Low-Disturbance Prediction of Scaling-Off Deterioration on the Shaded and Sunward Sides of Earthen Heritage Sites Based on Comprehensive Assessment of Weathered-Layer Characteristics

Multi-indicator investigations of scaling off at earthen heritage sites can increase disturbance, while prediction for shaded and sunward sides using few readily obtainable indicators remains underexplored. This study examines 10 Ming Great Wall sites in northern Shaanxi. Crust layer thickness, powdered layer thickness, mean annual rainfall, and plasticity index characterize weathered-layer development, soil properties, and environmental conditions. Equal-weight and entropy-weighted versions of the technique for order preference by similarity to an ideal solution (TOPSIS) generate composite deterioration scores and rankings for shaded and sunward sides, respectively. Six prediction models use input variables excluded from the composite assessment; rankings from held-out predictions or pairwise voting are compared with assessment rankings. The tournament leave-pair-out k-nearest-neighbor model (TLPO-KNN) for shaded sides and support vector regression (SVR) for sunward sides achieve the highest Spearman correlations, 0.872 and 0.903, respectively. Mean absolute rank errors are 1.0 and 0.8 rank positions, respectively. These models require only crust layer clay content, or clay content and damage ratio, respectively. Within this sample, few low-disturbance, readily obtainable indicators reproduce the main assessment ordering, supporting face-specific deterioration screening and priority wall-face verification.