Estimation Of Forest Parameters By Using PALSAR Data

Forest parameters are indispensable measurements in forestry investigation,including tree height,diameter at breast height(DBH),biomass,leaf area index and volume,etc.The traditional forest survey is featured with strong labor intensity and low efficiency,so it is difficult to investigate the parameters of large area forest.The visible light remote sensing technology plays an important role in forest investigation and forest parameter inversion.However,the optical remote sensing,which used in the estimation of forest parameters is seriously affected by the weather condition;at the same time,the optical remote sensing inversion can only be carried out on the planewithout detecting vertical structure of forest accurately.This paper selects Purple Mountain National Forest Park as the research object,which based on the data of two synthetic aperture radar data in 2011 and 2015,and takes advantages of six estimation models: Multiple linear regression,Artificial neural network,K-nearest neighbor algorithm,Decision and regression tree,Bagging algorithm and Random forest;Tree height,DBH and volume parameters of the forest in the studied area were estimated.This paper analyzes the distribution and growth conditions of Purple mountain in 2011 and 2015 were analyzed,and the forest parameters and the dynamic changes of forest growth in the last 5 years were analyzed.The results show that: the color characteristics of water body,building land and bare land,which can be seen from the polarization characteristic images,obtained from the four polarization target decomposition methods,are obvious and easy to distinguish.But the color characteristics of coniferous forest and broad-leaved forest are not obvious and difficult to distinguish.Cloude decomposition is superior to the other three decomposition methods for coniferous and broad-leaved forest.In the six estimation models,which can be seen from the accuracy of six Forest Parameter Estimation Models,the highest accuracy is random forest model for the average DBH,average tree height and volume of three forest parameters,but Multivariate linear regression model is low;in the three forest parameters,the average tree height estimation accuracy is the highest and the average breast diameter is the lowest.By calculating and analyzing the mean decrease accuracy and the mean decrease gini of polarization eigenvalue,terrain factor,human disturbance factor and radar vegetation index,the radar echo scattering characteristics and terrain factors are important environmental variables that affect the forest parameters in the study area.It can be resulted that the ordinary Kriging interpolation and Natural Neighbor interpolation precision are best,while the inverse distance weighted interpolation has general precision and trend surface interpolation has the lowest by comparison of four interpolation accuracy by using the trend surface interpolation,inverse distance weighted interpolation,ordinary Kriging interpolation and Natural Neighbor interpolation on the inversion results of forest parameters.Kriging interpolation and Natural Neighbor interpolation method are better than other interpolation methods for samples with large and spatially distributed data.It can be seen from the spatial distribution of forest parameters: average DBH,average tree height and volume per unit area of high stand are in the upper and middle slope of the north and south slope with high elevation and steep slope of the study area,while the medium stand is in the lower and lower slopes of the north and south slopes of the park and the lowest stand is located in the water,the construction land and the dominant area of the lawn,which are in the outer rim of Purple Mountain.Meanwhile,the zonal distribution pattern was gradually reduced from inside to outside in general.It can be seen from the spatial distribution map of forest parameters in 2011 and 2015: the average DBH,average tree height,unit stock volume increased in 2015,and the average DBH,average tree height and unit stock volume of broad-leaved forest were higher in high altitude area,however,the minimum average DBH(<5cm),the minimum average tree height(<4m)and the minimum unit stock volume(<30m3/hm2)increased significantly due to the transformation of the tourism landscape in the study area,the repair and construction of the building.