Estimating species misclassification with occupancy dynamics and encounter rates: A semi‐supervised, individual‐level approach

Abstract

1. Large-scale, long-term biodiversity monitoring is essential to conservation, land management and identifying threats to biodiversity. However, multispecies surveys are prone to various types of observation error, including false-positive/ false-negative detection and misclassification, where a species is thought to have been encountered but not correctly identified. Previous methods assume an imperfect classifier produces species-level classifications, but in practice, particularly with human observers, we may end up with extraspecific classifications including 'unknown', morphospecies designations and taxonomic identifications coarser than species. Disregarding these types of species misclassification in biodiversity monitoring datasets can bias estimates of ecologically important quantities such as demographic rates, occurrence and species richness. 2. Here we present a joint classification-occupancy model that accounts for species non-detection and misclassification. Our framework accommodates extinction and colonization dynamics, allows for additional uncertain 'morphospecies' designations and makes use of individual specimens with known species identities in a semi-supervised setting. We compare the performance of our model to a classification-only model that discards information about occupancy and encounter rate. We illustrate our model with an empirical case study of the carabid beetle (Carabidae) community at the National Ecological Observatory Network Niwot Ridge Mountain Research Station, near Boulder, CO, USA. We also use simulations to evaluate model performance through validation metrics where varying fractions of the data are confirmed. 3. The model supported imperfect classifier accuracy and favoured certain true species classifications strongly for some morphospecies. The model outperformed (e.g. precision) the reduced model that discarded occupancy information, and these differences were most pronounced for abundant species. 4. Spatial and temporal dynamics from modelled occupancy and encounter rates may inform species misclassification probability, but this idea has not yet been | 1529 Methods in Ecology and Evoluঞon SPIERS et al.

Coupling validation effort with in situ bioacoustic data improves estimating relative activity and occupancy for multiple species with cross‐species misclassifications

“…Site‐specific covariates can be included through a generalized linear model framework, , where represents an appropriate link function, represents a row vector of site‐level covariates and represents a vector of regression coefficients for species . Spatial or temporal dependence in occupancy among species can be induced by incorporating hierarchical regression coefficients (Kéry & Royle, 2008; Spiers et al, 2021).…”

“…7). These so‐called ‘coupled classification’ models rely on unambiguous detections to inform species‐specific classification probabilities, which are used to adjust ambiguous detections for false positives (Spiers et al, 2021; Wright et al, 2020). By ‘coupling’ ambiguous and unambiguous detections from the same survey event within an integrated model, classification rates are simultaneously estimated with occurrence and relative activity.…”

Towards estimating marine wildlife abundance using aerial surveys and deep learning with hierarchical classifications subject to error

“…We require three sources of data– 1) the censoring outcomes from the object detection algorithm at Level 1, 2) the CNN classifications of the objects in retained images at Level 3 and 3) paired CNN and human validation data to aid in estimating the CNN classification probability matrix (Wright et al, 2020; Spiers et al, 2022). We will describe each in turn.…”

“…These models were originally formulated for occupancy data that were “contaminated” with false-positive detections and the inferential goal is to estimate the false-positive rate, used to correct the occupancy parameter estimates for misidentified detections (Royle and Link, 2006). These false-positive occupancy models accommodate “target” and “non-target” classes (species) of detections; however, for multi-species surveys, jointly modeling all species and estimating their pairwise classification probabilities is often more appropriate because false-positives for one species may result in false-negatives for another and joint modeling better accommodates this type of data (Wright et al, 2020; Spiers et al, 2022). Further, species classification can be improved when coupled with an ecological model that jointly estimates the habitat associations and detection characteristics of multiple species simultaneously with the false-positive or pairwise species classification probabilities (“coupled classification”; Kéry and Royle, 2020; Spiers et al, 2022).…”

“…These false-positive occupancy models accommodate “target” and “non-target” classes (species) of detections; however, for multi-species surveys, jointly modeling all species and estimating their pairwise classification probabilities is often more appropriate because false-positives for one species may result in false-negatives for another and joint modeling better accommodates this type of data (Wright et al, 2020; Spiers et al, 2022). Further, species classification can be improved when coupled with an ecological model that jointly estimates the habitat associations and detection characteristics of multiple species simultaneously with the false-positive or pairwise species classification probabilities (“coupled classification”; Kéry and Royle, 2020; Spiers et al, 2022). Regardless of the approach, false-positive occupancy models require either known-identity validation samples (or site occupancy states) or informative priors for the classification probabilities (Royle and Link, 2006; Chambert et al, 2015; Wright et al, 2020).…”