Pi-SAR极化数据与K分布指数估算森林生物量与实验验证

Estimation of Forest Biomass and Experimental Validation Based on Pi-SAR Polarimetric Data and K-Distribution Index

用2002年和2003年日本Pi-SAR全极化数据,研究日本北海道苫小牧森林地区的森林生物量。雷达后向散射系数随森林生物量的增大而增大并迅速达到饱和,L波段雷达数据饱和点约为40 t/hm^2,X波段仅约为20t/hm^2。在SAR数据统计分布中,K分布的指数参数在饱和点以上仍随生物量的增大而增大,并且HV极化方式时相关性最高。根据交叉极化数据K分布的指数参数与森林生物量的关系,本文估算了23个观测点的森林生物量,结果表明平均准确率为85%。因此该算法可以作为一种新的估算森林生物量的手段。

Employing Pi-SAR polarimetric data acquired in 2002 and 2003,forest biomass estimation approach is stud- ied on Tomakomai forests located in Hokkaido,Japan.The purpose of this project is to develop effective approach for esti- mating forest biomass.The ground truth data of 19 test sites are in hand.In this test sites,one sample stand of 20m×20m are selected and tree height,age,basal area,diameter of breast height and tree species are measured,the biomass is then calculated.The conventional Radar cross section(RCS)method is first investigated.It is found that RCS increases with biomass and becomes saturated rapidly,under the situation of this paper.That is:the L-band RCS saturation levels are found approximately to the biomass of 40t/hm2,with the tree age of 30 years,the tree height of 8m,and the basal area of 30 m^2/hm^2.The RCS saturates at 20t/hm^2 for X-band data.Therefore,forest biomass beyond saturation level cannot be estimated utilizing RCS.To search the quantitative relation between high-resolution SAR data and forest parameters,sta- tistical analysis approach is utilized.The probability density function of image amplitude is then investigated,and among different distribution including Rayleigh,log-normal,Weibull and K-distributions,the K-distribution is found to fit best to the L-band data of all polarizations according to the Akaike information criterion(AIC).The relations between K-distribu- tion index and different tree parameters including biomass,tree age,height,basal area,are investigated.It is found that the tree biomass correlates best with the index parameter.Moreover,K-distribution index increases with biomass beyond RCS saturation level,and the highest correlation coefficient is obtained at cross-polarization.The regression model is de- veloped between K-distribution index and forest biomass at cross-polarization based on 19 test sites data.In August and September of 2005,we further collected ground truth data of 23 test sites.Based on the relation of K-distribution index of cross-polarization and forest biomass,the biomass estimation is made for the 23 test sites.The comparison of estimated bi- omass and measured ground truth data rerifies that the average accuracy of the estimation reaches 85%.It is concluded that,at least for the Hokkaido forests,this empirical model is an effective and superior way of estimating forest biomass from polarimetric SAR data compared with the conventional RCS model.