高级检索
    李明泽, 于欣彤, 高元科, 范文义. 基于SAR极化分解与Landsat数据的森林生物量遥感估测[J]. 北京林业大学学报, 2018, 40(2): 1-10. DOI: 10.13332/j.1000-1522.20170284
    引用本文: 李明泽, 于欣彤, 高元科, 范文义. 基于SAR极化分解与Landsat数据的森林生物量遥感估测[J]. 北京林业大学学报, 2018, 40(2): 1-10. DOI: 10.13332/j.1000-1522.20170284
    Li Ming-ze, Yu Xin-tong, Gao Yuan-ke, Fan Wen-yi. Remote sensing quantification on forest biomass based on SAR polarization decomposition and Landsat data[J]. Journal of Beijing Forestry University, 2018, 40(2): 1-10. DOI: 10.13332/j.1000-1522.20170284
    Citation: Li Ming-ze, Yu Xin-tong, Gao Yuan-ke, Fan Wen-yi. Remote sensing quantification on forest biomass based on SAR polarization decomposition and Landsat data[J]. Journal of Beijing Forestry University, 2018, 40(2): 1-10. DOI: 10.13332/j.1000-1522.20170284

    基于SAR极化分解与Landsat数据的森林生物量遥感估测

    Remote sensing quantification on forest biomass based on SAR polarization decomposition and Landsat data

    • 摘要:
      目的森林生物量是评价森林生态系统结构、功能和生产力的重要指标之一,区域尺度上的森林生物量的准确估测对了解森林现状和科学经营森林具有重要指导意义。本文旨在利用SAR影像结合Landsat5 TM影像对区域尺度上的森林生物量进行定量估测。
      方法首先利用极化分解的方法对极化合成孔径雷达(SAR)数据进行处理获得45个极化分解参数,然后将45个极化分解参数与6个Landsat5 TM波段参数共51个参数作为自变量,森林生物量W作为因变量构建统计回归模型,最后利用最优模型反演研究区的森林生物量。
      结果使用两种方法进行模型构建:(1)逐步回归法,利用逐步回归进行变量筛选,选出2个参数构建模型,模型R2为0.534,拟合精度为67.51%,RMSE为43.21 t/hm2;(2)最优子集法,用Bootstrap法进行变量筛选,共筛选出9个参数,然后用这9个参数进行最优子集回归,获得511个选模型,然后选择出最优子集模型,并用交叉验证法对模型进行验证,最终选出的最优子集模型包含的参数为TM_band4、Neumann_delta_mod、Neumann_psi、TSVM_psi、TSVM_tau_m3,模型R2为0.768 2,拟合精度为88.32%,拟合RMSE为14.98 t/hm2,验证精度为86.21%,验证RMSE为19.14 t/hm2,CP指数为5.249 5,赤池信息量AIC为256.504 5。本文最终使用最优子集法获得的模型进行反演,获得研究区的森林生物量分布图。
      结论结果表明:全极化C波段SAR数据结合Landsat5 TM光学数据构建遥感信息模型可以准确反演森林生物量。

       

      Abstract:
      ObjectiveForest biomass is one of the important indexes to evaluate the structure, function and productivity of forest ecosystem, and accurate forest biomass estimation on regional scale has great significance in understanding the current forest status and scientific forest management. This study aims to quantify regional scale forest biomass through the polarimetric SAR and Landsat5 TM.
      MethodFirstly, SAR data was polarized by polarization decomposition. Then 51 parameters from 45 polarization decomposition parameters and 6 TM bands were used as predictor variables with forest biomass W as response variables, the best model was used in the research area finally.
      ResultTwo methods were implemented for model construction: (1) stepwise regression, the final model includes two variables with R2 of 0.534, predicting accuracy of 67.51% and RMSE of 43.21 t/ha; (2) Optimal subset method, Bootstrap was applied to select 9 parameters. Then, we got 511 models by optimal subset method and cross-validation was used for model validation. The final model had 5 parameters(TM_band4, Neumann_delta_mod, Neumann_psi, TSVM_psi, TSVM_tau_m3), R2 of 0.768 2, simulating accuracy of 88.32%, simulating RMSE of 14.98 t/ha, test accuracy of 86.21%, test RMSE of 19.14 t/ha, Mallows'Cp of 5.249 5 and AIC of 256.504 5 t/ha. We used optimal subset method to build forest biomass estimation model and acquired forest biomass distributing map.
      ConclusionThe results show that C band polarimetric SAR and optical Landsat5 TM data can get accurate estimation of forest biomass.

       

    /

    返回文章
    返回