Highly pathogenic avian influenza (HPAI) remains a major threat to poultry production and animal-health preparedness, but its rarity makes spatially consistent local surveillance prioritization difficult. This study developed a spatial rare-event modelling framework to transform sparse HPAI detections into 3-kilometre surveillance-priority scores in Delaware and evaluated adjacent-region transfer to Maryland. HPAI presence was modelled using poultry-house infrastructure, wetland and hydrographic proximity, agricultural land cover, topography, and wild-bird occurrence predictors. Model performance was assessed using spatial-block validation in Delaware and transfer to a separate Maryland grid. Elastic-net logistic regression achieved the highest Delaware area under the receiver operating characteristic curve (AUC) point estimate of 0.935 (95% confidence interval [CI]: 0.868–0.986), but discrimination was not statistically distinguishable from ridge logistic regression. The highest-ranked 5%, 10%, and 20% of Delaware cells contained 63.6% (95% CI: 30.8–89.1%), 81.8% (48.2–97.7%), and 90.9% (58.7–99.8%) of the 11 HPAI-positive cells, respectively. In Maryland, the Delaware-trained model achieved an AUC of 0.898 (95% CI: 0.719–0.984) and captured 64.3% (95% CI: 35.1–87.2%) and 85.7% (57.2–98.2%) of the 14 HPAI-positive cells in the highest-ranked 5% and 10%, respectively. Poultry infrastructure provided most of the discrimination (AUC = 0.930); adding environmental predictors increased AUC to 0.935, a difference not statistically distinguishable from zero. These findings support surveillance-priority screening for poultry biosecurity, outreach, and preventive action without attributing disease to individual farms.