Abstract: The spatiotemporal distribution of individual tree age within a forest community is important for understanding ecological processes, such as competition, succession, and function, at different scales. However, traditional methods are expensive and inefficient, particularly at large scales. This study proposes a novel conceptual framework to obtain the Forest Community Age Spectrum (FCAS) and evaluates its feasibility by integrating an explainable machine learning model with high-resolution hyperspectral remote sensing of leaves. Focusing on Larix gmelinii, we used hyperspectral data to rapidly estimate tree age. The results showed that the hyperspectral model of mature leaves could accurately estimate tree age (best model performance: R² = 0.78, RMSE = 6.13, RPD = 2.12). The model performed best in the 400–1000 nm wavelength band because of leaf structure-sensitive wavelength (near 644.88 nm) band and Photosynthetic pigment wavelength bands (701–724 nm), and captured the entire age gradient within the 900–1700 nm wavelength band due to the presence of phenolic aldehyde and other secondary metabolite-sensitive wavelength bands (1460–1517 nm and 1600–1700 nm). Overall, this study successfully established a key methodological foundation for estimating tree age and, ultimately, constructing the FCAS. The framework provides a potential pathway for future FCAS-based research to quantify spatial age patterns and investigate mechanisms driving competition, succession, functional optimization, and carbon sequestration. These findings offer both an empirical basis and an operational tool for quantitatively linking forest age structure with core ecological processes through FCAS, representing a critical first step toward its realization.

Key words: hyperspectral, remote sensing, machine leaning, community, inversion, tree age, forest community age spectrum