摘要Insect diversity and abundance are in drastic decline worldwide,but quantify-ing insect populations to better conserve them is a difficult task.Mark-release-recapture(MRR)is widely used as an ecological indicator for insect populations,but the accuracy of MRR estimates can vary with factors such as spatial scale,sampling effort and mod-els of inference.We conducted a 3-year MRR study of B.thaidina in Yanzigou valley,Mt.Gongga but failed to obtain sufficient data for a robust population estimate.This prompted us to integrate B.thaidina life history information to parameterize agent-based models and evaluate the conditions under which successful MRR studies could be conducted.We evaluated:(1)the performance of MRR models under different landscape types,and(2)the influence of experimental design on the accuracy and variance of MRR-based esti-mates.Our simulations revealed systematic underestimates of true population parameters by MRR models when sampling effort was insufficient.In a total of 2772 simulations,subjective decisions in sampling protocol(e.g.,frequency,number of sampling locations,use of spatially explicit models,type of estimands)accounted for nearly half of the vari-ation in estimates.We conclude that MRR-based estimates could be improved with the addition of more field-specific parameters.