A Growth-Prior-Informed Forest Age Estimation Method Using Overlapping ICESat/GLAS and GEDI Footprints With Reduced Dependence on Field Calibration Data

Abstract

Forest age is a critical ecological variable for quantifying carbon sequestration, assessing biodiversity, and supporting sustainable forest management. Conventional remote sensing approaches for forest age estimation either rely heavily on field inventory data or require long-term optical archives, limiting their scalability and applicability across diverse forest types and age ranges. This study presented a novel light detection and ranging (LiDAR)-based framework that leverages the temporal complementarity of the Ice, Cloud, and land Elevation Satellite/Geoscience Laser Altimeter System (ICESat/GLAS, 2003–2009) and the Global Ecosystem Dynamics Investigation (GEDI, 2019-present). By reformulating classical height-age growth models into a temporally paired-height structure, canopy height at the later observation was expressed as a function of canopy height at the earlier observation and the elapsed time interval. Overlapping GLAS and GEDI footprints provide paired canopy height observations spanning more than a decade, enabling direct parameterization of growth models with reduced dependence on field-derived age calibration data. The framework was evaluated in two ecologically distinct regions. For Larix gmelinii in Northeast China, the estimated ages achieved an R2 of 0.86 with a root-mean-square error (RMSE) of 13.37 years. For loblolly pine in the southeastern United States, the corresponding R2 and RMSE were 0.71 and 5.74 years, respectively. Further simulations with random height measurement errors confirmed the resilience of the model, demonstrating strong robustness for loblolly pine and relative stability for L. gmelinii, with increased sensitivity only in later growth stages. Building on these results, the models were applied at the regional scale to generate 30 m resolution forest age maps for both study areas. At the regional scale, the estimated mean age in Study Area I was 55.8 years, broadly consistent with the estimates from two independent national products (59.6 and 64.5 years), and the young-forest proportion was 16.4%. In Study Area II, map-derived ages showed strong agreement with stand-level observations (R ${}^{2} =0.96$ , RMSE = 2.83 years). Overall, this LiDAR-driven framework substantially reduces dependence on dense field calibration and long-term optical time series, offering a complementary pathway for full-spectrum forest age estimation.