Objective Given the complex structure of natural mixed forests and the difficulty of accurately estimating top height, this study aims to utilize high-precision tree height data derived from Airborne LiDAR to systematically evaluate the applicability of multiple top height estimation methods in coniferous mixed forests, broad-leaved mixed forests, and coniferous–broad-leaved mixed forests, and to identify the optimal estimation method for each forest type.

Method Based on data from 42 natural mixed forest plots, combined with high-precision individual-tree height information extracted from Airborne LiDAR point clouds, ten top height estimation methods were developed by integrating three basic estimation approaches—the conventional estimation method, the adjusted largest trees method, and the U-estimator method (UE)—with four dominant tree selection criteria. Uncertainty analysis, difference testing, and correlation analysis were applied, and the results were compared with a general method applicable to all forest types to systematically evaluate the performance of these methods under different forest conditions.

Result (1) Across the three forest types, the estimated top height values differed among methods but exhibited a generally consistent pattern, following the trend of conventional estimation methods > adjusted maximum tree methods > U-estimator methods. In addition, incorporating tree species differences tended to result in lower top height estimates. (2) Differences among estimation methods varied with forest type. In broad-leaved mixed forests, only TH2 (the conventional estimation method selecting the highest trees without considering species differences) and TH10 (the U-estimator method considering both the highest trees and species differences) showed significant differences. In coniferous mixed forests, TH10 differed significantly from several other methods. The coniferous–broad-leaved mixed forests exhibited the most complex pattern, with multiple pairs of methods showing significant or highly significant differences. (3) All estimation methods were positively correlated across forest types, with the strongest inter-method associations observed in broad-leaved mixed forests (r = 0.71–0.99). (4) In all three forest types, the majority of top height estimation methods showed only weak positive relationships with the weighted stand density index (SDI), indicating that these methods were relatively insensitive to stand density. (5) Uncertainty analysis indicated that TH9 (the U-estimator method selecting only the largest-diameter trees) and TH2 performed best in coniferous mixed forests, while TH10 and TH6 exhibited the lowest uncertainty in broad-leaved mixed forests and coniferous–broad-leaved mixed forests, respectively, yielding the most reliable results. (6) Across the three forest types, all top height estimation methods showed moderate to very strong positive correlations with mean stand height, whereas their correlations with stand biomass and basal area increment were generally weak, revealing consistent patterns across methods.

Conclusion This study demonstrates that, for top height estimation in natural mixed forests, matching forest-type-specific algorithms based on Airborne LiDAR data is more scientifically reliable than applying a single universal method. Specifically, the conventional estimation method that selects only the highest trees without considering species differences (TH2) is recommended for coniferous mixed forests, the adjusted maximum tree method with the same selection criterion (TH6) for coniferous–broad-leaved mixed forests, and the U-estimator method that accounts for both the highest trees and species differences (TH10) for broad-leaved mixed forests. These findings further clarify the rational selection of top height estimation methods under natural mixed forest conditions and provide valuable reference for the application of Airborne LiDAR data in forests with complex structural characteristics.