基于随机森林和蒙特卡洛的高山松地上碳储量估测及不确定性分析

韩雪莲; 肖庆琳; 张加龙; 曹军; 黄屹杰; 王飞平

doi:10.12403/j.1001-1498.20220603

基于随机森林和蒙特卡洛的高山松地上碳储量估测及不确定性分析

Estimation and Uncertainty Analysis of Aboveground Carbon Storage of Pinus densata based on Random Forests and Monte Carlo

摘要

摘要:
目的为了解决碳储量的不确定性受单一类型变量的影响，而忽略了残差变异导致模型不确定性的问题，研究不同类型变量对碳储量估测模型的不确定性。

方法以香格里拉市高山松为研究对象，采用随机森林联立蒙特卡洛（RF-MC），基于不同变量组合建立回归模型，判断各模型的不确定性。

结果表明：（1）光谱和样地数据直接参与RF-MC模型的估测精度及不确定性最差。（2）在上一步的基础上引入纹理特征的模型预测结果优于引入DEM。（3）同时引入DEM和纹理特征的RF-MC模型效果最优（R²=0.892，RMSE=5.539 t·hm⁻²，MAE = 4.319 t·hm⁻², rRMSE=18.7%）。模型拟合度提高了0.343，模型的不确定性降低了19.43%。

结论基于多特征的RF-MC方法在碳储量估测中效果较好，不同类型的变量对碳储量估测精度及不确定性有一定的影响。

Abstract:
Objective To study the model uncertainty induced by different types of variables on carbon storage estimation.

Method Taking as the research object, Random forest joint Monte Carlo (RF-MC) was used to establish a carbon storage regression model of Shangri-La Pinus densata based on different variable combinations to determine the uncertainty of each model.

Result (1) Spectral and plot data directly contributed to the estimation accuracy and uncertainty of the RF-MC model. (2) The model prediction introducing texture features was better than introducing DEM. (3) The RF-MC model that introduced both DEM and texture features performed the best (R²=0.892, RMSE=5.539 t·hm⁻², MAE = 4.319 t·hm⁻², rRMSE=18.7%). The model fit improved by 0.343 and the uncertainty of the model decreased by 19.43%.

Conclusion The RF-MC method based on multiple characteristics performs well on carbon storage estimation. Different types of variables have certain impact on the accuracy and uncertainty of carbon storage estimation.

HTML全文

参考文献(42)

施引文献

资源附件(0)