Post-fire forest recovery trajectory characterized by a modified LandTrendr recovery detection method: A case study of Pinus yunnanensis forests.

• Propose a novel method for post-fire forest recovery extraction. • Modify LandTrendr algorithm for identifying finer post-fire recovery patterns. • Quantify post-fire forest recovery over large areas and long time series. • Reveal latitudinal and altitudinal gradients in P. yunnanensis forest recovery. Forest fires profoundly affect forest growth, and then alter forest ecosystem services and global carbon cycles. Quantitatively characterizing the trajectories of post-fire forest recovery or regrowth is crucial for understanding the effects of increasing wildfires from local to global scales. However, obtaining synoptic and large-scale patterns of post-fire recovery trajectories from remotely sensed data remains challenging. In this study, we propose a modified LandTrendr (LT) recovery detection method (mLT-Recovery) that integrates an optimal segmented LandTrendr algorithm (os-LT) with a recovery trajectory classification method (ReTClass). This novel approach was applied to map and classify the post-fire recovery trajectories of forests predominantly composed of Pinus yunnanensis (P. yunnanensis) , which is a typical fire-adapted tree species in southwest China. The os-LT derives optimal segmented trajectories for each pixel by limiting the maximum number of segments to 3, allowing adjustable trajectory length from the disturbance year to the latest year, and using RMSE instead of p-value from F-statistics as the criterion for selecting an optimal trajectory. The resulting trajectories are classified by the ReTClass based on trajectory morphology, number of segments, and two derivative metrics (i.e., Years to Recovery and Recovery Ratio). Compared to the LT, the os-LT increased the proportion of trajectories with three segments by 59.81 % and lowered the median RMSE of trajectories by 30.63 %. Generally, fire-disturbed P. yunnanensis forests exhibit a dominant recovery trajectory characterized by rapid initial recovery followed by stabilization. The average recovery duration varies substantially across different geographical zones: 8.88 years in plateau temperate humid/sub-humid zone, 6.61 years in mid-subtropical humid zone, and 5.71 years in southern subtropical humid zone, respectively. The outperformance of the mLT-Recovery proposed herein highlights a promising application prospect for accurately characterizing post-fire forest recovery trajectories and estimating carbon sequestration in forest ecosystems. [ABSTRACT FROM AUTHOR]