Species distribution modeling with expert elicitation and Bayesian calibration

Abstract

Species distribution models (SDM) are key tools in ecology, conservation, and natural resources management. They are traditionally trained with data on direct species observations. However, if collecting species data is difficult or expensive, complementary information sources on species distributions are needed. Expert knowledge has been demonstrated to improve SDM predictions in a number of such applications but there is still no consensus on methods to integrate information from several experts into a single coherent species distribution prediction. Moreover, since expert assessments are inherently subjective and prone to biases, expert-driven SDMs should calibrate their assessments. We propose a method to tackle these challenges by extending the hierarchical Bayesian integrated species distribution modeling framework to expert informed species distribution modeling. We treated map-like expert assessments as data and integrated them with calibration data on species recordings. Our integrated SDM has model components to estimate experts' reliability and to adjust for potential biases in their assessments. After integrated inference, we used the model to make predictions over a study area. We tested our approach with an extensive simulation study and a real world case study comprising ten expert assessments and survey data on pikeperch larvae from a coastal area of the Gulf of Finland. Expert assessments significantly improved species distribution predictions compared to predictions conditioned on survey data only. They also improved parameter inference, thus strengthening the ecological interpretation of the results. The skill of the experts, and biases in their assessments, varied considerably in the case study though, emphasizing the importance of formal expert calibration provided by our model. Our results show that expert elicitation can be an efficient tool for improving species distribution model predictions. Our approach is especially useful for applications where any type of species data are expensive to collect but local species experts can easily be reached.