stellenbosch University. We are deeply grateful to everyone who contributed observations and identifications to the iNaturalist project. These include but are not limited to Clare Archer (@sedgesrock), Brian du Preez (@mr_fab), Sandra Falanga (@sandraf), Bianke Fouche (@biankefouche), Jeremy Gilmore (@jeremygilmore), David Hoare (@dhoare), Suzaan Kritzinger-Klopper (@zaniekk), Caroline Mashau (@carolinemashau), Phil McLean (@fynbosphil), Erick Munro (@erickmunro), Tony Rebelo (@tonyrebelo), Norbert Sauberer (@brothernorbert), Charles Stirton (@charles_stirton), Diana Studer (@dianastuder), Hedi Stummer (@venturefoth), Shaun Swanepoel (@shauns), Pieter Winter (@pieterwinter), Mark van Dalsen (@vandalsen), @linkie, and many more. A full list of identifiers and observers that contributed to the project can be found at https://www.inaturalist.org/projects/non-native-plants-of-western-cape-urban-areas.
The authors have declared that no competing interests exist.
No ethical statement was reported.
No use of AI was reported.
CPG is supported by a postgraduate scholarship from the National Research Foundation (PMDS230601112319). LJP acknowledges support from the Centre for Invasion Biology, Department of Botany and Zoology, Stellenbosch University, and from the Natural Sciences and Engineering Research Council of Canada (grant 386151). CH is supported by the NRF (grant 89967) and the European Union’s Horizon Europe Research and Innovation Programme (B3 – Biodiversity Building Blocks for policy, ID 101059592). Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Commission. Neither the EU nor the EC can be held responsible for them.
Data cleaning was performed by DMR, CPG, and LJP. Supervision was performed by DMR, CH, and LJP. Writing of associated software was performed by CPG. All authors contributed to the conceptualisation and writing of the manuscript.
Christiaan P. Gildenhuys https://orcid.org/0009-0006-5226-3250
Luke J. Potgieter https://orcid.org/0000-0001-7790-2721
Cang Hui https://orcid.org/0000-0002-3660-8160
David M. Richardson https://orcid.org/0000-0001-9574-8297
All code and data used in this study are available in Zenodo (10.5281/zenodo.15210704) and the latest versions will be made available in GitHub (https://github.com/christiaan-g/sANNIC_workflow). All additional figures referred to are available in the Suppl. material
Defining Western Cape urban boundaries
The delineation of urban boundaries is challenging and context/region specific (
Boundaries of the Western Cape urban centres were delineated using the latest urban land cover dataset (generated in 2020) for the Western Cape from the South African National Land Cover database (available at: https://egis.environment.gov.za/sa_national_land_cover_datasets). Using the “built-up” feature class from this dataset, a 500 m buffer was drawn around all continuous built-up areas covering at least five hectares (5 000 m2). This was dissolved, i.e. all overlapping buffered areas were merged, before a negative 500 m buffer was applied. This process effectively removed internal boundaries between adjacent built-up areas in close proximity, incorporating intra-urban non-built-up areas such as parks and road verges, and smoothing the outer boundaries while preserving the general shape of the urban centre in alignment with the original “built-up” feature class outline.
Any erroneous “towns” including power infrastructure, military bases, and airports were removed. Though these could be considered urban centres they are unlikely to have good coverage of iNaturalist records (due to restricted access) and were not considered in this study. The remaining urban centres were ranked by size, and the 100 largest were retained. Urban centres ranged from ~ 821 km2 (Cape Town) to 1.7 km2 (Bot River) in extent. Smaller areas below our cut-off would have been difficult to differentiate from large agricultural infrastructure and were unlikely to have sufficient iNaturalist coverage. Fig.
The shapefiles of the urban centres were grouped according to local districts and uploaded to iNaturalist as “places”. Only the City of Cape Town Metropolitan Municipality was handled differently to accommodate restrictions on place uploads imposed by iNaturalist. Because of the large number of records in this area, it was separated into smaller areas before uploading. All manipulations were performed in QGIS version 3.28.3 (
Plant taxa that could not be identified to species level in iNaturalist records for the Western Cape, South Africa. The potential/suggested species names are the taxa most often suggested (listed in the order of the most commonly suggested) by iNaturalist users (and known to be present in the region), ranked according to their frequency in the dataset (https://www.inaturalist.org/projects/naturalising-alien-plants-of-western-cape-urban-areas [accessed: 3 November 2024]).
| Taxon | Rank | Reason | Potential/suggested species names |
|---|---|---|---|
| Alocasia | Genus | Unclear distinction between various species | A. macrorrhizos, A. brisbanensis, A. Odora |
| Amaranthus | Genus | Unclear distinction between many species (there are several native species, but most taxa in urban centres are most likely non-native) | A. deflexus, A. curentus, A. hybridus, A. viridis, A. blitum |
| Avena | Genus | Species impossible to distinguish from iNaturalist images (details of spikelet morphology) | A. fatua, A. barbata, A. sativa, A. Sterilis |
| Bambusoideae | Subfamily | Many taxa; difficult to distinguish without flowers | Phyllostachys aurea, Bambusa vulgaris, B. balcooa |
| Brugmansia | Genus | Difficult to distinguish species from typical iNaturalist images | B. suaveolens, B. arborea, B. insignis, B. × candida |
| Eucalyptus | Genus | Species difficult/impossible to identify at juvenile stage or without flowers/fruit | E. conferruminata, E. camaldulensis, E. cladocalyx, E. botryoides, E. gomphocephala, E. cinerea, E. grandis, E. globulus, E. polyanthemos, E. microcorys, E. melliodora |
| Fraxinus | Genus | Unclear distinction between species | F. pennsylvanica, F. excelsior, F. americana, F. angustifolia |
| Gamochaeta | Genus | Unclear species distinctions | G. americana, G. pensylvanica |
| Hedera | Genus | Unclear distinction between H. helix and H. canariensis without juvenile foliage | H. helix, H. canariensis |
| Narcissus | Genus | Possible hybridisation and several cultivars | N. tazetta, N. papyraceus |
| Neltuma | Genus | Difficult to distinguish | N. glandulosa, N. juliflora, N. odorata, N. velutina |
| Oenothera | Section | Indistinguishable species and hybridisation | O. drummondii, O. stricta, O. indecora, O. biennis, O. laciniata, O. affinis, etc. |
| Pinus | Genus | Difficult to identify at juvenile stage and from image quality typically available on iNaturalist | P. radiata, P. pinaster, P. pinea, P. halepensis, P. canariensis |
| Sagina | Genus | Unclear species distinctions | S. apetala, S. micropetala, S. Procumbens |
| Salsola | Genus | Difficult to distinguish | S. kali, S. tragus, S. australis |
| Sambucus | Genus | Two species difficult to distinguish | S. nigra, S. canadensis |
| Vitis | Genus | Most likely V. vinifera but other difficult to distinguish species possibly present | V. vinifera, (V. labrusca) |
The most abundant families in the inventory
Data type: docx