Search for: All records

Citizen science biodiversity data present great opportunities for ecology and conservation across vast spatial and temporal scales. However, the opportunistic nature of these data lacks the sampling structure required by modeling methodologies that address a pervasive challenge in ecological data collection: imperfect detection, i.e., the likelihood of under-observing species on field surveys. Occupancy modeling is an example of an approach that accounts for imperfect detection by explicitly modeling the observation process separately from the biological process of habitat selection. This produces species distribution models that speak to the pattern of the species on a landscape after accounting for imperfect detection in the data, rather than the pattern of species observations corrupted by errors. To achieve this benefit, occupancy models require multiple surveys of a site across which the site's status (i.e., occupied or not) is assumed constant. Since citizen science data are not collected under the required repeated-visit protocol, observations may be grouped into sites post hoc. Existing approaches for constructing sites discard some observations and/or consider only geographic distance and not environmental similarity. In this study, we compare ten approaches for site construction in terms of their impact on downstream species distribution models for 31 bird species in Oregon, using observations recorded in the eBird database. We find that occupancy models built on sites constructed by spatial clustering algorithms perform better than existing alternatives. 

Nest predation rates critically influence avian biodiversity and evolution. In the north temperate zone, increased nest failure along edges of forest fragments is hypothesized to play a major role in the disappearance of bird species from disturbed landscapes. However, we lack comprehensive syntheses from tropical latitudes, where biodiversity is highest and increasingly threatened by habitat fragmentation and disturbance. We assembled data from five decades of field studies across the global tropics (1,112 populations of 661 species) and used phylogenetic models to evaluate proposed predictors of nest success. We found significant effects of several traits, including adult body mass and nest architecture. Contrary to results from many temperate locations, anthropogenic habitat disruption did not consistently reduce nest success; in fact, raw nest success rates were lower in large tracts of primary forest than in disturbed or fragmented landscapes. Follow-up analyses within species, using a subset of 76 species for which we had estimates of nest survival in habitats with different levels of disruption, confirmed that neither disturbance nor fragmentation significantly influenced nest success. These results suggest that nest predation alone cannot explain observed declines in avian biodiversity in tropical forest fragments, raising new questions about the demographic processes that drive extinction in the tropics. 

Geospatial problems often involve spatial autocorrelation and covariate shift, which violate the independent, identically distributed assumption underlying standard cross-validation. In this work, we establish a theoretical criterion for unbiased crossvalidation, introduce a preliminary categorization framework to guide practitioners in choosing suitable cross-validation strategies for geospatial problems, reconcile conflicting recommendations on best practices, and develop a novel, straightforward method with both theoretical guarantees and empirical success. 

Geostatistical learning problems are frequently characterized by spatial autocorrelation in the input features and/or the potential for covariate shift at test time. These realities violate the classical assumption of independent, identically distributed data, upon which most cross-validation algorithms rely in order to estimate the generalization performance of a model. In this paper, we present a theoretical criterion for unbiased cross-validation estimators in the geospatial setting. We also introduce a new cross-validation algorithm to evaluate models, inspired by the challenges of geospatial problems. We apply a framework for categorizing problems into different types of geospatial scenarios to help practitioners select an appropriate cross-validation strategy. Our empirical analyses compare cross-validation algorithms on both simulated and several real datasets to develop recommendations for a variety of geospatial settings. This paper aims to draw attention to some challenges that arise in model evaluation for geospatial problems and to provide guidance for users. 

This paper discusses opportunities for developments in spatial clustering methods to help leverage broad scale community science data for building species distribution models (SDMs). SDMs are tools that inform the science and policy needed to mitigate the impacts of climate change on biodiversity. Community science data span spatial and temporal scales unachievable by expert surveys alone, but they lack the structure imposed in smaller scale studies to allow adjustments for observational biases. Spatial clustering approaches can construct the necessary structure after surveys have occurred, but more work is needed to ensure that they are effective for this purpose. In this proposal, we describe the role of spatial clustering for realizing the potential of large biodiversity datasets, how existing methods approach this problem, and ideas for future work. 

The growth of biodiversity data sets generated by citizen scientists continues to accelerate. The availability of such data has greatly expanded the scale of questions researchers can address. Yet, error, bias, and noise continue to be serious concerns for analysts, particularly when data being contributed to these giant online data sets are difficult to verify. Counts of birds contributed to eBird, the world’s largest biodiversity online database, present a potentially useful resource for tracking trends over time and space in species’ abundances. We quantified counting accuracy in a sample of 1,406 eBird checklists by comparing numbers contributed by birders (N = 246) who visited a popular birding location in Oregon, USA, with numbers generated by a professional ornithologist engaged in a long-term study creating benchmark (reference) measurements of daily bird counts. We focused on waterbirds, which are easily visible at this site. We evaluated potential predictors of count differences, including characteristics of contributed checklists, of each species, and of time of day and year. Count differences were biased toward undercounts, with more than 75% of counts being below the daily benchmark value. Median count discrepancies were −29.1% (range: 0 to −42.8%; N = 20 species). Model sets revealed an important influence of each species’ reference count, which varied seasonally as waterbird numbers fluctuated, and of percent of species known to be present each day that were included on each checklist. That is, checklists indicating a more thorough survey of the species richness at the site also had, on average, smaller count differences. However, even on checklists with the most thorough species lists, counts were biased low and exceptionally variable in their accuracy. To improve utility of such bird count data, we suggest three strategies to pursue in the future. (1) Assess additional options for analytically determining how to select checklists that include less biased count data, as well as exploring options for correcting bias during the analysis stage. (2) Add options for users to provide additional information that helps analysts choose checklists, such as an option for users to tag checklists where they focused on obtaining accurate counts. (3) Explore opportunities to effectively calibrate citizen-science bird count data by establishing a formalized network of marquis sites where dedicated observers regularly contribute carefully collected benchmark data.