143 Convolutional Variational Autoencoder (VAE-CNN), which learns intrinsic geological patterns directly from multi-source datasets without requiring explicit non-deposit labeling. To evaluate spatial generalization, the study area is subdivided into western and eastern sectors: deposits in the western sector are used to define the porphyry latent signature, while those in the eastern sector are reserved for independent validation. The methodology follows a two-stage approach. First, the VAE-CNN is trained to learn latent representations of spatial patterns across the study area. Then, subcrops associated with confirmed porphyry deposits are projected into this latent space and aggregated to define a reference signature. The similarity between this signature and all subcrops is subsequently quantified, enabling the identification of regions with comparable geological characteristics. 2. STUDY AREA The study is conducted in the Yukon a geologically prospective region in Canada known for hosting porphyry copper systems and a wide range of mineral occurrences. The analyzed area covers approximately 219.9 km × 236.7 km (Figure 2). Tthe dataset of known deposits was spatially partitioned into two sectors: • Western sector: used for extraction of the porphyry signature, incorporating 25 known deposits. • Eastern sector: reserved exclusively for validation, containing 60 known deposits 3. DATA COMPILATION AND PREPROCESSING 3.1. Data Sources Geoscientific datasets were compiled from publicly available repositories provided by the Yukon Geological Survey and satellite-based sources, including the Alaska Satellite Facility (ASF) for remote sensing data. The primary datasets used in this study include:
RkJQdWJsaXNoZXIy MTM0Mzk2