基于HLS数据的森林蓄积量遥感反演

Remote sensing inversion of forest volume based on HLS data

通过对比陆地卫星8号(Landsat 8)与协同陆地卫星和哨兵2号(HLS)数据在波段信息及森林蓄积量建模效果的差异,探索HLS数据源在森林蓄积量反演中的应用潜力。以内蒙古自治区旺业甸林场为研究区,以Landsat 8和HLS为数据源提取遥感变量,结合森林蓄积量样地实测数据,利用线性逐步回归(LSR)和逐步随机森林(SRF)进行特征变量筛选,分别构建线性和非线性模型开展森林蓄积量反演,并进行精度验证及对比分析。结果表明:HLS影像中的7个波段与Landsat 8影像各对应波段均表现为极显著相关,相关系数绝对值均大于0.7(P<0.01),在波段信息上具有较好的一致性;基于Landsat 8及HLS影像数据构建多元线性回归和随机森林模型,其最优模型决定系数(R^(2))分别为0.55和0.54,均方根误差(RMSE)分别为65.66和68.15 m^(3)·hm^(-2),结果无显著性差异(P>0.05)。利用HLS影像进行森林蓄积量估测,可得到与Landsat 8相似的效果,且HLS数据无需进行繁琐的预处理,能显著提高森林蓄积量估测效率,为大尺度森林蓄积量遥感反演和森林资源监测提供参考。

By comparing the differences between Landsat 8 and Harmonized Landsat and Sentinel-2(HLS)data in terms of band information and forest volume modeling,the application potential of the HLS source in forest volume inversion was explored.Considering Wangyedian Forest Farm in the Inner Mongolia Autonomous Region as the study area,Landsat 8 and HLS images were used as data sources to extract remote sensing variables.In combination with the measured data of forest volume sample plots,linear stepwise regression and stepwise random forest were used to screen characteristic variables.Linear and nonlinear models were constructed to perform forest volume inversion,followed by accuracy verification and comparative analysis.The results showed that the seven bands in HLS were highly significantly correlated with each corresponding band of the Landsat 8 image,and the absolute values of the correlation coefficients were all greater than 0.7(P<0.01),indicating good consistency in terms of band information.Based on the Landsat 8 and HLS images,multiple linear regression and random forest models were constructed.The R^(2)values of these optimized models were 0.55 and 0.54,while the RMSE values were 65.66 and 68.15 m^(3)·hm^(-2),respectively,indicating no significant differences(P>0.05).Therefore,using HLS imagery for forest volume estimation can yield results similar to those of Landsat 8;further,HLS data do not require tedious preprocessing,which can significantly improve the efficiency of forest volume estimation.Thus,this study provides a reference for large-scale forest volume remote sensing inversion and forest resource monitoring.