基于高分遥感数据的森林郁闭度估测方法研究
|
摘要
森林郁闭度是研究森林生态系统和了解森林资源状况的重要参数,而传统的实地测量方法效率较低下,且仅能获取小范围的一些具有代表性的数据,不利于研究大范围或区域内郁闭度的空间分布及变化。为了估算森林郁闭度,分析其与遥感影像因子之间的相关性,本文以河北省滦平县巴克什营镇和长山峪镇为研究区域,采用分辨率较高的高分一号(GF-1)数据,结合SRTM DEM数据的地形因子,对该地区的森林郁闭度进行反演。本文在系统整理分析和评价国内外森林郁闭度相关研究文献的基础上,选择了红波段、近红外波段、亮度、绿度、黄度等14个因子作为自变量参与构建多元逐步回归(Multivariable Stepw ise Regression,MSR)、随机森林(Random Forest,RF)和Cubist三种模型,对该地郁闭度进行估测。实验结果表明,基于机器学习的随机森林和Cubist算法结果要优于传统的多元逐步回归算法,各项评价指标显示其中Cubist回归算法在该研究区的拟合效果最好。多元逐步回归(MSR)算法成熟简单,应用广泛,但模型不稳定,反演精度不高,不适用于大区域的郁闭度估算;随机森林(RF)处理大数据速度快,但高值低估和低值高估的情况比较严重,增大了郁闭度估测误差; Cubist在预测连续值方面很成功,使用最近邻样本来调整规则预测结果,模型较稳定,能够得到较为准确的预测数值,不过需要花费很长时间进行计算。
Forest canopy density is an important parameter for studying forest ecosystem and understanding forest resources.Traditional field measurement methods are inefficient and can only obtain some representative data in a small range,which is not conducive to the study of spatial distribution and change of forest canopy density in a large area or region. In order to estimate forest canopy density and analyze its correlation with remote sensing image factors,this paper takes Bakeshiying Town and Changshanyu Town in Luanping County of Hebei Province as the research area,and uses high resolution GF-1 data,combined with SRTM DEM data topographic factors,to invert forest canopy density in this area. On the basis of systematically analyzing and evaluating the related literatures of forest canopy density at home and abroad,14 factors,such as red band,near infrared band,brightness,greenness and yellowness,are selected as independent variables to participate in the construction of three models: multi-variable Stepwise Regression( MSR),Random Forest( RF) and Cubist,and the canopy density is improved. The experimental results show that the results of random forest and Cubist algorithm based on machine learning are better than those of traditional multiple stepwise regression algorithm. The evaluation indexes show that Cubist regression algorithm has the best fitting effect in this research area.Multivariate stepwise regression( MSR) algorithm is mature and simple,and widely used,but the model is unstable,the inversion accuracy is not high,and it is not suitable for estimating canopy density in large areas; Random Forest( RF) can process large data quickly,but the situation of overestimation and underestimation is serious,which increases the estimation error of canopy density;Cubist is very successful in predicting continuous values,and uses the nearest neighbor sample to adjust the rules. The prediction results show that the model is stable and can get more accurate prediction values,but it takes a long time to calculate.
Forest canopy density is an important parameter for studying forest ecosystem and understanding forest resources.Traditional field measurement methods are inefficient and can only obtain some representative data in a small range,which is not conducive to the study of spatial distribution and change of forest canopy density in a large area or region. In order to estimate forest canopy density and analyze its correlation with remote sensing image factors,this paper takes Bakeshiying Town and Changshanyu Town in Luanping County of Hebei Province as the research area,and uses high resolution GF-1 data,combined with SRTM DEM data topographic factors,to invert forest canopy density in this area. On the basis of systematically analyzing and evaluating the related literatures of forest canopy density at home and abroad,14 factors,such as red band,near infrared band,brightness,greenness and yellowness,are selected as independent variables to participate in the construction of three models: multi-variable Stepwise Regression( MSR),Random Forest( RF) and Cubist,and the canopy density is improved. The experimental results show that the results of random forest and Cubist algorithm based on machine learning are better than those of traditional multiple stepwise regression algorithm. The evaluation indexes show that Cubist regression algorithm has the best fitting effect in this research area.Multivariate stepwise regression( MSR) algorithm is mature and simple,and widely used,but the model is unstable,the inversion accuracy is not high,and it is not suitable for estimating canopy density in large areas; Random Forest( RF) can process large data quickly,but the situation of overestimation and underestimation is serious,which increases the estimation error of canopy density;Cubist is very successful in predicting continuous values,and uses the nearest neighbor sample to adjust the rules. The prediction results show that the model is stable and can get more accurate prediction values,but it takes a long time to calculate.
引文
[1]龚韡,吴聪连,陈奕浪.信息技术在林业管理中的应用[J].南方农业,2017,11(7):35-36.
[2]孙华.SPOT5在森林资源调查中的应用研究[D].长沙:中南林业科技大学,2006.
[3]赵国帅.无人机遥感在林业中的应用与需求分析[J].福建林业科技,2017,44(1):136-140.
[4]车腾腾,冯益明,吴春争.“3S”技术在精准林业中的应用[J].绿色科技,2010(10):158-162.
[5]柏艺.基于3种不同数据源的川西南常绿阔叶林郁闭度估算模型研究[D].雅安:四川农业大学,2016.
[6]邹杰,诸葛祥东.森林郁闭度及其测定方法[J].黑龙江科技信息,2011(35):290.
[7]李永宁,张宾兰,秦淑英,等.郁闭度及其测定方法研究与应用[J].世界林业研究,2008,21(1):40-46.
[8]杜文峰,王凤臻,李庆.提高郁闭度调查精度的几点建议[J].林业资源管理,1999(3):62-64.
[9]王强,斯琴毕力格,肖海涛,等.森林郁闭度测定方法探讨与展望[J].内蒙古林业调查设计,2014,37(1):82-83,108.
[10]覃先林,陈尔学,李增元,等.基于MODIS数据的森林覆盖变化监测方法研究[J].遥感技术与应用,2006,21(3):178-183.
[11]吴胜义,赵强国,石昊楠.SPOT5遥感影像在林业二类调查中的应用[J].福建林业科技,2011,38(3):25-31.
[12]陈尔学.合成孔径雷达森林生物量估测研究进展[J].世界林业研究,1999,9(6):18-23.
[13]NELSON R,KRABILL W,MACLEAN G.Determining forest canopy characteristics using airborne laser data[J].Remote Sensing of Environment,1984,15(3):201-212.
[14]樊晶,杨燕琼.基于遥感的森林健康度分析—以东莞桉树林为例[J].林业与环境科学,2017,33(1):40-45.
[15]庞勇,李增元.基于机载激光雷达的小兴安岭温带森林组分生物量反演[J].植物生态学报,2012,36(10):1095-1105.
[16]黄燕平,陈劲松.基于SAR数据的森林生物量估测研究进展[J].国土资源遥感,2013,25(3):7-13.
[17]NELSON R,KRABILL W,TONELLI J.Estimating forest biomass and volume using airborne laser data[J].Remote Sensing of Environment,1988,24(2):247-267.
[18]WULDER MA,NORMAN P,WITTE C.Special issue:Li DAR remote sensing of forest structure and terrain[J].Can.J.For.Res,2003,29(15):527-678.
[19]SUN G Q,RANSON K J,KIMES D S,et al.Forest vertical structure from glas:An evaluation using LVIS and SRTMdata[J].Remote Sensing of Environment,2008,112(1):107-117.
[20]HUDAK A T,EVANS J S,SMITH A MS.Li DAR utility for natural resource managers[J].Remote Sensing,2009,1(4):934-951.
[2]孙华.SPOT5在森林资源调查中的应用研究[D].长沙:中南林业科技大学,2006.
[3]赵国帅.无人机遥感在林业中的应用与需求分析[J].福建林业科技,2017,44(1):136-140.
[4]车腾腾,冯益明,吴春争.“3S”技术在精准林业中的应用[J].绿色科技,2010(10):158-162.
[5]柏艺.基于3种不同数据源的川西南常绿阔叶林郁闭度估算模型研究[D].雅安:四川农业大学,2016.
[6]邹杰,诸葛祥东.森林郁闭度及其测定方法[J].黑龙江科技信息,2011(35):290.
[7]李永宁,张宾兰,秦淑英,等.郁闭度及其测定方法研究与应用[J].世界林业研究,2008,21(1):40-46.
[8]杜文峰,王凤臻,李庆.提高郁闭度调查精度的几点建议[J].林业资源管理,1999(3):62-64.
[9]王强,斯琴毕力格,肖海涛,等.森林郁闭度测定方法探讨与展望[J].内蒙古林业调查设计,2014,37(1):82-83,108.
[10]覃先林,陈尔学,李增元,等.基于MODIS数据的森林覆盖变化监测方法研究[J].遥感技术与应用,2006,21(3):178-183.
[11]吴胜义,赵强国,石昊楠.SPOT5遥感影像在林业二类调查中的应用[J].福建林业科技,2011,38(3):25-31.
[12]陈尔学.合成孔径雷达森林生物量估测研究进展[J].世界林业研究,1999,9(6):18-23.
[13]NELSON R,KRABILL W,MACLEAN G.Determining forest canopy characteristics using airborne laser data[J].Remote Sensing of Environment,1984,15(3):201-212.
[14]樊晶,杨燕琼.基于遥感的森林健康度分析—以东莞桉树林为例[J].林业与环境科学,2017,33(1):40-45.
[15]庞勇,李增元.基于机载激光雷达的小兴安岭温带森林组分生物量反演[J].植物生态学报,2012,36(10):1095-1105.
[16]黄燕平,陈劲松.基于SAR数据的森林生物量估测研究进展[J].国土资源遥感,2013,25(3):7-13.
[17]NELSON R,KRABILL W,TONELLI J.Estimating forest biomass and volume using airborne laser data[J].Remote Sensing of Environment,1988,24(2):247-267.
[18]WULDER MA,NORMAN P,WITTE C.Special issue:Li DAR remote sensing of forest structure and terrain[J].Can.J.For.Res,2003,29(15):527-678.
[19]SUN G Q,RANSON K J,KIMES D S,et al.Forest vertical structure from glas:An evaluation using LVIS and SRTMdata[J].Remote Sensing of Environment,2008,112(1):107-117.
[20]HUDAK A T,EVANS J S,SMITH A MS.Li DAR utility for natural resource managers[J].Remote Sensing,2009,1(4):934-951.