A linear mixed-effects model of biomass and volume of trees using Landsat ETM+ images

Using forest inventory data and Landsat ETM+ data, linear fixed-effects models and linear mixed-effects models are developed based on the allometric growth model. The surface area of the normalized difference vegetation index (NDVIsa) is developed from the triangulated irregular network (TIN) with the aid of image-processing and the three-dimensional analysis extensions of Environmental Systems Research Institute's ArcView GIS software. The NDVIsa is used as the predictor, and it implies the area of trees for both site and area. Linear fixed-effects and linear mixed-effects models based on the allometric growth model are developed to fit the relationships between either biomass or volume and NDVIsa. Linear mixed-effects model with both intercept and slope having random-effects best fits the data, and the relatively high R2 (about 0.57) was achieved. This linear mixed-effects model is significantly different from the linear fixed-effects model and the linear mixed-effects model with only intercept having random-effects. The best fitted linear mixed-effects model discovers different spatial characteristics of biomass and volume of trees across the whole state of Georgia. The Piedmont ecoregion has positive allometric characteristics, the Upper Coastal Plain has negative allometry, whereas the other three ecoregions have isometric characteristics. At last, the linear mixed-effects models were compared with the extreme situation, fitting linear fixed-effects models within each region. Model diagnostics indicated that the linear mixed-effects models were the best modeling approach.