| Forest attribute maps can provide rich information on forest resources and contribute to forest sustainable management policy.Airborne Laser Scanning,as a new technology,has been widely used in forest attribute estimation and mapping at various scales.In this study,a subtropical region of 2376,000 km~2 in southern China with rich species and complex landscape was selected as the study area.Based on airborne lidar data and plot measured data the vertical distribution map of laser point cloud and the continuous canopy vertical profile were obtained by fitting the frequency distribution of point cloud height and coverage,and the vertical structure of pine forest,Chinese fir forest and eucalyptus forest was classified by visual interpretation method.The rule-based exhaustive method was used to establish the multiplicative nonlinear regression model of stand volume,basal area,mean tree height and the diameter at breast height independent estimation for each vertical structure type.For pure pine forest and pure Chinese fir forest,the simultaneous equations of mean tree height,basal area and stand volume were established by using multivariate nonlinear error variable method,and were compared with independent estimation models.The results show that:1)Before classification,the relative root mean square error(r RMSE)of the estimation models of stand volume,basal area,mean tree height and the diameter at breast height of pine forest were 21.01%、19.04%、10.76%、20.57%,respectively,and that of Chinese fir forest were 22.02%、22.27%、13.73%、19.19%,respectively.In eucalyptus forest,19.13%、18.30%、9.08%、11.40%,and in broadleaf forest,36.19%,30.00%、16.43%、26.77%,respectively.2)After canopy vertical structure classification,the interpretation rate of Li DAR variables in the forest attribute estimation model is higher than that without classification,and the r RMSE of the model is lower than that of the non-classification.Calculated according to the weighted average of sample plots,The relative root mean square error(r RMSE)of the original vertical structure classification for pine forest decreased by 21.64%、16.93%、13.13%、14.75%,and that for Chinese fir forest decreased by 7.99%、18.32%、14.93%、3.88%,respectively.Eucalyptus forest decreased by 1.30%、1.66%、6.65%、3.86%,and broadleaf forest by 14.59%、14.06%、1.10%、21.28%,respectively.3)After clustering the original vertical structure types of each forest type into two types,the accuracy of stand volume,basal area,mean tree height and the diameter at breast height estimation models were improved to a certain extent compared with those without classification.According to the weighted average calculation of sample size,pine forest increased by 10.75%、5.93%、10.22%、10.73%,Chinese fir forest increased by 8.19%、17.27%、11.52%、3.88%,and broadleaf forest increased by 4.87%、5.54%、4.49%、5.55%.4)In the simultaneous equation group,the r RMSE of the stand volume,basal area and mean tree height models of Chinese fir and pine forests increased by no more than 5%compared with the independent estimation,and the average difference between the estimated values was less than±2.0%,but most of them had significant differences.5)The performance of each simultaneous equation group was very similar,and there was basically no significant difference in the estimation results.The simultaneous equation group with basal area and mean tree height as endogenous variables to estimate volume performed slightly better.The estimation of forest attributes based on stand vertical structure classification is helpful to improve the estimation accuracy of forest attributes in large-scale airborne lidar forest resources inventory and monitoring.The more the classification types are,the greater the improvement of estimation accuracy is.The simultaneous equations ensure the mathematical relationship between various forest attributes,which is in line with the theoretical basis of forest measurement and is more conducive to forest resource management and application. |
PDF Full Text Request