Abstract
Understanding the landscape determinants that escalate Lyme disease (LD) risk through various times and regions is vital for appraising disease susceptibility and shaping precise intervention and prevention strategies. This research introduces a novel data-driven framework to identify potential indicators from an extensive array of potential variables. We then deployed an advanced spatiotemporal weighted regression (STWR) model to investigate how landscape fragmentation metrics correlate with the spatiotemporal variability of LD incidence rate in Connecticut towns. We proposed a data-driven filtering framework to select five variables from a large data pool. The analysis unveils that LD incidence rates exhibit heightened sensitivity to proportional or exponential shifts in landscape fragmentation; logarithmic and squared transformations of landscape metrics shed light on lesser effects and venue for potential parabolic relationships. Observations also disclose significant spatial trends, showing elevated LD incidence rates in locales with vast, uninterrupted deciduous forests, alongside contributions from wetland ecosystem-related variables to the rise in disease occurrence. Compared with Geographically Weighted Regression (GWR), the STWR model proved more potent and reliable with higher R2 and lower estimated standard errors (SE). The STWR model is highly flexible in terms of spatiotemporal variations in data. The STWR results further reversely indicate the changes made by the Center for Disease and Prevention (CDC) in the case classification of LD in 2008. The integration of data-driven and model-driven approaches in this study delivers a robust framework that combines empirical pattern detection with theoretical insight, enhancing the robustness and predictive power of ecological studies.
•An innovative data-driven framework was proposed for variable filtering.•The significant patterns in the STWR results help identify heterogeneities in Lyme disease transmission.•STWR model is highly flexible in the spatiotemporal variation of data.•The synergy between data-driven and model-driven techniques unlocks new insights and explores latent relationships.