Filling surveillance gaps: Bayesian INLA models for predicting tick distributions in data-sparse regions

Ticks impose major health and economic losses on the livestock sector of Pakistan, yet uncertainty-aware maps of tick burden remain scarce. We focused on the two most common disease transmitting tick species, Rhipicephalus microplus and Hyalomma anatolicum, to produce exposure-adjusted district-level abundance estimates and predictions for unsampled areas in Punjab and Khyber Pakhtunkhwa (KPK). We compiled heterogeneous tick count records and standardized them per 100,000 animals. District-level climate and physiographic covariates were summarized via principal components analysis. Bayesian spatial models were fit in R-INLA using Gaussian likelihoods and BYM2 over a hybrid adjacency matrix. Competing non-spatial and spatial models were compared, and the best model was used to generate posterior predictions and 95% credible intervals for unsampled districts. Spatial models outperformed non-spatial alternatives and calibrated well. Model robustness was confirmed through eight independent 80/20 train-test splits, showing strong generalization with consistent predictions across seeds. For unsampled areas, R. microplus exhibited a pronounced north-south gradient with high predicted means but wide intervals in the northern highlands, indicating information gaps. H. anatolicum predictions were highest and most precise in southern Punjab. Sensitivity analysis highlighted a dominant spatial component, with modest contributions from PC1 and PC2. The Bayesian spatial models using the Besag-York-Mollie framework delivered comparable, exposure-adjusted tick abundance maps while quantifying uncertainty to guide surveillance. Results suggest a need for immediate control in confirmed hotspots and recommend targeted field sampling in high-uncertainty districts. The workflow generalizes to other vectors, pathogens, and regions for evidence-based livestock health planning.