基于混合效应模型的曲靖市云南松林地上生物量遥感估测

Remote Sensing Estimation on Aboveground Biomass for Pinus yunnanensis Forests in Qujing City Using Mixed Effect Models

以曲靖市云南松林为研究对象,基于2016年曲靖市二类调查数据以及同时期Landsat 8 OLI遥感影像,利用随机选取的小班样地提取遥感因子统计值建立数据集,基于不同曲线拟合定量研究曲靖市云南松林生物量估测的光饱和值;并以线性逐步回归模型为基本模型,考虑区域和龄组随机效应建立云南松林生物量遥感估测模型,以期减小生物量遥感估测中光饱和引起的估测误差。结果表明:利用三次模型拟合研究区云南松生物量饱和值为167 t/ha;不同效应水平的混合模型的拟合精度均优于一般逐步回归模型。在独立性样本检验中混合效应模型的预估精度(91.556%)要高于一般逐步回归模型的预估精度(83.826%);从生物量分段残差检验结果与研究区生物量反演结果上看,混合效应模型相较于一般逐步回归模型有着更大的估测范围,在一定程度上能够解决高值低估和低值高估问题,减小光学遥感影像数据存在数据饱和的影响。

Taking the Pinus yunnanensis in Qujing City,Yunnan Province as the research object,based on forest management inventory(FMI)data in 2016 and the simultaneous Landsat 8 OLI remote sensing images,subcompartment points were created randomly.Then,datasets were built through extracting subcompartment's ststistics values based on remote sensing indexes.Quantitative study of the light saturation value of P.yunnanensis biomass estimation based on different curve fitting.A linear stepwise regression model was established,and based on this model,the mixed effects models considering region and age group random effect were established to reduce the estimation error caused by light saturation in biomass remote sensing estimation.The results showed that the biomass saturation value of P.yunnanensis forests was 167 t/ha in the study area.The fitting accuracy of the mixed effects models are better than the basic linear stepwise regression model,and the prediction accuracy of the mixed-effects model with both random effects in the independent sample test(p=91.556%)is higher than that of the basic stepwise regression model(p=83.826%).The results of biomass segmentation residual test and the biomass inversion results of the study area showed that the mixed-effects model has a larger estimation range than the general stepwise regression model,and the mixed effects models can reduce the uncertainties from the high-value underestimation and low-value overestimation to a certain extent,and decrease the impact of data saturation on optical remote sensing image data.