Abstract: Abstract: The airborne light detection and ranging (LiDAR) has already been widely used in forest inventory investigation with the advantage of obtaining multiple forest information. The canopy height model (CHM) derived from LiDAR data is a key model, which is used frequently to retrieve forest parameters, such as the tree height, crown width, diameter at breast height, crown density, volume and biomass and so on. However, there exist some abnormal or sudden pits in CHMs, which will have an influence on forest parameters extraction. A method of removing pits from LiDAR-derived CHMs with fusion of the CHMs of the first return point cloud by inverse distance weighted (IDW) interpolation, which is layered according to different thresholds of canopy height, is proposed in the article. In general, the search radius of IDW interpolation is usually set to 1-1.5 times of the original points cloud interval. Three plots are chosen as the experimental data, which respectively represent the needle leaf forest, broadleaf forest, mixed needleleaf and broadleaf forest. Mean pixel values of difference image between CHM and CHM0 is 3.31, 4.20, 5.88 m respectively in the three plots, which illustrate that the pits are close to the upper canopy, and mainly generated by the first echo point coming from the foliage inside canopy of tree. Maximum pixel values in the difference images were 12.97, 14.99, 29.00 m respectively, which is very close to the measured maximum tree height and maximum height of the normalized point cloud. This indicates that a small amount of CHM pits are generated by the first echo points close to the ground. In other words, the first echo generated CHM pits probably roots in shrubs, low saplings and bare land. Comparative analysis with the profile of CHM0, CHM and the original points cloud was performed. CHM removed pits tallies exactly with the original point cloud, which shows that the algorithm developed in this paper changes very little on height of canopy cloud, and keeps the surface structure shape of the original canopy, and loses very little information of canopy. At the same time, we have carried out the experiments that remove pits from LiDAR-derived CHMs when the point clouds in the plots were thinned. And then the result was compared with the smoothing filter algorithm. By comparison，the results show that the proposed algorithm can effectively remove pits of CHM and reserve forest canopy border clearly and canopy gaps. The CHM can accurately express the forest canopy shape, and the proposed algorithm is suitable for different forest types, with the adaptability to point cloud density. The effect of removing pits is better than the smoothing filter algorithms such as the median filtering, Mean filtering and Gaussian filtering. The pit-free and optimized CHM contributes to the subsequent extraction of forest parameters. However, due to the IDW interpolation searching along any direction with a certain radius range, there exists a certain degree of smoothing in canopy, and the small forest gaps are slightly filled, and the height of forest gaps slightly increased, and the crown edge is slightly dilated.