Wang, Yan; Wang, Hengbin; Yang, Jingyi; Li, Zhibin; Liu, Zhe; Zhang, Xiaodong; Li, Shaoming; Galleguillos, Mauricio; Zhao, Yuanyuan
Abstract
Forest type and forest age constitute important variables for carbon budget quantification, biodiversity conservation, and ecosystem management. However, current data acquisition predominantly relies on localized forest inventories or manual image interpretation, while large-scale, high-resolution mapping of natural and planted forests and their ages still faces two critical challenges: the dependence of conventional classification methods on labor-intensive, contemporaneous training samples, and the reliance of existing forest age estimation approaches on single-algorithm-based strategies. In this study, we proposed a framework that automatically updated historical training samples for forest type mapping and employed multitemporal remote sensing series with differentiated fusion strategies to estimate forest age. First, the continuous change detection and classification (CCDC) algorithm was applied to identify breakpoints in time series and automatically update historical samples. Then, the dynamically updated samples and long-term trend features extracted using the LandTrendr (LT) algorithm were used to train a random forest classifier, achieving an overall accuracy of 89.54%. Finally, based on the Landsat normalized burn ratio time series from 1985 to 2023, we independently applied both LT and CCDC algorithms and generated eight forest age maps through different strategies. Among these, the maximum-value fusion approach-selecting the oldest valid age per pixel-outperformed others, achieving a +/- 2-year estimation accuracy of 84.39%. This study presents the first high-resolution maps of forest type and age in Chile by integrating multitemporal remote sensing data with dynamic sample updating, filling a critical gap in forest type-age mapping and supporting carbon accounting and long-term ecosystem monitoring.
A_Dynamic_Time-Series_Framework_for_Mapping_Natural_and_Planted_Forests_and_Estimating_Forest_Age_at_30-m_Resolution.pdf