应用Landsat8 OLI和GBRT对高山松地上生物量的估测

Estimating Above Ground Biomass of Pinus densata Based on Landsat8 OLI and Gradient Boost Regression Tree

依据香格里拉市Landsat8 OLI提取的因子和外业调查的高山松样地地上生物量,利用梯度提升回归树(GBRT)建立了地上生物量与遥感因子的估测模型,并与随机森林、多元线性回归、偏最小二乘方法进行了对比。结果表明:纹理信息对生物量有重要影响,其中熵、相关性和Landsat8 OLI近红外波段的信息对生物量的影响最大;采用GBRT进行建模,当迭代次数大于200次时,偏差降低减缓,GBRT建模方法的精度评价指标(R2=0.96,rRMSE=8.80%,P=73.88%)均优于其他3个模型。应用Landsat数据进行森林地上生物量估测的不确定因素较多,GBRT可作为高山松及其他树种地上地上生物量遥感估测的另一新方法。

The Landsat8 OLI band indexes and the aboveground biomass（ AGB） of the Pinus densata field sampling plots in Shangri-la were used to build AGB estimation model with Gradient boost regression tree（ GBRT）. The methods of GBRT,multiple linear regression,partial least square regression and random forests were compared. Texture features were highly influenced the AGB; Entropy,Correlation and near infrared band of Landsat8 OLI were the most influenced AGB. With GBRT,when the iterations were more than 200 times,the deviation reduced gradually. Accuracy assessment with GBRT（R^2= 0.96,rRMSE = 8.80%,P = 73. 88%） performed better than other three methods. There are lots of uncertainties in estimating forest AGB with Landsat data,but GBRT would be a new method to estimate AGB of P. densata and other tree species.