currently in progress. Overall, this research demonstrates that ANN-based susceptibility mapping provides a scalable, quantitative, and operationally relevant solution for geological risk management. By enabling resolution-adaptive hazard assessment from national planning to mine-site operations, the methodology supports safer infrastructure design, improved geotechnical decision-making, and more responsible mineral production across Peru. KEYWORDS Machine Learning, Landslide, Geological Risk, ANN, Downscaling 1. INTRODUCTION Landslide susceptibility, defined as the propensity of an area to experience a specific landslide process, is a critical component of hazard assessment (INGEMMET, 2010). In Peru, landslides represent a persistent threat to infrastructure, communities, and mining operations due to steep Andean topography, active tectonics, and intense seasonal precipitation (PuenteSotomayor et al., 2021). For the mining sector, slope instability directly affects haul roads, waste dumps, tailings storage facilities, access platforms, and associated infrastructure, potentially disrupting mineral production chains and increasing operational risk. Accurate susceptibility mapping is therefore essential not only for national planning but also for ensuring operational continuity and geotechnical risk management in mining environments. Landslide susceptibility zonation methods are commonly classified as qualitative (direct) and quantitative (indirect) approaches (Mandal & Mondal, 2019). In Peru, official susceptibility assessments have traditionally been developed at Macro-Regional (≥1:250,000) and Regional (≈1:100,000) scales following national geomorphological zoning standards. Macro-Regional maps provide broad territorial characterization based on lithological, structural, and morphoclimatic domains, whereas Regional maps refine susceptibility using expert-weighted thematic overlays of slope, geology, geomorphology, and land cover (Villacorta et al., 2012; Luque Poma et al., 2021). Although these maps provide comprehensive spatial coverage, their reliance on heuristic weighting schemes and fixed cartographic scales limits their quantitative transferability to engineering and mining applications, where higher spatial resolution and probabilistic consistency are required. Quantitative approaches attempt to reduce subjectivity by statistically relating landslide occurrences to conditioning factors. Bi-variate statistical models such as weights-of-evidence have been applied at local scale in northern Peru (Vilchez Mata & Medina Allcca, 2008), while recent machine learning frameworks have demonstrated improved predictive performance at regional scale (Kumar et al., 2023). However, no country-wide Artificial Neural Network (ANN) model
RkJQdWJsaXNoZXIy MTM0Mzk2