High-Resolution Lithofacies Identification in Volcanically Influenced Lacustrine Shales: A Case Study of the Chang 73 Sub-Member, Yanchang Formation, East Gansu Area, Ordos Basin

Abstract

Lithofacies identification in the Chang 73 Sub-Member is challenged by tephra input and thin interbeds, which distort conventional well logging responses and increase vertical heterogeneity. Consequently, classification based directly on conventional well logs is often unreliable for fine-scale reservoir characterization. This study presents a geologically constrained hierarchical method for lithofacies identification in a volcanically influenced shale interval. The proposed method includes tuffaceous shale indicator (TSI)-based partitioning of tephra response intervals, constrained prediction of mineral components in different response domains, and sequence-based lithofacies classification. Specifically, two convolutional neural network (CNN)-bidirectional long short-term memory (BiLSTM) models are developed: a regression architecture for key mineral-content prediction within different tephra response domains, and a sequence-classification architecture for lithofacies identification. The lithofacies classification model integrates conventional well logs, predicted mineralogical attributes, and lamina density (LD) derived from electrical image logging. The results demonstrate that the method improves lithofacies prediction relative to conventional direct classification methods. TSI partitioning reduces tephra response interference in mineral prediction, while LD strengthens the identification of thin interbeds that cannot be resolved reliably from conventional well logs alone. The combined use of compositional, textural, and sequential information improves both model performance and geological consistency. This method extends discrete core-based lithofacies interpretation to continuous log intervals and provides a more robust basis for reservoir characterization and shale oil sweet-spot evaluation in the Chang 73 Sub-Member. The method is applicable to shale reservoirs where tephra input significantly affects well logging responses and lithofacies variability.