Bird occurrence records for Mexico were obtained from the Global Biodiversity Information Facility (GBIF), including both scientific collections and citizen science observations classified as “present.” We retained multiple record types (specimens and observations) and applied geographic filtering, allowing the full temporal extent of available data. Taxonomy was harmonized following HBW and BirdLife International standards (2022), using accepted scientific names and resolving discrepancies through automated and manual validation. To focus on native biodiversity patterns, invasive and domestic species were excluded. Duplicate records were removed to ensure data integrity. Full details on data filtering, taxonomic validation, and exclusion criteria are provided in Appendix A.

Geographic validation was conducted using the CoordinateCleaner package (Zizka et al., 2019) to identify and remove records with potentially erroneous or imprecise coordinates. We excluded occurrences associated with common spatial errors (e.g., zero coordinates, political centroids, capitals, institutions, and spatial outliers), as well as records with coordinate uncertainty greater than 100 km. To balance data quality and temporal coverage, we retained records collected between 1900 and 2024. In a second step, occurrences falling outside species’ known distribution ranges, based on BirdLife International maps with an additional 0.5 ° buffer, were excluded as likely spatial or taxonomic outliers. This conservative filtering prioritizes spatial reliability while preserving broad historical coverage. Detailed procedures are provided in Appendix A.

Citizen science documented 1,068 species from 95 families, while scientific collections recorded 1,015 species from 92 families. Several families (e.g., Muscicapidae, Oceanitidae, Phylloscopidae) were represented exclusively in citizen science data after geographic validation against IUCN range maps.

Most species in both datasets were classified as Least Concern by the IUCN, although citizen science included 10 additional Near Threatened species (Table 1). Detection ratios indicated that most shared species were better represented in citizen science (Appendix A, Fig. S2). However, 42 species exhibited detection ratios >1, indicating substantially greater representation in scientific collections, likely reflecting rarity, restricted distributions or low detectability by amateur observers.

Citizen science recorded 69 species absent from scientific collections, although only 4% had more than 100 records; Laterallus exilis was the most frequently recorded among them (1,846 records; Appendix B, Table S1). Conversely, 16 species were exclusive to scientific collections, with only 3% exceeding 20 records; Turdus confinis was the most represented (Appendix B, Table S2). Threatened species were unevenly represented. Scientific collections included six threatened species (three Endangered, two Vulnerable, one Near Threatened), all with fewer than five records. Citizen science documented 19 threatened species, including one Critically Endangered (Gymnogyps californianus) and six Vulnerable, although some (e.g., Ardenna bulleri) were represented by single observations.

The ten most frequently recorded species differed markedly between datasets (Appendix B, Tables S3–S4). Haemorhous mexicanus dominated scientific collections (4,467 records), whereas Quiscalus mexicanus (334,983 records) overwhelmingly dominated citizen science observations.

Sampling intensity in scientific collections was most strongly associated with proximity to roads (0.039), cities (0.043), and protected areas (0.010), whereas rivers had minimal influence (Fig. 1a). Citizen science was more strongly influenced by cities (0.062) and roads (0.052), with weaker effects of protected areas and rivers (Fig. 1b). Despite differences in the relative strength of individual predictors, geographic sampling patterns were highly similar between data sources (Spearman’s ρ = 0.98).

Fig. 1.

Sampling bias in bird data: (a) scientific collections, and (b) citizen science observations. Error bars represent 95% credible intervals from the Bayesian model. The middle panel shows the sampling rate as a function of distance to the nearest instance of each bias factor based on the inferred model for scientific collections (c) and citizen science observations (d). (f) Spatial representation of the geographic bias for scientific collections and citizen science. Lighter colors represent higher sampling bias for roads, cities, natural protected areas and rivers. Maps are at a spatial resolution of 0.0083 arc-seconds.

Cumulative accessibility effects revealed broadly similar spatial bias patterns, with stronger biases (values closer to zero) concentrated in Central Mexico. However, the relative influence of individual predictors differed, indicating partially distinct accessibility dynamics. For example, scientific collections clustered along highways in Guerrero and around Mérida in Yucatán, while citizen science provided broader spatial coverage in these regions. In highly biodiverse states such as Oaxaca and Chiapas, both datasets showed limited coverage outside accessible areas. Northern states (e.g., Coahuila, Chihuahua, Sonora, Durango) and southern Campeche remained largely under sampled in both datasets (Fig. 1f).

Citizen science provided more recent and spatially extensive coverage of bird richness, whereas scientific collections exhibited pronounced spatial and temporal gaps, particularly in Northern and Southwestern Mexico. Species richness peaked along mountain ranges, reaching maxima of 388 species in citizen science and 337 species in scientific collections. Despite high record numbers, survey completeness could not be estimated for all spatial units (Fig. 2).

Fig. 2.

Survey completeness in scientific collections (left panel) and citizen science (right panel). Darker colors indicate higher documented species richness (top row) and greater inventory completeness (bottom row). The top row maps show spatial units where species richness could be measured. The bottom row shows inventory completeness, with polygons outlining well-sampled grid cells (completeness ≥80%). Bar charts within each map summarize the proportion of grid cells by species richness and completeness categories for each dataset.

In scientific collections, 33.7% of spatial units were covered with a median survey completeness of 76.9% (± 8.9). Only these, 35.5% were well-sampled, 63.8% moderately sampled, and 0.7% poorly sampled. These areas were patchily distributed, often aligning with mountain ranges and urban centers, including urban Yucatán. Citizen science covered 55% of spatial units with a higher median completeness of 88.1% (± 9.4). Of these, 71% were well-sampled, 28.6% moderately sampled, and 0.1% poorly sampled. Well-surveyed spatial units extended across Baja California, the Pacific coast, and Central Mexico, with additional coverage near the west coast and Yucatán (Fig. 2).

Scientific collections provided more historical depth but declined after 1960 (Fig. 3S), with a median record age of 65 years (± 23.6) (Fig. 3). In 71.9% of spatial units, species with the most recent records had not been recorded for over 50 years. Recent records (<50 years) were concentrated in Central Mexico. In contrast, citizen science data had a much younger median record age of 3 years (± 9.5), with 95.4% of grid cells containing data from 2004 or later (Fig. 3). Isolated cells with unusually old citizen science records were detected in eastern mountain ranges, coastal zones, and Gulf of California islands (Fig. S5).

Fig. 3.

Temporal bias for scientific collections and citizen science. Darker colors indicate younger inventories. Inset histograms show the distribution of the proportion of grid cells within each bin, with the dotted line marking the median value. Grid cell resolution: 15 arc-minutes.

Combining completeness and collection recency revealed that 50% of cells were well-sampled by citizen science but associated with outdated scientific records, identifying high-priority areas for updating institutional collections. An additional 24% were well-sampled with recent scientific records, while 19% showed moderate completeness and outdated collections, suggesting areas where representativeness could be improved. The remaining 7% included cells with either moderate or low completeness but recent scientific records.

We combined scientific collection completeness with a binarized map of record recency. As expected, many of the resulting cells overlapped with those identified as well-sampled by citizen science, but this comparison also highlights additional spatial units where citizen science lacks sufficient data to calculate completeness (Fig. 4). In this analysis, 44% of cells had medium completeness and outdated records, 24% had high completeness with recent records, and 12% had high completeness but with outdated records. This last group of cells are high-priority targets for resampling field campaigns, as it includes areas with solid baseline data that have not been updated in decades.

Fig. 4.

Integration of survey completeness and temporal recency patterns. Temporal recency is derived from scientific collections. Grid cells are classified into six categories based on the combination of survey completeness (well-, moderately-, or poorly-sampled) and record recency (recent vs. outdated). Categories correspond to: (1) well-sampled with outdated records, (2) well-sampled with recent records, (3) moderately-sampled with outdated records, (4) moderately-sampled with recent records, (5) poorly-sampled with outdated records, and (6) poorly-sampled with recent records. Hashed and outlined cells represent scientific collection completeness × scientific collection recency, whereas colored cells represent citizen science completeness integrated with scientific collection recency, allowing direct comparison between data sources. Grid cell resolution is 15 arc-minutes.

Finally, 11% of grid cells were well-sampled and up-to-date in both datasets, making these ideal candidates for establishing long-term bird monitoring programs. These cells are scattered across the country, with notable concentrations in Central and Southern Mexico, as well as on some Pacific islands (Fig. S6).

Neither dataset includes all 1,094 bird species reported for Mexico by BirdLife International (2024). Citizen science lacked ∼30 species, and scientific collections lacked geographically useful records for 79 species. These absences likely reflect data limitations—such as filtering criteria, incomplete digitization or lack of open-access sharing—rather than true species absence (Hortal et al., 2015; Meyer et al., 2016). Citizen science data disproportionately represent species that are conspicuous, large-bodied, common, and/or of least concern (Callaghan et al., 2021). This bias reduces their utility for comprehensive biodiversity assessments, especially for rare or cryptic species that are harder for amateur observers to detect or identify (Daru and Rodriguez, 2023; Fontaine et al., 2022). Additionally, the absence of voucher specimens limits opportunities for taxonomic verification and reduces their value for long-term research or confirmation of key ecological patterns (Bonney et al., 2014; Lang et al., 2019).

Regarding geographic biases, both datasets show strong spatial clustering around cities and roads, consistent with previous findings (Boakes et al., 2010; Callaghan et al., 2021). Citizen science efforts generally rely on accessible areas with adequate infrastructure to facilitate observer participation, including transportation, safety, and communication networks (Echeverri et al., 2022; Ocampo-Peñuela et al., 2025). Thus, citizen science records tend to concentrate near urban and recreational areas, reflecting the influence of accessibility and leisure-related motivations (Geldmann et al., 2016; Moerman and Estabrook, 2006). In contrast, scientific collections exhibit relatively greater representation in remote locations, particularly within protected areas and along riparian corridors, where professional collectors often work under institutional mandates.

Riparian zones appear underrepresented in citizen science data, likely due to limited physical access, safety concerns or lack of recreational appeal. These areas may also be bypassed when observers focus on more visible or charismatic species typically found in open or urban habitats (Echeverri et al., 2022; Ocampo-Peñuela et al., 2025). Therefore, initiatives such as birdwatching tourism could be designed to promote the exploration of riparian ecosystems and help bridge this gap and improve spatial coverage (Ocampo-Peñuela et al., 2025).

Despite differences, both datasets are heavily concentrated in Central Mexico, while northern regions and biodiverse southern states like Oaxaca and Chiapas remain under-sampled, likely due to accessibility challenges (Boakes et al., 2010; Geldmann et al., 2016). As a result, neither dataset fully compensates for the other in these key but poorly documented regions (Daru and Rodriguez, 2023; Geldmann et al., 2016).

Scientific collections offer long-term historical records (median record age: 65 years), but 71.9% of grid cells include records over 50 years old. While this limits their ability to reflect present-day biodiversity patterns (Lang et al., 2019), the issue stems more from chronic underfunding and reduced institutional capacity than from the collections themselves (Ladle and Hortal, 2013). In contrast, citizen science contributes recent data (median record age: 3 years), addressing the temporal shortfall and enabling near real-time biodiversity monitoring (Boakes et al., 2010). Integrating both datasets is therefore crucial for assessing biodiversity dynamics over time. For instance, combining historical specimens with recent observations has proven valuable for tracking environmental change in Mexican endemic birds (Peterson et al., 2015).

Citizen science can significantly contribute to biodiversity research by providing recent, large-scale observations that complement formal inventories (Callaghan et al., 2019; Lau et al., 2019). Although both citizen and institutional efforts are typically clustered within accessible regions, citizen observations can help fill geographic and taxonomic gaps when aligned with research and conservation objectives (Wetzel et al., 2018). The main complementarity of both datasets lies in the temporal dimension: citizen science supplies recent records with broad participation, while scientific collections provide historical depth and taxonomic rigor. Continued institutional fieldwork remains essential for building complete and accurate biodiversity inventories, especially in under sampled areas.

While citizen science has not yet significantly expanded coverage into remote regions, it holds strong potential if guided by targeted protocols, environmental education, and institutional coordination (Callaghan et al., 2019; Rowley et al., 2019; Fontaine et al., 2022). In this study, we provide a spatial framework to help prioritize areas for bird sampling—whether to document local diversity, update inventories or launch new monitoring efforts. Species-focused citizen science projects, especially when paired with educational outreach, can boost participant skills and promote broader engagement in biodiversity conservation (Fontaine et al., 2022; Lau et al., 2019; Steven et al., 2019).

We followed the HBW taxonomy (BirdLife International, 2020) that aligns with IUCN distribution polygons. Although other global taxonomies such as eBird (Clements et al., 2021) and BirdTree (Jetz et al., 2012) differ, these deviations remain within ∼20% and generally yield comparable ecological outputs (Tobias et al., 2022).

Regional taxonomies introduce additional inconsistencies. Many Mexican birdwatchers rely on the American Ornithologists’ Society (AOS) taxonomy (Chesser et al., 2024), which tends to be more conservative in species splits compared to HBW. For example, HBW treats Turdus confinis (San Lucas Robin) as distinct from T. migratorius, whereas AOS merges them. This taxonomic discrepancy likely contributes to the underreporting of T. confinis in citizen science datasets from Baja California. Moreover, species appearing only in citizen science are sometimes absent from scientific collections due to differences in taxonomic standards. Using more conservative systems may lump endemic taxa under broader names, reducing their visibility in integrated databases. These inconsistencies highlight the importance of harmonizing taxonomy across datasets to ensure the inclusion of endemic and range-restricted species.