基于无人机和卫星遥感的胡杨林地上生物量估算

基于光谱信息的森林地上生物量遥感模型多存在精度不高的问题，如何更准确地获取森林地上生物量是遥感领域的研究热点。该研究以位于塔里木河下游的河岸胡杨林为例，探讨在无人机摄影测量技术支持下，使用高分辨率卫星遥感技术，通过面向对象影像分析和回归分析等技术，获取区域尺度下胡杨冠幅、树高和密度等森林结构参数，在此基础上，通过生长方程计算得到区域尺度森林地上生物量。在30、50、100和250 m 4个空间尺度上，与无人机数据的估算结果相比，高分辨率卫星遥感数据的地上生物量估算结果高22%～26%，其误差主要来自于树冠生物量部分。随着空间尺度增大，基于卫星遥感的地上生物量回归模型R2也随之增大，其中在100 m尺度上，地上生物量回归模型R2为0.851，表明使用高分辨率卫星遥感技术可以在较大的区域尺度上获得较高的森林地上生物量估算精度。地上生物量回归模型的标准化系数分析表明，对森林地上生物量估算精度影响最大的因素是密度和树高，冠幅影响最小，并且随着空间尺度增大，密度的影响有增加趋势，树高的影响有减少趋势。研究结果可为使用无人机和卫星遥感技术研究森林地上生物量提供参考。

Estimation of above ground biomass of Populus euphratica forest using UAV and satellite remote sensing

Forest biomass is the main component of terrestrial carbon sequestration, to serve as an important indicator of forest ecological function.Currently, there is still some limitations on the above ground biomass (AGB) of forestusing spectral informationfrom remote sensing technology. In this study, taking populus euphratica forest in the lower reaches of Tarim River as an example, theunmanned aerial vehicle (UAV) lowaltitude remote sensing and very high-resolution (VHR) satellite remote sensing wereused to estimate theforest AGB using forest structure information. Some more advanced UAV and image segmentation techniques were used to improve the accuracy of crown diameter, thereby to improve the accuracy of AGB estimation in the future. The AGB of populus euphratica was divided into trunk biomass and crown biomass. An allometric equation was used to calculatewith the parameters of tree height, diameter at breast jeight (DBH), and crowndiameter. The actual procedure was as follows: Digital surface model (DSM) and digital terrain model (DEM) were firstly obtained using UAV oblique photogrammetry and geographic information system(GIS) interpolation,together with the canopy height model (CHM). Secondly, anobject-oriented image analysis(OBIA) was used to acquire the tree height and crown-diameter. Finally, an allometric equation was used to calculatethe AGB by UAV measured data. The VHR WorldView-2 (WV2) normalized difference vegetation index(NDVI) image was calculatedby the OBIA and GIS overlay technologies, thereby to extractthe crown-diameter asresult.Specifically, the WV2 tree height was obtained from the regression model that built between 32 spectral relectance, texture, general features, and UAV-measured tree height. The AGB by WV2 measured was calculated uisngan allometric equation. Acomparison of UAV- and field-measured datashowedthat: The coefficients of determination (R2) of crown diameter, height, density, and AGB were 0.783, 0.866, 0.941, and 0.816, respectively.A high goodness-of-fit was also proved that the UAV-measurement can be expected to replace the field-measurent in plot size. Tree height fromthe WV2-measured was overestimated by 2.2%-3.2%, resulting in the trunk biomass was higher by 10%-13%, compared with theUAV-measured data.The crown diameter of WV2-measured was significantly overestimated by 27%-30%, resulting in the canopy biomass was overestimated by 58%-71%.Therefore, the density was under estimated by 1.8%-6.5%. The AGB of WV2-measured was overestimated by 22%-26%, compared with the UAV-measured data, wherethe error mainly came from the canopy biomass. Acomparison of WV2- and UAV-measured data on the four scale grid size of 30 to 250mshowedthatthe R2 of crown diameter, height, density, and AGB increasedwith the increasing of statistical grid size, whereas, the R2 of AGB was 0.851 at the scale of 1 hectare, which was usually used as a AGB statistical standard size. The forest structure information can be obtained by the VHR remote sensing through the OBIA with the support of UAV-measured data, and a good AGB accuracy can be obtained on a coarse scale. The linear regression models were establishedbetween AGB and crown diameter, height, density obtained by the UAV- and WV2-measured data.The coefficients of tree density and tree height were larger than those of crown diameter, indicating that tree density and tree height were the most important factors affecting the AGB on four scale, while the crown diameter has the least effect. There was an increasing trend in the influence of density, whereas a decreasing trendin the effect of tree height,with the increase of statistical grid size.