Correcting Chance Coincidence and Observation-Unit Mismatch in Cross-Sensor Validation of Event-Based Remote Sensing Products

Cross-sensor validation of event-based remote-sensing products often converts spatiotemporal matches directly into detection-performance estimates, although candidate competition, reference reuse, chance coincidence, and non-equivalent observation units alter the measured quantity. We present a validation method that constructs the complete candidate graph, resolves matches by global one-to-one assignment, estimates the temporal background by repeating the same assignment after physically bounded shifts, and propagates uncertainty across event definitions and block resamples. Across 48 simulations with known event identities, the method reduced mean absolute error from 7.9 to 3.1 percentage points; conventional many-to-one matching with an unsuitable temporal shift produced errors of up to 49.1 percentage points. Applied to 51,290 FY-4A/LMI groups and ground-based lightning records over the East China Sea during June–September 2022, the any-neighbour rate was 39.1%, the one-to-one rate was 19.8–20.9%, and 75.8% of conventional pairs reused a ground detection. A same-season eastern-lowland–Tibetan Plateau comparison showed that local background estimation can reverse the apparent regional contrast. The method provides a reproducible, uncertainty-resolved basis for cross-sensor validation of event-based remote-sensing products.