The LiDAR frame data can be processed by LiDAR SLAM algorithms to obtain target point cloud of forest plots. As a SLAM algorithm, LOAM can locate and map by extracting line and surface features, which has the advantages of fast computing speed and better robustness than ICP algorithm. However, it is difficult to use this algorithm directly in forest inventory due to poor lines and surface features in forests. A forest LiDAR SLAM system was developed to address the problem as follows: the modules of secondary de-distortion and secondary registration were introduced into the workflow of the new SLAM system to improve the robustness and accuracy of location and mapping;and the priori information, such as LiDAR device measurement accuracy and positional estimation accuracy, was introduced into the optimization algorithm of the de-distortion and registration to improve SLAM accuracy. Four 32m×32m forest sample plots were scanned by 32line LiDAR, and the raw data for the plot point clouds were processed by using the forest SLAM system. An indirect evaluation of the mapping accuracy of the system in forests was completed by comparing the tree position and DBH extracted from the point clouds with the reference data. The results showed that the mean error of the tree position estimates in the x-axis and y-axis directions were -0.004m and -0.011m, respectively. The root mean square error (RMSE) in the x-axis and y-axis directions were 0.081m and 0.083m, respectively. The deviation of the DBH estimates was 0.25cm (relative deviation of 1.18%) and the RMSE was 1.03cm (relative RMSE of 5.53%). And the estimates had higher accuracy compared with the LOAM system estimates. The results showed that the forest LiDAR SLAM system can be used for processing of the plot data scanned by multi-line LiDAR. It was a potential solution for accurate forest inventory.