The use of geospatial data and Bayesian Networks to assess the risk status of Mexican amphibians

Abstract

Amphibians are undergoing alarming population declines worldwide, leading to the worst extinction crisis of their history. Here, we use quantitative habitat loss data to assess 144 amphibian species for which we have environmental niche models and environmental remotely sensed information. Moreover, we evaluated previous expert-based assessments replicability using Bayesian Networks (BN) and quantitative habitat loss data as the source of information. BN demonstrated that expert-based assessments are unable to be replicated when using quantitative habitat loss data; the average accuracy for the BN classification by experts was low (42.4% ± 1.5%) with a high percentage of incorrectly classified species (57.6% ± 1.5%). Our assessment offered a high average accuracy (96.1% ± 2%), and a low percentage of incorrectly classified species (3.9% ± 2%). Thus, we propose that qualitative data and expert knowledge should be used together to formalize objective and quantitative evaluation models through BNs to obtain better risk assessments. Only through the use of accurate assessments can we accurately reflect the conservation status of different species groups. Keywords: Amphibian conservation, Geospatial data, Bayesian networks, Species risk assessment