Using remote sensing methods to accurately estimate aboveground biomass carbon stock (ABGCS) in forest canopy layers and light saturation value of carbon storage, aiming to replace the cumbersome procedures of traditional large-area surveys, providing references and basis for carbon storage estimation, and improving the efficiency of sustainable forest management. In this study, the ABGCS in Jiayin County, Yichun City, Heilongjiang Province in 2017 was selected as the research object. Landsat 8 OLI remote sensing images and forest resource two-class survey data were used to construct parameter models of stepwise multiple regression model (SMR), non-parameter models of BP neural network model (BP-NN), random forest model (RF), support vector regression model (SVR) to estimate and reverse the spatial distribution of ABGCS in Jiayin County. The research results showed that the estimation accuracy of non-parameter models was significantly higher than that of parameter models. Among them, the fitting accuracy of the three non-parameter models (BP-NN, RF, SVR) was increased by 25.0%, 12.2%, and 7.3%, respectively, compared with the parameter model (SMR). By comprehensive comparison of the evaluation indexes of the four models in ten-fold cross-validation, the performance of the models was analyzed: BP-NN>RF>SVR>SMR, among which the BP-NN model fitted the largest R2 (0.785) and the smallest RMSE (3.572 t/hm2), MSE (12.757 t/hm2), MAE (2.687 t/hm2). From the perspective of carbon storage residual segmentation test results, all four models exhibited varying degrees of overestimation and underestimation of carbon storage. The BP-NN model had the smallest ME and MRE values in each carbon storage segment, indicating strong generalization ability. The light saturation value of ABGCS was determined to be 63.056 t/hm2 using a cubic model, which was close to the predicted ABGCS light saturation value by BP-NN (64.232 t/hm2). Therefore, the BP-NN model has a relatively ideal effect in estimating ABGCS in Jiayin County, providing important basis for dynamic monitoring and research of forest carbon storage.
使用R语言中的随机森林包进行装袋法估计,首先确定决策树的数目。将参数中的决策树节点数设置为11(mtry=11),表示在该例中使用全部的特征变量,结果显示,随机森林模型函数默认估计500棵决策树,在每个节点均使用全部的11个变量,根据袋外观测值计算的袋外数据(Out of Band,OOB)为22.71;而准R2=0.659,使用plot画出袋外误差曲线图,如图4所示。由图4可知,随着决策树数目的增加,袋外误差呈现下降趋势,当决策数目大于300时,误差就趋于平稳,这时继续增大决策树数目,也不会使之下降。因此,决策树的数目大于300即可。
ZHENGJ J, HUANGS Y, JIAX,et al.Spatial variation and controlling factors of temperature sensitivity of soil respiration in forest ecosystems across China[J].Chinese Journal of Plant Ecology,2020,44(6):687-698.
[3]
BALIMAL H, KOUAMEF N, BAYENP,et al.Influence of climate and forest attributes on aboveground carbon storage in Burkina Faso,West Africa[J].Environmental Challenges,2021,4:100-123.
[4]
BONANG B.Forests and climate change:forcings,feedbacks,and the climate benefits of forests[J].Science,2008,320(5882):1444-1449.
SHUY, GUOJ Y, ZHOUM,et al.Carbon measurement parameters of larchplantation in the Greater Xing′an Mountains based on IPCC method[J].Research at Temperate Forestry,2022,5(1):30-35,47.
MAOX G, WANGJ W, FANW Y.Spatiotemporal variation analysis of forest biomass based on remote sensing and ground statistics[J].Journal of Beijing Forestry University,2016,38(2):10-19.
[9]
穆喜云.森林地上生物量遥感估测方法研究[D].呼和浩特:内蒙古农业大学,2015.
[10]
MUX Y.Research on the remote sensing estimation method of forest aboveground biomass[D].Hohhot:Inner Mongolia Agricultural University,2015.
LID R, WANGC W, HUY M,et al.General review on remote sensing-based biomass estimation[J].Geomatics and Information Science of Wuhan University,2012,37(6):631-635.
[13]
OUG L, LVY Y, XUH,et al.Improving forest aboveground biomass estimation of pinus densata forest in Yunnan of southwest China by spatial regression using Landsat 8 images[J].Remote Sensing,2019,11(23):2750.
[14]
CHENQ, MCROBERTSR E, WANGC,et al.Forest aboveground biomass mapping and estimation across multiple spatial scales using model-based inference[J].Remote Sensing of Environment,2016,184:350-360.
[15]
ZHAOP P, LUD S, WANGG X,et al.Examining spec-tral reflectance saturation in landsat imagery and corresponding solutions to improve forest aboveground biomass estimation[J].Remote Sensing,2016,8(6):469.
[16]
FANGJ Y, CHENA P, PENGC H,et al.Changes in forest biomass carbon storage in China between 1949 and 1998[J].Science,2001,292(5525):2320-2322.
[17]
ZHANGC H, JUW M, CHENJ M,et al.China′s forest biomass carbon sink based on seven inventories from 1973 to 2008[J].Climatic Change,2013,118(3/4):933-948.
[18]
肖兴威.中国森林资源清查[M].北京:中国林业出版社,2005.
[19]
XIAOX W.Forest resources inventory in China[M].Beijing:China Forestry Press,2005.
[20]
TUIAD, VERRELSTJ, ALANSOL,et al.Multioutput support vector regression for remote sensing biophysical parameter estimation[J].IEEE Geoscience and Remote Sensing Letters,2011,8(4):804-808.
[21]
ZHAOY, WANGK.Fast cross validation for regularized extreme learning machine[J].Journal of Systems Engineering and Electronics,2014,25(5):895-900.
[22]
DONGL J, LIX B, PENGK.Prediction of rockburst classification using random forest[J].Transaction of Nonferrous Metals Society of China,2013,23(2):472-477.
LINC D, PENGG L.Application of random forest on selecting evaluation index system for enterprise credit assessment[J].Journal of Xiamen University:Natural Science Edition,2007,46(2):199-203.
LIP, ZENGL K, SHUIA Z,et al.Design of the BP neural network prediction system based on MATLAB[J].Computer Applications and Software,2008,25(4):149-150.
SHENH Y, WANGZ X, GAOC Y,et al.Determining the number of BP neural network hidden layer units[J].Journal of Tianjin University of Technology,2008,24(5):13-15.
SUNX L.A remote sensing estimation model of Alpine pine forest based on Landsat8-OLI[D].Kunming:Southwest Forestry University,2016.
[35]
ZHANGX, JIAW, SUNY,et al.Simulation of spatial and temporal distribution of forest carbon stocks in long time series-based on remote sensing and deep learning[J].Forests,2023,14(3):483.
WANGF P, ZHANGJ L.Review of forest carbon reserve estimation study based on carbon satellite[J].World Forestry Research,2022,35(6):30-35.
[38]
DONGL H, ZHANGL J, LIF R.Developing two additive biomass equations for three coniferous plantation species in northeast China[J].Forests,2016,7(7):136.
LIZ J, LIUY F, RANJ,et al.Optical remote sensing estimation and saturation point analysis of forest biomass in Yuanjiang Basin,Yunnan Province Based on Landsat 8 OLI[J].Journal of Southwest Forestry University:Natural Science,2023,43(1):126-136.
TIANH L, ZHUJ H, HEX,et al.Prediction biomass carbon stock of arbor forest of three provinces in northeastern China based on random forest model[J].Scientia Silvae Sinicae,2022,58(4):40-50.
FANW Y, ZHANGH Y, YUY,et al.Comparison of three models of forest biomass estimation[J].Journal of Plant Ecology,2011,35(4):402-410.
[45]
WANGG X, ZHANGM Z, GERTNERG Z,et al.Uncertainties of mapping aboveground forest carbon due to plot locations using national forest inventory plot and remotely sensed data[J].Scandinavian Journal of Forest Research,2011,26(4):360-373.
[46]
FLEMINGA L, WANGG X, MCROBERTSR E.Comparison of methods toward multi-scale forest carbon mapping and spatial uncertainty analysis:combining national forest inventory plot data and landsat TM images[J].European Journal of Forest Research,2015,134(1):125-137.