APPLICATION OF SPATIAL REGRESSION MODELS FOR FOREST BIOMASS ESTIMATION IN GUIZHOU PROVINCE, SOUTHWEST CHINA.

At the regional scale, many studies have been devoted to the construction of biomass models to improve the accuracy of regional biomass estimation, while only a few studies were carried out to investigate spatial influence. Therefore, the current research examined the spatial autocorrelations as well as variations between forest biomass and forest variables using 419 forest biomass plots sampled in the Guizhou Province in the year 2010. Besides, 4 global models, including the ordinary least squares model (OLS), linear mixed model (LMM), spatial lag model (SLM), spatial error models (SEM), together with geographically weighted regression model (GWR, the local model), were fitted to the associations of forest biomass with basal area, height and age of the stand. As suggested by our findings, distinct spatial autocorrelations as well as variations exist between forest biomass and these variables. OLS is not appropriate for modeling. SLM and SEM efficiently accounted for the spatial autocorrelations within model residual; however, they were unable to manage spatial heterogeneities. However, LMM and GWR, which had combined spatial variations as well as dependence during the modeling process, performed well in data fitting and response variable predicting. Of them, GWR reduced spatial heterogeneity to a greater extent than LMM. [ABSTRACT FROM AUTHOR]