Mapping three decades of forest structural changes in Japan using Landsat time series and airborne LiDAR data

Understanding nationwide forest structural changes is crucial for sustainable forest management and policy formulation. Airborne Light Detection and Ranging (LiDAR) can provide accurate three-dimensional forest structural information; however, it is challenging to implement large-scale spatially explicit mapping. Integrating LiDAR with Landsat time series data offers the potential to map forest structure consistently over space and time. However, existing studies have not fully leveraged the rich temporal information derived from multiple change detection algorithms. In this study, we integrated large-scale airborne LiDAR data and Landsat time series data with U-Net convolutional neural network models to map forest structural attributes (i.e., canopy height, stand volume, aboveground biomass, and basal area) across Japan at 30 m resolution annually from 1990 to 2021. We utilized airborne LiDAR data covering a total of 4.6 million ha (2017–2021) and extended plot attributes from National Forest Inventory data. Landsat-derived predictors from LandTrendr and the Continuous Change Detection and Classification (CCDC) algorithms were used as inputs to the U-Net models. The developed models achieved R 2 ranging from 0.73 to 0.76, with a root mean squared error of 3.26 m (17.1%) for canopy height, 128.11 m 3 /ha (29.6%) for stand volume, 47.80 Mg/ha (21.5%) for aboveground biomass, and 8.46 m 2 /ha (16.5%) for basal area. The ablation studies revealed that predictors from both the LandTrendr and CCDC were effective in improving model performance, highlighting the value of integrating multiple change detection algorithms. The total estimated stand volume for 2021 was 9.57 ± 0.31 billion m 3 , reflecting an increase of 140.2% from that in 1990. The generated maps revealed substantial spatial and temporal variations, with lower increasing rates observed in western Japan. This study provides a scalable framework for mapping nationwide forest dynamics over three decades, supporting data -informed forest management.