Estimating mangrove forest coverage based on unmanned aerial vehicle (UAV) LiDAR point cloud data

Objective This paper aims to quickly and accurately obtain mangrove fractional canopy coverage based on unmanned aerial vehicle (UAV) LiDAR data, which would provide important reference for evaluating the effectiveness of mangrove ecological restoration.

Method The location of this research is on Lingtou Island in Taiping Town, Zhanjiang City of Guangdong Province, southern China. Based on ULS data and ground field data of 40 sample plots, linear regression was used to fit the measured and estimated fractional coverage of mangrove, and the determination coefficient (R²), root mean square error (RMSE) and estimation accuracy (E_A) were calculated. The estimation accuracies of mangrove fractional coverage through four different algorithms based on the first return proportionality model (FRRM), the all return proportionality model (ARRM), pulse return intensity proportionality model (PRIRM), and canopy height model (CHM) were compared. The correlations between sample site fractional coverage, sample site laser point cloud density, sample site LiDAR height characteristic variable and fractional coverage estimation error were analyzed. Finally, the optimal coverage estimation model with the best accuracy was selected to estimate mangrove coverage in the study area, and mapping was carried out.

Result (1) The estimation accuracy of mangrove fractional coverage based on FRRM model was the highest (R² = 0.970 1, RMSE = 0.032 5, E_A = 93.01%), and the estimation error was the lowest with an average underestimation of 1.04%. The second was the algorithm based on ARRM model (R² = 0.977 4, RMSE= 0.033 6, E_A = 92.58%), the third was the algorithm based on CHM model (R² = 0.945 0, RMSE = 0.044 0, E_A = 90.54%), and the accuracy of PRIRM model was the lowest (R² = 0.9509, RMSE = 0.0610, E_A = 88.17%). (2) The estimation errors of PRIRM model were significantly negatively correlated with both coverage and height characterization variables, and were generally overestimated. Meanwhile, there was no significant correlation between sample site fractional coverage and estimation error of algorithm based on FRRM, ARRM and CHM models. (3) The results of LiDAR sampling sensitivity analysis showed that a raster size of 3 m was most suitable as a raster unit for mapping the coverage of the study area.

Conclusion The estimation accuracy of mangrove fractional coverage based on the four models is high, and the estimation accuracy of fractional coverage based on FRRM model could be the highest, the estimation results are reliable, which can provide support for the scientific management and ecological restoration of mangrove forests on Lingtou Island, Guangdong Province of southern China.