Abstract:

Objective: In order to provide technical support for improving the efficiency and accuracy of forest resources investigation, this paper was implemented to explore an efficient method to map the mean height of forest(MHF) with few field plot survey data, airborne light direction and ranging(LiDAR) sampling data and ZY3 stereo imaging pairs covering the full target region. Method: In Guangxi Zhuang Autonomous Region, two subfarms(Jiepai and Dongsheng) of Gaofeng forest farm were selected as our study areas. Full covered airborne LiDAR and ZY3 stereo imaging pairs combined with few field plot data were collected in 2018. DEM(digital elevation model) extracted from airborne LiDAR was treated as a historical under canopy topography, and twelve strips sampled from the airborne LiDAR data were simulated as the real airborne LiDAR sampling flight areas. DEM, twelve strips of LiDAR data and field plot data were used in the following processes. First of all, the MHF for these sampled airborne LiDAR areas were produced with field plot data, which was used as reference data(Y) for modeling. After that, an independent variable(CHM_ZY3) was obtained by subtracting the DEM from the DSM(digital surface model) extracted from ZY3 stereo imaging pairs. Finally, MHF estimation results got from ordinary least squares(OLS), k-nearest neighbor(KNN) and regression Kriging(RK) models were compared, which were used to estimate MHF combined with CHM_ZY3 data. Result: It was showed that root mean square error(RMSE) of OLS and KNN was 1.88 m and 1.96 m, the estimate accuracy(EA) was 87.18% and 86.64%, respectively. The RMSE and EA of RK_KNN was 1.86 m and 87.32%, respectively, however these parameters for RK_OLS reached to 1.84 m and 87.42%, correspondingly. Conclusion: It could be concluded that all the 2 categories and 4 types of models tested can effectively estimate MHF, and among these models, RK based models(RK_OLS and RK_KNN) are better than non-spatial models(OLS and KNN), while RK_OLS has the best performance. With known under canopy topography information, synergy utilization of the few field plot data, ZY3 stereo imaging pairs(fully covering the target area) and airborne LiDAR strips(acquired by sampling the target area) can estimate the MHF efficiently and precisely. The method proposed can provide important supports for improving the efficiency and accuracy of forest resource inventory in the future.

Key words: light direction and ranging(LiDAR), ZY3, mean height of forest, regression Kriging(RK)