Improving Forest Aboveground Biomass Estimation of <i>Pinus densata</i> Forest in Yunnan of Southwest China by Spatial Regression using Landsat 8 Images

• Main Authors: Guanglong Ou, Yanyu Lv, Hui Xu, Guangxing Wang
• Format: Article
• Language: English
• Published: MDPI AG 2019-11-01
• Series: Remote Sensing
• Subjects: aboveground biomass; spatial autocorrelation; spatial heterogeneity; spatial variability modeling; landsat 8-oli images; uncertainty; <i>pinus densata</i> forest
• Online Access: https://www.mdpi.com/2072-4292/11/23/2750

Uncertainties in forest aboveground biomass (AGB) estimates resulting from over- and underestimations using remote sensing data have been widely studied. The uncertainties may occur due to the spatial effects of the plot data. In this study, we collected AGB data from a total of 147 <i>Pinus densata</i> forest sample plots in Yunnan of southwestern China and analyzed the spatial effects on the estimation of AGB. An ordinary least squares (OLS) and four spatial regression methods were compared for the estimation using Landsat 8-OLI images. Through the spatial analysis of AGB and residuals of model predictions, it was found that the spatial autocorrelation and heterogeneity of the plot data could not be ignored. Compared with the OLS, the impact of the spatial effects on AGB estimation could be reduced slightly by the spatial lag model (SLM) and the spatial error model (SEM) and greatly reduced by the linear mixed effects model (LMM) and geographically weighted regression (GWR) based on the distributions of prediction residuals, global Moran&#8217;s I, and <i>Z</i> score. The spatial regression models had better performance for model fitting and prediction because of the reduction in overestimations and underestimations for the forests with small and large AGB values, respectively. However, the reductions in the overestimations and underestimations varied depending on the spatial regression models. The GWR provided the most accurate predictions with the largest <i>R</i>² (0.665), the smallest root mean square error (34.507), and mean relative error (&#8722;9.070%) by greatly reducing the AGB interval for overestimations occurring and significantly increasing the threshold of AGB from 150 Mg/ha to 200 Mg/ha for underestimations. Thus, GWR offered the greatest potential of improving the estimation of <i>Pinus densata</i> forest AGB in Yunnan of southwestern China.