Lightning geolocation accuracy as the dominant error source in evaluating strike-prediction models for transmission lines

This study evaluates ten Electro-Geometric Models (EGMs) and one Leader Progression Model (LPM) for predicting lightning strikes to a 220 kV transmission line. Using 15 years of EUCLID lightning-location data (2011–2025) processed through IEC 62858 flash grouping and polarity-specific IC current filtering, we identified 7577 flash events from 18,333 stroke records along an 83.9 km line with eight confirmed fault-causing attachments. Our central finding comes from a one-factor-at-a-time Monte-Carlo analysis (200 realizations per error source): among lightning location, peak current, and stroke-type/amplitude classification, only location error materially changes whether individual confirmed events are detected—for both the geometric (EGM_Young) and physics-based (LPM) models—while peak-current and classification error change it for neither. Because the median location error (89 m) is comparable to or larger than the 43–88 m EGM attractive radii, event-level detection is bounded by geolocation accuracy rather than model choice, and 7 of 10 EGMs detect none of the confirmed attachments. We propose detection ratio and predicted attachment rate as metrics better suited to limited confirmed events with undefined true negatives. The LPM and EGM_Young attain the highest detection ratio (2/8); the LPM predicts 39% fewer total attachments (249 versus 411) at 4.8 × higher computational cost, reported as a model property rather than evidence that either is more accurate. This count ordering is invariant across all three error sources, whereas the detection result, resting on only two detected events, is reported as a sensitivity analysis rather than a full variance decomposition. These models can also screen candidate high-risk segments for maintenance prioritization, subject to the stated limitations.