Assessing Effect of Ground Sampling Intensity on Estimating Forest Quantitative Characteristics Using Fusion of Airborne Laser Scanner Data and UltraCam-D Images

In this study we evaluated the effect of ground sampling intensity on estimation of stand volume, tree density and basal area using random forest, support Vector regression and k-NN algorithms for the part of Shast Kalate of Gorgan. We applied a systematic random sampling method to collect field data with 150×200 meter network (3.33% intensity sampling). So that 308 circular with 17.84 (0.1 ha) meters radius plot were measured in study area. In addition to the data collected, in compartment number of 16 and 21, we applied a systematic random sampling method to collect field data with 75×100 meter network. So that 134 plot circular with 17.84 (0.1 ha) meters radius were measured in 2 compartments. After removal of all outliers and creating DTM and DSM, all height and density related metrics of first and last pulse were produced. Also, after orthorectification digital aerial images, all texture measures were produced.The results of comparison of intensity sampling in stand volume, tree density and basal area estimation using fusion Lidar data and Digital aerial images showed that with increasing intensity sampling, RMSE% is reduced and with reducing intensity sampling, the RMSE% is increased. Although the results of 3.33%, 1.66% and 0.83% intensity sampling were not very different. According to low difference in RMSE% of the resulting all intensity sampling (3.33%, 1.66% and 0.83%), the 0.83% intensity sampling can be used to estimate the stand volume, tree density and basal area. Therefore, there is a possibility of the estimation of stand volume, tree density and basal area using Laser scanner data and UltraCam-D images with minimum cost, reasonable accuracy and less plots compared to 3.33% intensity.
In this study we evaluated the effect of ground sampling intensity on estimation of stand volume, tree density and basal area using random forest, support Vector regression and k-NN algorithms for the part of Shast Kalate of Gorgan. We applied a systematic random sampling method to collect field data with 150×200 meter network (3.33% intensity sampling). So that 308 circular with 17.84 (0.1 ha) meters radius plot were measured in study area. In addition to the data collected, in compartment number of 16 and 21, we applied a systematic random sampling method to collect field data with 75×100 meter network. So that 134 plot circular with 17.84 (0.1 ha) meters radius were measured in 2 compartments. After removal of all outliers and creating DTM and DSM, all height and density related metrics of first and last pulse were produced. Also, after orthorectification digital aerial images, all texture measures were produced.The results of comparison of intensity sampling in stand volume, tree density and basal area estimation using fusion Lidar data and Digital aerial images showed that with increasing intensity sampling, RMSE% is reduced and with reducing intensity sampling, the RMSE% is increased. Although the results of 3.33%, 1.66% and 0.83% intensity sampling were not very different. According to low difference in RMSE% of the resulting all intensity sampling (3.33%, 1.66% and 0.83%), the 0.83% intensity sampling can be used to estimate the stand volume, tree density and basal area. Therefore, there is a possibility of the estimation of stand volume, tree density and basal area using Laser scanner data and UltraCam-D images with minimum cost, reasonable accuracy and less plots compared to 3.33% intensity.