iNaturalist Users Exhibit Distinct Spatiotemporal Sampling Preferences, with Implications for Biodiversity Science and Project Planning

Downloading and filtering observation data

Using custom code derived from the R package rinat (Barve and Hart 2022), we downloaded all verifiable iNaturalist observations (observed by December 31st, 2023 and uploaded by October 12th, 2024) in the focal region and removed those with obscured coordinates (Figure 1a). We focused on the southeastern USA, a biodiversity hotspot (Noss et al. 2015) that draws over 140 million tourists annually (Visit Florida Research 2025), contains areas with some of the nation’s highest poverty rates (Baker 2020), and has extensive private land ownership (Butler and Wear 2013). Core analyses were conducted in R version 4.2.3 (R Core Team 2023), and we used the following R packages: tidyverse (Wickham et al. 2019), sf (Pebesma and Bivand 2023), terra (Hijmans 2024), nnet (Venables and Ripley 2002), and data.table (Barrett et al. 2025).

Figure 1

Associating observations with land context

We associated each observation from the focal region with land context categories: land cover, protected areas, urban areas, urban parks, and low-income urban neighborhoods. We established land cover categories following methods described in Callaghan et al. (2019), aggregating land cover classes from the National Land Cover Database (USGS 2023) into urban green area, natural green area, agriculture, open-urban, low-intensity developed, and medium/high-intensity developed. Urban green area and natural green area were differentiated using the 2020 TIGER/Line Urban Areas boundaries (USCB 2020).

We designated land protection status using the “Fee” feature class of the Protected Areas Database (USGS 2022). Urban areas were designated using 2020 TIGER/Line Urban Areas boundaries (USCB 2020). We used the ParkServe dataset to designate which urban observations were made in parks versus neighborhoods (TPL 2024). Finally, we used low-income census tract polygons from the American Community Survey (USCB 2022) to specify urban observations made in low-income areas.

General sampling biases in the focal region

To quantify overall spatial sampling biases, we first defined the proportion of observations expected in each land class based on area. We calculated the proportion of land within the focal region (for land cover, protected areas, and urban areas) and within focal urban areas (for urban parks and low-income neighborhoods) that met each land context categorization. We then calculated the actual proportion of observations made within each land class and compared these values with the expected proportions.

To quantify weekend versus weekday bias, we calculated the expected proportion of observations on weekends versus weekdays and determined whether each observation in the focal region was made on a weekend or weekday using the wday() function in the R package lubridate (Grolemund and Wickham 2011). We calculated the proportion of observations made on weekends versus weekdays and compared these values with the expected proportions. Weekend versus weekday bias was included here because it can strongly impact phenological analyses, which are often conducted using contributory science data (Courter et al. 2013).

Assigning user categories

We compiled a list of users whose observations were included in the primary dataset. To assemble a complete dataset of global observations made by these users, we combined previously downloaded observations from Campbell et al. (2023) with additional data directly downloaded from iNaturalist, again using custom code based on the R package rinat (Barve and Hart 2022). We then counted each user’s total number of verifiable global observations as of October 12, 2024 and assigned each user a percentile relative to others. The top 5% of observers by number of observations were classified as highly active and all others were classified as casual. We determined traveler status by calculating the proportion of each user’s observations made in the southeastern USA relative to their global observation count. We classified users who had made half or more observations in the focal region as locals and those who made less than half as travelers.

This method relies on the assumption that observers make the majority of their observations in the immediate area in which they reside. Although there may be edge-cases in which observers live near the study region boundary or moved their primary residence during the study period, the large number of active observers included (283,392) ensures that such cases are unlikely to influence overall patterns. Misclassifications would be most likely to reduce the observed differences between groups, making our results somewhat conservative.

Assessing how participation level and traveler status shape sampling

iNaturalist records

We downloaded a total of 8,897,124 observations. After removing records with obscured coordinates, 7,533,946 remained in the focal region, contributed by 283,392 unique users. We downloaded and compiled 58,435,335 observations made globally by these users prior to October 12, 2024.

General sampling biases in the focal region

We found that observations in five land cover classes were overrepresented in the focal region: low-intensity developed areas (20.7% observed, 4.6% expected), medium/high-intensity developed areas (16.4% observed, 2.3% expected), open urban areas (16.3% observed, 7.1% expected), and urban green spaces (9.2% observed, 2.2% expected) (Supplemental file 2: Table 2). Agricultural areas (5.5% observed, 20.2% expected) and natural green areas (31.9% observed, 63.6% expected) were underrepresented.

Protected land (39.4% observed, 14.4% expected) and urban areas (46.8% observed, 8.9% expected) were overrepresented, while unprotected land (60.6% observed, 85.6% expected) and rural areas (53.2% observed, 91.1% expected) were underrepresented. Within urban areas, urban parks (19.3% observed, 2.5% expected) were overrepresented, while urban neighborhood areas (80.7% observed, 97.5% expected) were underrepresented. Higher-income census tracts (68.3% observed, 67.3% expected) within urban areas were slightly overrepresented, and low-income census tracts (31.7% observed, 32.7% expected) slightly underrepresented. Weekends (35% observed, 28.6% expected) were overrepresented and weekdays (65% observed, 71.4% expected) underrepresented.

Assigning user categories

Of the 283,392 users, we defined the most active 5% as highly active and the remaining 95% as casual based on their number of global observations (Figure 1c). Highly active users made an average of 3,419 total observations globally (median: 1,223, min: 494, max: 276,674, sd: 7,855.2), while casual users made an average of 37 total observations globally (median: 9, min: 1, max: 493, sd: 73.5). Highly active users were responsible for 59.7% of observations in the focal region and casual users 40.3% (Figure 1b). 61.6% of users were categorized as locals and 38.4% as travelers. Locals made 62.9% of all observations in the focal region and travelers made 37.1%. At the intersection of participation level and traveler status, 61.1% users were classified as casual locals, 33.9% as casual travelers, 0.5% as highly active locals, and 4.5% as highly active travelers. Highly active locals made the most observations (31.9%), followed by casual locals (31%), highly active travelers (27.8%) and casual travelers (9.3%) (Figure 1c).

Assessing how participation level and traveler status shape sampling

Quality grade

The best-supported model of observation quality grade included participation group, traveler status, and their interaction as predictors (Supplemental file 3: Table 4). We found that highly active users were more likely to make observations that reach research grade, and casual users were more likely to make observations that remain as needs ID or are designated as casual grade (Figure 2; Supplemental file 3: Table 5). Travelers were more likely to make research-grade observations than residents, and less likely to make casual or needs ID observations.

Figure 2

Land cover

The best-supported land cover model contained participation group, traveler status, and their interaction as predictors (Supplemental file 3: Table 4, 7). Highly active users, regardless of traveler status, were less likely to sample in developed landscapes and more likely to sample in natural green areas compared with casual users (Figure 3a). Travelers, regardless of participation level, were more likely to sample in natural green areas and less likely to sample in areas with low development compared with locals. In developed areas, highly active travelers observed more than expected based on their participation and traveler status group effects alone. The category with the most variation in observation behavior was natural green areas, where highly active travelers observed the most and casual locals the least. With increasing human development, the difference between casual and highly active users generally widened, with these groups making similar proportions of observations in urban green and open urban areas, but casual users making higher proportions in developed areas (Figure 3a).

Figure 3

Protected land

The best-supported protected land model included participation group, traveler status, and their interaction as predictors (Supplemental file 3: Table 4). Highly active users had 2.24 times higher odds of observing on protected land than casual users (p < 0.0001), while travelers had 1.29 times higher odds compared with locals (p < 0.0001; Figure 3b; Supplemental file 3: Table 7). Highly active travelers were slightly less likely to observe in protected areas than expected from the combined effects of participation level and traveler status (p < 0.0001; Supplemental file 3: Table 7). All four user groups observed more than expected on protected land. Highly active travelers were the most biased toward protected land, observing 36.2% more than expected, followed by highly active locals (+30.3%), casual travelers (+17.4%), and casual locals (+11.7%) (Figure 4).

Figure 4

Highly active users seek out biodiversity-rich areas, while casual users primarily incorporate sampling into daily life

We found that casual and highly active users have different observational habits and likelihoods of making observations that reach research grade. In particular, highly active observers generally go out of their way to sample in natural and protected areas, both within and outside cities, and are more likely to add observations on weekends than casual users. Such areas are likely to harbor higher, unique diversity given that developed areas tend to act as strong filters for such diversity (Aronson et al. 2014). To be clear, highly active users also contribute numerous records in developed areas and during the week, but what sets them apart from casual users is their high proportion of observations in biodiverse, natural areas.

The preference of highly active observers toward these natural locations, presumably often during leisure time, aligns with their increased motivation to contribute to conservation research (Larson et al. 2020; Lowe et al. 2025), and with previous work focused on other contributory science initiatives. For example, Rosenblatt et al. (2022) found that specialist birders, who have stronger skills and devote more time to birding, travel farther from home to observe. Similarly, Bowler et al. (2022) observed that participants with more experience were more likely to conduct planned searches, and that these searches were more likely to occur in natural and protected areas than urban areas. Our findings, in combination with these works, suggest highly active users make intentional and focused efforts to collect biodiversity data.

Highly active users fill critical gaps in areas broadly under-sampled on iNaturalist, such as rural areas and natural green spaces, but also perpetuate biases toward protected land and urban parks. Still, observations from these biodiverse areas are particularly valuable for research on rare or threatened species, range shifts, ecosystem response to environmental change, and urban-natural comparisons. Consistent with previous work (Dimson and Gillespie 2023), we found observations made by highly active users were more likely to reach research grade than those by casual users. This may be because highly active users include more detailed images or metadata with their observations, or might initially provide more specific identifications (e.g., genus rather than family or order). As the primary source of research-grade records, highly active users play a key role in generating high-quality biodiversity data for research. However, these users also tend to neglect non-park urban areas and unprotected rural lands, which are most likely privately owned. These areas may host lower overall biodiversity, reducing their sampling potential and attractiveness to observers, but they also might simply be places highly active users can’t access or don’t frequent.

Our results suggest that observers who use iNaturalist more casually tend to primarily incorporate sampling into their daily lives. These casual users sample predominantly in urban areas, strongly perpetuating biases toward developed land that are well-documented in community-collected biodiversity data (Di Cecco et al. 2021). These findings align with those in Dimson and Gillespie (2023), who found that short-term observers who were residents of Hawaii were the least likely to sample outside developed areas. This is likely because casual users sample primarily close to home or during their routines and are less likely to go on nature-specific excursions, a conclusion further strengthened with the result that these users, especially locals, are less biased toward weekends than highly active users, and are more likely to make casual observations that reflect cultivated organisms. We note, however, the lack of direct information about where observers reside, and that these conclusions are inferences based on patterns in their sampling behavior.

Although casual users perpetuate data biases toward developed areas, they critically fill gaps in areas less served by highly active users. Within cities, casual users sample proportionally more than highly active users in non-park neighborhood areas. As observations in urban parks are generally overrepresented on iNaturalist, these non-park observations are particularly valuable for sampling broadly across cities and can help researchers more effectively understand how biodiversity responds across the broadest urban gradients (Callaghan et al. 2020). Some of these non-park observations are cultivated plants on private property based on a spot-check of records to determine cultivation status of casual-grade records. While these may be considered less valuable, records of cultivated species can offer valuable insight into the interaction between cultivated and natural biodiversity, the resiliency of urban green areas to environmental change, and environmental equity across cities (Leong et al. 2018). Still, the primary goal of iNaturalist is to collect biodiversity information on wild organisms, and cultivated observations can sometimes pose challenges for ecological analyses, particularly when the “cultivated” tag is not properly selected (López-Guillén et al. 2024).

In addition to filling data gaps within urban environments, casual users fill gaps in non-protected rural areas. In the southeastern USA, the majority of forested land is privately owned (Butler and Wear 2013) and heavily underrepresented on iNaturalist. Contributory biodiversity monitoring is especially critical in these areas, as they are generally not covered by traditional ecological monitoring. Casual users, perhaps because they represent a broader pool of participants who live in, know about, or have access to these areas, are helping fill these data gaps.

Locals fill gaps in low-income urban areas regardless of participation group

Regardless of participation group, local iNaturalist users were more likely to observe in low-income urban areas than travelers. Casual locals observed in these areas slightly more than expected by land area, representing the only case in our analyses in which a user group actively sampled against the overall data bias (Figure 4).

It is well-documented that contributory science platforms are impacted by socioeconomics, with historically redlined and lower-income areas underrepresented and yielding a less complete picture of biodiversity (Ellis-Soto et al. 2023; Estien et al. 2024). We did not find large differences in representation across income classes, likely in part due to the broad classifications of low-income versus higher-income areas in our analyses, but more importantly because iNaturalist users local to the southeastern USA helped mitigate biases toward higher income-areas. Filling gaps in underrepresented areas not only provides valuable data for research but also strengthens the capacity of communities in these areas to advocate for better environmental protections (Ellis-Soto et al. 2023). Here, we focused broadly on locals across the southeastern USA, but our work provides preliminary evidence that better engaging residents may be key to improving data availability and knowledge. Additionally, while this was not our focus here, facilitating participation from locals might reduce not only spatial but also taxonomic biases because locals, particularly those who are highly active, may be more attuned to regional biodiversity (Dimson and Gillespie 2023).

Implications for contributory project planning

Our results emphasize that a broader pool of contributors, varying in location, preferences, motivations, and abilities, enhances the scope of data collected and supports iNaturalist’s mission to “connect people worldwide with nature while advancing biodiversity science” (iNaturalist 2025). While highly active users are often prioritized in contributory science projects, our findings highlight the importance of recruiting new participants and residents of under-sampled areas, regardless of whether they will eventually become highly active. Strategies focused solely on cultivating highly active users may risk overlooking other contributors who fill critical data gaps.

Our work offers insights into which user groups are likely, willing, or able to observe in certain environments, and can be used to target specific audiences depending on project goals. Previous research has demonstrated that contributory project participants can be influenced to observe in under-sampled areas or areas of conservation concern thorough behavioral nudging, an intervention with the goal of behavioral changes (Callaghan et al. 2023; Thompson et al. 2023). Our results can help augment these efforts by implementing targeted nudges based on behavioral attributes of participants (Hart et al. 2022). For example, casual users may be more likely to respond to nudges asking them to sample near their neighborhood, while highly active users may be more willing or able to travel to rural areas of conservation concern. It is unclear whether the effectiveness of behavioral nudging varies by user group, and this presents a valuable question to explore in future work.

Implications for the field of citizen science

Our results illustrate how considering multiple behavioral axes and their relation to each other can disentangle participation patterns when assessing how volunteers contribute to citizen science. We found differences in sampling patterns across both the participation level and traveler status behavioral axes, generally supporting the notion that motivations of participants are complex and multifaceted (Dowthwaite et al. 2025; Robinson et al. 2021; West et al. 2021). As mentioned above, because different groups have different motivations and contextual conditions, project design needs to consider how and when to appeal to the different groups (Bowser et al. 2014; Lee et al. 2018). As an example, different maps could be provided to the different user types, highlighting specific areas to sample (Skarlatidou et al. 2024). This user-focused design concept is relevant not only to biodiversity-focused contributory science programs, but also citizen science programs more broadly (Fogg-Rogers et al. 2024). Understanding how different user groups contribute, and what motivates or enables their participation, provides a pathway to refine both analyses and project design, forming an important component of future research. When conducting these multi-axes analyses, axes may vary based on the size, complexity, goals, and subject matter of the citizen science project, participant information available to the researcher, or specific hypotheses or questions held by the researcher or project organizers.

Implications for biodiversity research

Understanding that different kinds of observers are more or less likely to observe in various environments and at different times is key to understanding the processes which underlie opportunistic spatial and temporal data biases. Our work suggests that spatial and temporal biases(e.g., weekends versus weekdays) may be intimately linked, which is valuable to consider when accounting for oversampling in phenology analyses, an area of research rapidly expanding using opportunistic data (Li et al. 2021). Previous work has shown that different biodiversity trend analyses vary in robustness to variation in user sampling behavior (Pocock et al. 2023b).

These same user-structured spatial biases in sampling also likely impact region-specific taxonomic coverage. Previous literature suggests highly active and casual users also differ in their taxonomic preferences. Deitsch et al. (2024), for example, found less-experienced iNaturalist users were more likely to observe large, brightly colored, invasive spiders compared with more-active users. In sum, our work implies that taxonomic biases, in addition to temporal biases, may not be consistent across space, which is relevant for researchers aiming to draw comparisons across land contexts or gradients.

Generalizability and future directions

While our work encompasses more than seven million observations and a broad geographic area, our findings describe patterns and processes within the context of the southeastern USA. We expect many of our findings (i.e., highly active users preferentially targeting biodiverse and protected areas and casual users incorporating observations into daily routines) to be broadly transferable to other regions. This is because behaviors stem from general motivational and logistical constraints (e.g., time availability, travel capacity, proximity to home) that shape opportunistic biodiversity recording regardless of geography (Bowler et al. 2022). While the particular land-use configurations of the southeastern USA are unique, the underlying behavioral drivers are likely common across contexts. However, future studies, building upon the work of Dimson and Gillespie (2023) and our analyses, are still needed to explore whether user groups maintain distinct sampling patterns across different land, cultural, and socioeconomic contexts.

Although we think it likely that many of these conclusions are relevant to other contributory science platforms, previous work suggests differences in participant behavior, motivation, and preferences across platforms (Boakes et al. 2016; Crimmins et al. 2021). Further work to capture dynamics on other platforms would help extend these conclusions to the global citizen science field more broadly and clarify which are unique to iNaturalist.

Here, we focused on two dimensions of user behavior and broadly categorized users. While this was appropriate to answer our questions, future analyses could be extended to include additional and continuous behavioral axes (August et al. 2020). An additional next step to more clearly link motivation to sampling behavior is to survey participants and analyze how their motivations, abilities, and background predict sampling patterns. Finally, it is unknown whether users maintain their sampling preferences over time or shift their observation patterns as they transition to being more highly active. Our data summarizes over those differences, but individual user traces could be used to examine this question more directly.