Assessing Species Distribution Models for Fine-scale Predictions of Ixodes Scapularis, what are we missing?

Ticks and tick-borne diseases are increasingly threatening public health in the United States, emphasizing the importance of accurately predicting their distribution to develop effective management strategies. However, modeling tick distributions can be challenging due to their three-host life cycle, clustered dispersion, and dependence on specific microhabitats. In this study, we compared three modeling methods for predicting the distribution of blacklegged ticks ( Ixodes scapularis ) across three urban parks in Maryland: presence-only Maximum Entropy (MaxEnt), presence-only Log Gaussian Cox Processes (LGCP) utilizing a latent stochastic partial differential equation (SPDE), and a presence-absence GLMM with an SPDE, based on site-specific, field-collected non-detections. We aimed to assess whether a spatially continuous presence-absence GLMM-SPDE could serve as an alternative or complement to the popular MaxEnt model, potentially offering better computational efficiency and predictive accuracy. Results indicated that both MaxEnt and LGCP models predicted tick distributions moderately well, although the MaxEnt model tended to overpredict presence in fragmented urban environments. The presence-absence model achieved the highest accuracy (mean AUC = 0.854 ± 0.04; CBI = 0.985), effectively identifying occupied sites while maintaining reasonable specificity, primarily when park-specific thresholds were used. These findings demonstrate that integrating a continuous spatial autocorrelation structure enables presence-only GLMMs to perform adequately. However, the most precise predictions in diverse urban areas come from field-collected presence-absence data. Therefore, we recommend using spatially explicit binomial SPDE-based GLMMs that require field ecologists to check drags or flags during tick sampling systematically and to record both absences and presences.