A machine learning framework for estimating the probability of blacklegged tick population establishment in eastern Canada using Earth observation data.

Ixodes scapularis ticks are the primary vector of Lyme disease (LD) in North America, and their range has expanded into southeastern and southcentral Canada with climate change. This study presents a comprehensive machine learning (ML) framework to estimate the probability of blacklegged tick population establishment as measured using active tick surveillance data. Environmental predictor variables were derived from Earth observation (EO) data at multiple spatial scales to assess their individual contributions in the prediction models. Among the tested ML algorithms, XGBoost emerged as the top-performing model, achieving high sensitivity (0.83) and specificity (0.71) in predicting population establishment. Performance was optimized when using predictor variables derived from a 1 km radius around surveillance sites. Top predictors included cumulative annual degree-days above 0°C and maximum temperature of warmest month, reflecting the importance of temperature in enabling tick survival and reproduction. Additional predictor variables of high importance included silty soil (lower clay content) with slightly higher than average SOC and pH, and land cover types that contained broadleaf forests (percent mixed forest, percent broadleaf) and less urban areas. By integrating ML with open access EO data, this study demonstrates that accurate, easily updatable risk maps can be produced to support public health management of LD, and more broadly, the growing threat of tick-borne diseases in a changing climate.