Objective  This study aims to establish a remote sensing based parametric model for estimation of aboveground biomass (AGB) of Pinus densata for permanent sample plots, which can be used for rapid and accurate biomass estimation in the future with previous sample plots, or obtaining biomass quickly with less field work.

  Method  Based on the change of remote sensing images and permanent sample plots, linear mixed model was used to improve the accuracy of biomass estimation. Based on the permanent sample plots in the 7 survey years of 1987, 1992, 1997, 2002, 2007, 2012, 2017 from the national forest inventory and corresponding years of Landsat TM and OLI images, firstly the images were preprocessed including radiometric correction, atmospheric correction, geometric correction and topographic correction. The original bands, ratio factors, vegetation indices, image enhancement information, textures, fraction after spectral mixture analysis, leaf area index were extracted. Then, the changes of remote sensing spectral variables were derived. According to the distribution of Pinus densata from the forest management inventory, the topographic factors were selected as the fixed and random effects for the linear mixed model. The multiple linear regression, non-linear regression, geographically weighted regression, and linear mixed model were used to establish the static models of the AGB estimation for Pinus densata. The change models with and without tree height participation were developed based on the change of remote sensing spectral variables. Finally, the different modeling and validation results were compared and validated, and the optimal results were selected as the estimation model and validated.

  Result  (1) Comparing the static data for modeling and validation, the linear mixed model with the plot number as fixed effect and the slope grade as random effect got the highest R² of 0.75. The prediction result showed that either using the remaining 20 training datasets or the observed data in the year of 2017 for validation, the prediction accuracy was low. (2) Comparing the change data for modeling and validation, the linear mixed model with the plot number as fixed factor, slope grade as random factor and remote sensing change factors as independent variables performed the best with R² of 0.70, the predicted P value was (68.86 ± 11.93)%. When increasing the change of average tree height, the fitting R² was 0.79, the P value was (73.39 ± 6.18)%. (3) The change model with or without the participation of tree height got a fitting and prediction accuracy of 80%, and its prediction accuracy reached the prediction accuracy of non-parametric models.

  Conclusion  The accuracy of fitting and prediction based on the change variables is improved compared with the static model. The accuracy for estimating AGB of Pinus densata has been greatly improved with linear mixed model, remote sensing and topographic factors. The developed model of estimating the AGB of Pinus densata with remote sensing change factors has effectively compensated the deficiency of the static optical images, and it can be used for estimation of other years after validation.