Fig 1.
The Bird Conservation Regions (BCRs) of North America.
BCRs are ecologically distinct regions with similar bird communities. BCRs were used to cluster checklists into groups with similar likelihoods of species encounter. Data from six BCRs were used in this study: BCR 9- Great Basin, BCR 23- Prairie Hardwood Transition, BCR 30- New England/Mid-Atlantic Coast, BCR 31- Peninsular Florida, BCR 32- Coastal California, and BCR 37- Gulf Coastal Prairie. These BCRs were selected to include a range of eBird participation (Table 1) and varying degrees of diversity and patterns of bird occurrence. Map provided by the North American Bird Conservation Initiative (www.nabci-us.org).
Table 1.
The number of observers and number of checklists submitted for each Bird Conservation Region.
Fig 2.
Species accumulation curves for all individual observers in a BCR.
Each line represents the species accumulation curve derived from the mixed model fit to data from that BCR, for every individual observer in that BCR, and calculated with the standardized covariates of Sep 1st, 7am start time, travelling 1km and average percentage land cover. The fitted line for each observer is plotted to the maximum count period duration in the data from that observer. Species accumulation curves that decrease for some observers may indicate different biases in attention to birding. For example checklists under 1 hour may be more concentrated birding, whereas checklists over 1 hour may combine birding with another activity such as hiking or fishing.
Fig 3.
Representative Individual Species Accumulation Curves and Indices.
Actual counts of species reported and observer-specific SACs for one individual observer classified as within the lowest quartile a), middle quartiles b), and highest quartile c). As the duration an observer spends collecting data for each checklist increases, the number of species observed increases. However as the duration lengthens, the rate of species accumulation decreases. The black dot on each curve is the SAC Index—the estimated number of species that individual would see during one hour of birding. Individuals were selected from the group who submitted at least 300 checklists between 2002 and 2012.
Fig 4.
Distribution of individual SAC Indices.
The expected number of species observed in 1 hour for all observers in a BCR. Individual data submission scores are ranked from lowest to highest and the light gray region represents the lower quartile of observers, and the dark gray region the upper quartile of observers.
Fig 5.
Comparison of detection rates of individual species of birds for observers in the lower quartile and the upper quartiles of SAC indices.
Detection rates are the proportions of checklists that record a given species and lines represent 95% bootstrap confidence intervals. The red line is a model fitted to the logit detection rates. The gray line indicates the line of equality, where detection rates for the two groups are equal; the highest quartile had statistically significant detection rates for the majority of species.
Fig 6.
Bird species detected at the most similar and dis-similar rates by observers in the lowest quartile and highest quartile of SAC index values.
Barplots from BCR 23 of the 20 species for which detection rates are proportionally most similar (left-hand panels) and the 20 species for which detection rates are proportionally most different (right-hand panels). Detection rate is the proportion of checklists that record a given species and error bars represent 95% bootstrap confidence intervals. The 20 species for which the two groups have proportionally most similar detection rates are generally species that are fairly easy to identify by sight. The 20 species that the two groups have proportionally most different detection rates are generally species that are difficult to identify, easier to identify by sound, or often be seen as a high-flying silhouette without many distinguishing features.
Fig 7.
Bird species detected at the most similar and dis-similar rates by observers in the lowest quartile and highest quartile of SAC index values.
Barplots from BCR 31 of the 20 species for which detection rates are proportionally most similar (left-hand panels) and the 20 species for which detection rates are proportionally most different (right-hand panels). Detection rate is the proportion of checklists that record a given species and error bars represent 95% bootstrap confidence intervals. The 20 species for which the two groups have proportionally most similar detection rates are generally species that are fairly easy to identify by sight. The 20 species that the two groups have proportionally most different detection rates are generally species that are difficult to identify, easier to identify by sound, or often be seen as a high-flying silhouette without many distinguishing features.
Table 2.
The mean and median number of checklists submitted for the lowest quartile and highest quartile of eBird participants.
Fig 8.
The change in SACs as a function of the cumulative participation in eBird.
We estimated average changes in shapes of species accumulation curves with increasing number of checklists submitted to eBird from our BCR-specific models of species accumulation curves, to visualize whether observers report more species after they have submitted more eBird checklists. Note that while increased participation leads to a higher rate of accumulation of species, this effect is highest for beginning participants and slows with increased participation.
Fig 9.
Species distribution model accuracy with and without data submission scores (AUC a and Kappa b.).
There are 20 species models from each of the 6 BCRs: 10 for species, which are hard to identify, and 10 for species that are easier to identify, as defined by differences between the highest and lowest quartiles of data submission scores (Figs 6 and 7). In most cases the inclusion individual observer data submission scores improved model accuracy.