基于神经网络和不同立地质量的森林蓄积量遥感估测
|
摘要
运用BP神经网络技术建立区分立地质量等级森林蓄积量遥感估测模型并探讨其适用性。以2009年黑龙江省伊春市凉水自然保护区森林资源二类调查数据为基础数据划分森林的立地质量等级,以森林蓄积量为研究对象,基于该地区LANDSAT-TM影像以及DEM数据提取遥感因子,采用BP神经网络方法构建区分立地质量的森林蓄积量遥感估测模型,并引入回归分析方法和不区分立地质量的模型予以比较。结果表明,基于不同立地质量等级的模型明显好于不区分立地质量等级的估测模型,且BP神经网络模型较回归分析模型可以更好地预测森林的蓄积量。经过对比检验,基于不同立地质量等级的BP神经网络模型性能优异,验证总体预测精度高达97%以上、实测值与预测值的R~2在0.94左右;不区分立地质量等级的BP神经网络模型的R~2为0.89,预测精度95%左右。同等条件下,BP神经网络模型较多元线性回归模型的预测精度约提高了3%～5%,R~2值提高了0.1左右。
This paper discussed a remote sensing estimation model for the forest stock volume with different site qualities by using BP neural network technology, and its applicability. Based on the data of the second type of forest resources in the Liangshui Nature Reserve of Yichun City, Heilongjiang Province in 2009, the site qualities of the forest were divided. With the forest stock volume as research object, the remote sensing factor was extracted based on the LANDSAT-TM image and DEM data in the region. The BP neural network method was used to construct a remote sensing estimation model for forest stocks with different site qualities. A regression analysis method and a non-separate quality model were introduced for comparison. The results show that the models based on different site qualities are better than the estimation models without the distinction of site qualities and that the BP neural network model can better predict the forest accumulation volume than the regression analysis model. The comparison indicates that the BP neural network model based on different site qualities has excellent performance, the overall prediction accuracy being over 97% and the measured value and predicted R~2 being around 0.94, while the R~2 of the BP neural network model without distinguishing site qualities is 0.89 and the prediction accuracy is about 95%. Under the same conditions, the prediction accuracy of the BP neural network model is about 3%-5% higher than that of the multiple linear regression model, and the R~2 value is enhanced by about 0.1.
This paper discussed a remote sensing estimation model for the forest stock volume with different site qualities by using BP neural network technology, and its applicability. Based on the data of the second type of forest resources in the Liangshui Nature Reserve of Yichun City, Heilongjiang Province in 2009, the site qualities of the forest were divided. With the forest stock volume as research object, the remote sensing factor was extracted based on the LANDSAT-TM image and DEM data in the region. The BP neural network method was used to construct a remote sensing estimation model for forest stocks with different site qualities. A regression analysis method and a non-separate quality model were introduced for comparison. The results show that the models based on different site qualities are better than the estimation models without the distinction of site qualities and that the BP neural network model can better predict the forest accumulation volume than the regression analysis model. The comparison indicates that the BP neural network model based on different site qualities has excellent performance, the overall prediction accuracy being over 97% and the measured value and predicted R~2 being around 0.94, while the R~2 of the BP neural network model without distinguishing site qualities is 0.89 and the prediction accuracy is about 95%. Under the same conditions, the prediction accuracy of the BP neural network model is about 3%-5% higher than that of the multiple linear regression model, and the R~2 value is enhanced by about 0.1.
引文
[1]刘琼阁,彭道黎,涂云燕.基于偏最小二乘回归的森林蓄积量遥感估测[J].中南林业科技大学学报,2014,34(2):81-84.LIU Qiongge,PENG Daoli,TU Yunyan.Estimation of forest stock volume based on partial least squares regression[J].Journal of Central South University of Forestry & Technology,2014,34(2):81-84.
[2]DIAS A T C D,MATTOS E A M,VIEIRA S A,et al.Aboveground biomass stock of native woodland on a Brazilian sandy coastal plain:Estimates based on the dominant tree species[J].Forest Ecology & Management,2006,226(1):364-367.
[3]TROTTER C M,DYMOND J R,GOULDING C J.Estimation of timber volume in a coniferous plantation forest using Landsat TM[J].International Journal of Remote Sensing,1997,18(10):2209-2223.
[4]WIJAYA A,KUSNADI S,GLOAGUEN R,et al.Improved strategy for estimating stem volume and forest biomass using moderate resolution remote sensing data and GIS[J].Journal of Forestry Research,2010,21(1):1-12.
[5]刘俊,孟雪,温小荣,等.基于不同立地质量的森林蓄积量遥感估测[J].西北林学院学报,2016,31(1):186-191.LIU Jun,MENG Xue,WEN Xiaorong,et al.Remote sensing estimation of forest volume based on different site qualities[J].Journal of Northwest Forestry University,2016,31(1):186-191.
[6]NELSON R,KRABILLW,MACLEAN G.Determining forest canopy characteristics using airborne laser data[ J] .Remote Sensing of Environment,1984,15(3):201-212.
[7]GEMMELL F M.Effects of forest cover,terrain,and scale on timmiber volume estimation with Thematic Mapper data in a rocky mountain site[ J].Remote Sensing of Environment,1995,51(2):291-305.
[8]FAZAKAS Z,NILSSON M,OLSSON H.Regional forest biomass and wood volume estimation using satellite data and ancillary data[J].Agricultural and Forest Meteorology,1999,99:417-425.
[9]程武学,倪勇,杨存建,等.基于小班对象的森林蓄积量遥感估测模型合理性诊断[J].湖北农业科学,2011,50(23):4945-4948.CHENG Wuxue,NI Yong,YANG Cunjian,et al.Rationality diagnosis of remote estimation model of forest volume based on small class objects[J].Hubei Agricultural Sciences,2011,50(23):4945-4948.
[10]王臣立,牛铮,郭治兴,等.基于植被指数和神经网络的热带人工林地上蓄积量遥感估测[J].生态环境学报,2009,18(5):1830-1834.WANG Chenli,NIU Zheng,GUO Zhixing,et al.Estimating tropical forest stock volume based on vegetation indices and neural network model[J].Ecology and Environmental Sciences,2009,18(5):1830-1834.
[11]许炜敏,陈友飞,陈明华,等.基于BP神经网络的杉木林蓄积量估测研究[J].森林与环境学报,2012,32(4):310-315.XU Weimin,CHEN Youfei,CHEN Minghua,et al.Research on the estimation of Chinese fir volume based on BP neural networks[J].Journal of Fujian College of Forestry,2012,32(4):310-315.
[12]吴达胜.基于多源数据和神经网络模型的森林资源蓄积量动态监测[D].杭州:浙江大学,2014.
[13]林昌庚,周春国,林俊钦,等.关于地位级表[J].林业资源管理,1997(5):31-34.LIN Changgeng,ZHOU Chunguo,LIN Junqin,et.al.About the status level table[J].Forestry Resource Management,1997(5):31-34.
[14]周春国,徐雁南.地位级表编制的新方法[J].华东森林经理,1997(2):11-13.ZHOU Chunguo,XU Yannan.A new method for the establishment of status table[J].East China Forest Management,1997(2):11-13.
[15]黄杏模.谈最佳立地指数导向曲线的评定[J].中南林业调查规划,1990(3):45-47.HUANG Xingmo.Evaluation of guidance curve for best site index[J].Central South Forest Inventory and Planning,1990(3):45-47.
[16]王艳婷,李崇贵,郝利军.用岭估计估测以分类为前提的森林蓄积量[J].东北林业大学学报,2014,42(9):39-42.WANG Yanting,LI Chonggui,HAO Lijun.Forest volume estimation on the premise of classification by ridge estimate[J].Journal of Northeast Forestry University,2014,42(9):39-42.
[17]刘俊,孟雪,温小荣,等.基于不同立地质量的松树林蓄积量遥感估测[J].林业资源管理,2015(5):55-60.LIU Jun,MENG Xue,WEN Xiaorong,et al.Remote sensing estimation of pine volume based on different site quality[J].Forestry Resource Management,2015(5):55-60.
[18]GAO J.Quantification of grassland properties:How it can benefit from geoinformatic technologies[J].International Journal of Remote Sensing,2006,27(7):1351-1365.
[19]柯宗建,张培群,董文杰,等.最优子集回归方法在季节气候预测中的应用[J].大气科学,2009,33(5):994-1002.KE Zongjian,ZHANG Peiqun,DONG Wenjie,etal.Application of optimal subset regression method in seasonal climate prediction[J].Chinese Journal of Atmospheric Sciences,2009,33(5):994-1002.
[20]MALLOWS C L.Some comments on CP[J].Technometrics,1973,15(4):661-675.
[21]何芳,王小川,肖森予,等.基于MIV-BP型网络实验的房地产项目风险识别研究[J].运筹与管理,2013,22(2):229-234.HE Fang,WANG Xiaochuan,XIAO Senyu,et al.Research on risk recognition of real estate projects based on MIV-BP Neural network test[J].Operations Research and Management Science,2013,22(2):229-234.
[2]DIAS A T C D,MATTOS E A M,VIEIRA S A,et al.Aboveground biomass stock of native woodland on a Brazilian sandy coastal plain:Estimates based on the dominant tree species[J].Forest Ecology & Management,2006,226(1):364-367.
[3]TROTTER C M,DYMOND J R,GOULDING C J.Estimation of timber volume in a coniferous plantation forest using Landsat TM[J].International Journal of Remote Sensing,1997,18(10):2209-2223.
[4]WIJAYA A,KUSNADI S,GLOAGUEN R,et al.Improved strategy for estimating stem volume and forest biomass using moderate resolution remote sensing data and GIS[J].Journal of Forestry Research,2010,21(1):1-12.
[5]刘俊,孟雪,温小荣,等.基于不同立地质量的森林蓄积量遥感估测[J].西北林学院学报,2016,31(1):186-191.LIU Jun,MENG Xue,WEN Xiaorong,et al.Remote sensing estimation of forest volume based on different site qualities[J].Journal of Northwest Forestry University,2016,31(1):186-191.
[6]NELSON R,KRABILLW,MACLEAN G.Determining forest canopy characteristics using airborne laser data[ J] .Remote Sensing of Environment,1984,15(3):201-212.
[7]GEMMELL F M.Effects of forest cover,terrain,and scale on timmiber volume estimation with Thematic Mapper data in a rocky mountain site[ J].Remote Sensing of Environment,1995,51(2):291-305.
[8]FAZAKAS Z,NILSSON M,OLSSON H.Regional forest biomass and wood volume estimation using satellite data and ancillary data[J].Agricultural and Forest Meteorology,1999,99:417-425.
[9]程武学,倪勇,杨存建,等.基于小班对象的森林蓄积量遥感估测模型合理性诊断[J].湖北农业科学,2011,50(23):4945-4948.CHENG Wuxue,NI Yong,YANG Cunjian,et al.Rationality diagnosis of remote estimation model of forest volume based on small class objects[J].Hubei Agricultural Sciences,2011,50(23):4945-4948.
[10]王臣立,牛铮,郭治兴,等.基于植被指数和神经网络的热带人工林地上蓄积量遥感估测[J].生态环境学报,2009,18(5):1830-1834.WANG Chenli,NIU Zheng,GUO Zhixing,et al.Estimating tropical forest stock volume based on vegetation indices and neural network model[J].Ecology and Environmental Sciences,2009,18(5):1830-1834.
[11]许炜敏,陈友飞,陈明华,等.基于BP神经网络的杉木林蓄积量估测研究[J].森林与环境学报,2012,32(4):310-315.XU Weimin,CHEN Youfei,CHEN Minghua,et al.Research on the estimation of Chinese fir volume based on BP neural networks[J].Journal of Fujian College of Forestry,2012,32(4):310-315.
[12]吴达胜.基于多源数据和神经网络模型的森林资源蓄积量动态监测[D].杭州:浙江大学,2014.
[13]林昌庚,周春国,林俊钦,等.关于地位级表[J].林业资源管理,1997(5):31-34.LIN Changgeng,ZHOU Chunguo,LIN Junqin,et.al.About the status level table[J].Forestry Resource Management,1997(5):31-34.
[14]周春国,徐雁南.地位级表编制的新方法[J].华东森林经理,1997(2):11-13.ZHOU Chunguo,XU Yannan.A new method for the establishment of status table[J].East China Forest Management,1997(2):11-13.
[15]黄杏模.谈最佳立地指数导向曲线的评定[J].中南林业调查规划,1990(3):45-47.HUANG Xingmo.Evaluation of guidance curve for best site index[J].Central South Forest Inventory and Planning,1990(3):45-47.
[16]王艳婷,李崇贵,郝利军.用岭估计估测以分类为前提的森林蓄积量[J].东北林业大学学报,2014,42(9):39-42.WANG Yanting,LI Chonggui,HAO Lijun.Forest volume estimation on the premise of classification by ridge estimate[J].Journal of Northeast Forestry University,2014,42(9):39-42.
[17]刘俊,孟雪,温小荣,等.基于不同立地质量的松树林蓄积量遥感估测[J].林业资源管理,2015(5):55-60.LIU Jun,MENG Xue,WEN Xiaorong,et al.Remote sensing estimation of pine volume based on different site quality[J].Forestry Resource Management,2015(5):55-60.
[18]GAO J.Quantification of grassland properties:How it can benefit from geoinformatic technologies[J].International Journal of Remote Sensing,2006,27(7):1351-1365.
[19]柯宗建,张培群,董文杰,等.最优子集回归方法在季节气候预测中的应用[J].大气科学,2009,33(5):994-1002.KE Zongjian,ZHANG Peiqun,DONG Wenjie,etal.Application of optimal subset regression method in seasonal climate prediction[J].Chinese Journal of Atmospheric Sciences,2009,33(5):994-1002.
[20]MALLOWS C L.Some comments on CP[J].Technometrics,1973,15(4):661-675.
[21]何芳,王小川,肖森予,等.基于MIV-BP型网络实验的房地产项目风险识别研究[J].运筹与管理,2013,22(2):229-234.HE Fang,WANG Xiaochuan,XIAO Senyu,et al.Research on risk recognition of real estate projects based on MIV-BP Neural network test[J].Operations Research and Management Science,2013,22(2):229-234.