Assessing post-harvest forest recovery using Landsat time series in temperate planted and natural forests

• Spectral and structural recovery indicators were assessed using Landsat time series. • Y2R100% metric effectively captured forest recovery (≥5 m, ≥30 %) across regions. • XGBoost predicted coniferous height recovery with R 2 of 0.69–0.82 and RMSE ∼2 m. • 6–33 % of planted forests remained structurally unrecovered 20 years after harvest. Monitoring forest recovery has important implications for effective forest management. Although Landsat time series data has the potential to characterize forest recovery, the applicability for monitoring recovery conditions in both planted and natural forests across regions remains unexplored. In this study, we evaluated post-harvest spectral and structural recovery metrics derived from Landsat time series using airborne LiDAR data in different regions of temperate planted and natural forests in Japan. We used the Normalized Burn Ratio (NBR) from Landsat time series data to calculate Years to Recovery (Y2R) metrics as indicators of spectral recovery and to identify forest conditions that meet the thresholds of 5 m canopy height and 30 % canopy cover. To assess structural recovery, machine learning models using extreme gradient boosting (XGBoost) regression were developed to predict canopy height and time-series growth in coniferous planted forests. The spectral recovery assessment demonstrated that Y2R threshold based on 100 % (Y2R100%) can be used as an indicator of forest recovery across regions and post-harvest forest types (i.e., coniferous/broadleaved) across Japan. The tree height predicted by the XGBoost models had R 2 and RMSE of 0.69–0.82 and 2.07–2.58 m, respectively. The predicted tree height growth showed statistical equivalence with the dominant height from the middle site index in the yield tables for the main planted species at ages 15–20 years. The rate of forest recovery from the Y2R100% and growth prediction from XGBoost models highlighted areas of replantation failure or locations that had not recovered in natural forests. This study shows the potential combination of spectral and structural recovery for monitoring post-harvest forest recovery.