The Journal of Engineering Geology ISSN:1226-5268(Print) 2287-7169(Online) 지질공학

This study evaluated the spatial agreement of debris-flow runout simulated using a random walk model (RWM) with unmanned aerial vehicle (UAV) observations of the 2023 Beolbang-ri debris-flow event in Yecheon, South Korea. Five observed debris-flow initiation points identified from UAV and field surveys were used as release locations to reduce uncertainty associated with initiation. The simulations were conducted on a 10 m grid, and computational stability was examined by comparing 10³ and 10⁴ random walks per case. Sensitivity analyses were subsequently performed for the mobility parameter 𝛺 and normalized impact frequency (NIF) threshold. Spatial agreement between the observed and modeled footprints was quantified using Precision, Recall, F1-score, and intersection over union (IoU), while the longitudinal extent was evaluated using axial runout. Increasing the number of random walks from 10³ to 10⁴ resulted in only minor changes in IoU and F1-score, whereas computation time increased by approximately 8.5 times. Within the tested parameter space, the highest spatial agreement was obtained at 𝛺 = 9° and an NIF threshold of 0.125. Under these conditions, the observed and modeled footprint areas were 0.0931 and 0.1321 km², respectively, with Precision = 0.440, Recall = 0.624, F1-score = 0.516, and IoU = 0.348. The modeled footprint was approximately 41.9% larger than the observed footprint, indicating a tendency toward spatial overprediction. In contrast, the observed and modeled axial runout extents were 1,738.3 and 1,779.4 m, respectively, corresponding to an overestimation of only 2.37%. These results indicate that the RWM reproduced the overall travel direction and longitudinal runout more closely than the detailed two-dimensional footprint. The calibrated parameters represent an event-specific solution rather than a universally optimal parameter set. The proposed UAV-based evaluation framework can support quantitative assessments of RWM performance for broad-scale debris-flow runout screening and provides a basis for future multi-event calibration and independent validation.