Multi-scale regional forest carbon density estimation based on regression and sequential Gaussian co-simulation

Abstract

By applying nonlinear regression of a unary cubic equation and sequential Gaussian co-simulation to Forest Inventory (plot) data in Xianju county, Zhejiang, from 2008, and Landsat TM image data collected in the same region in 2007, this research estimated the above-ground forest carbon density and its distributions at 30m×30m and 270m×270m resolutions, and analyzed the results comparatively. The results showed that the above-ground forest carbon density of Xianju county was continuously distributed, and was surrounded by high carbon density forestland, and the majority of the intermediate region was filled with low carbon density non-forestland. Using the random sampling method, the total carbon estimate is 5,289,437.11Mg. At 30m×30m resolution, with nonlinear regression of a unary cubic equation, the total carbon is 5,246,749.81Mg, and the R2 of the model is 0.1353. At the same scale, with sequential Gaussian co-simulation, the total carbon is 5,692,875.69Mg, and the R2 of the model is 0.6203. Compared with the results in the 270m×270m resolution, the former total carbon amount is larger, the range of distribution is wider, and the model's precision is higher. Comparing the two methods, the results estimated by the sequential Gaussian co-simulation are better than those of the unary cubic nonlinear regression. The result of sequential Gaussian co-simulation, which considers the spatial distribution of carbon density, is closer to that estimated from plot data, and better represents the continuous change of the carbon distribution.