Accounting for Misclassification in Multispecies Distribution Models

TL;DR

This paper introduces a flexible modeling framework that accounts for species misclassification in distribution data, improving accuracy and uncertainty estimates, especially relevant for citizen science biodiversity datasets.

Contribution

It presents a novel, adaptable model that corrects for misclassification errors in species distribution data and estimates associated uncertainties.

Findings

Model effectively corrects misclassification biases.

Accurately predicts true species distribution states.

Demonstrated on butterfly citizen science data.

Abstract

1. Species identification errors may have severe implications for the inference of species distributions. Accounting for misclassification in species distributions is an important topic of biodiversity research. With an increasing amount of biodiversity that comes from Citizen Science projects, where identification is not verified by preserved specimens, this issue is becoming more important. This has often been dealt with by accounting for false positives in species distribution models. However, the problem should account for misclassifications in general. 2. Here we present a flexible framework that accounts for misclassification in the distribution models and provides estimates of uncertainty around these estimates. The model was applied to data on viceroy, queen and monarch butterflies in the United States. The data were obtained from the iNaturalist database in the period 2019 to 2020. 3. Simulations and analysis of butterfly data showed that the proposed model was able to correct the reported abundance distribution for misclassification and also predict the true state for misclassified state.