Decoupling Spectral Confusion in Cold-Temperate Mountain Forests: A Topography–Phenology–Spectral Decoupling Classification Framework

Abstract

Complex topography and canopy spectral saturation during the growing season exacerbate spectral confusion in mountainous forest remote sensing, limiting the accuracy of forest-type mapping in mid- to high-latitude regions. To address this issue, we propose the topography–phenology–spectral decoupling classification framework (TPS-DCF), a knowledge-informed approach that integrates topographic information as ecological constraints on forest-type distribution rather than treating it solely as an auxiliary input. The framework employs a hierarchical classification strategy that combines phenology-guided background separation, topographic-threshold-based decoupling, and multidimensional feature synergy. The Oken River Basin in the Greater Khingan Mountains, China, was selected as the study area. TPS-DCF was evaluated using long-term Sentinel-2 imagery acquired from 2017 to 2025 together with multidimensional topographic data. Based on an independent validation dataset (n= 90), TPS-DCF achieved an overall accuracy of 92.22% and a Kappa coefficient of 0.895. The classification accuracies of deciduous broadleaf, deciduous coniferous, and mixed forests all exceeded 90%. Sensitivity analysis identified 12° as the slope threshold associated with the lowest local misclassification rate among the tested thresholds. Ablation experiments further showed that integrating topographic, phenological, and spectral information improved local classification accuracy by more than 15% within low-slope valley ecotones (0°–4°) relative to the spectral-only baseline. These results indicate that combining topographic, phenological, and spectral information can improve the discrimination and interpretability of spectrally similar forest types in mountainous landscapes.