A dual-pathway framework for mapping forest age in complex mining landscapes by multi-source remote sensing data and tree growth patterns

Accurate forest age estimation is critical for sustainable forest management, carbon sequestration assessment, and biodiversity conservation. While traditional dendrochronology offers high accuracy, it lacks spatial scalability. The multi-source remote sensing data presents a powerful solution for mapping forest age across large landscapes. This study developed and validated a dual-pathway framework to generate high-resolution forest age maps, specifically designed to improve accuracy in complex mining landscapes containing both disturbed and undisturbed forests. We collected field data from 231 sample plots in the Huodong mining area, using dendrochronology for age determination and LiDAR scanning to derive plot mean height and biomass. The application of two pathways as follows: (1) For disturbed forest, the LandTrendr algorithm was applied to a 39-year Landsat time-series (1985−2023) on the Google Earth Engine platform to identify the year of the stand-replacing disturbance and recovery. (2) For undisturbed forest, we first generated wall-to-wall canopy height map by integrating Sentinel-1, Sentinel-2, Digital Elevation Models, and Global Ecosystem Dynamics Investigation (GEDI) within Gradient Boosting Regression Trees (GBRT) models. Subsequently, eight stratified age-biomass-height (A-B-H) models were developed to predict age from canopy height based on forest types and terrain aspects. The GBRT-derived canopy height map demonstrated acceptable model accuracy (coefficient of determination, R² = 0.716; root mean square error, RMSE = 2.325 m), with an average height of 11.85 m. The A-B-H models using Logistic, Gompertz, and power regression achieved R2 values ranging from 0.623 to 0.776. By synthesizing dual pathway, our framework produced a comprehensive forest age map with a high overall accuracy (R² = 0.817; RMSE = 5.971 years) and a mean forest age of 32.64 years. These results confirm that combining long-term disturbance history with multi-source remote sensing and growth models provides a robust and scalable solution for age estimation across human-impacted forest landscapes. Our framework effectively distinguishes between the age of post-mining recovery forests and that of undisturbed forests, a distinction that is unachievable using a single method. This methodology offers a transferable approach for creating reliable forest age products essential for ecological monitoring and management.