温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,汇文网负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。
网站客服:3074922707
结合
遥感
林龄
因子
亚热带
森林
蓄积
估算
方法
周小成
doi:10.11707/j.1001-7488.LYKX20210712结合遥感林龄因子的亚热带森林蓄积量估算方法周小成1黄婷婷1李媛1肖祥希2朱洪如3陈芸芝1冯芝淸4(1.福州大学空间数据挖掘与信息共享教育部重点实验室卫星空间信息技术综合应用国家地方联合工程研究中心福州 350108;2.福建省林业科学研究院福州 350012;3.福建省林业调查规划院福州 350003;4.福建金森林业股份有限公司将乐 353300)摘要:【目的】应用 XGBoost 算法建立包含林龄的遥感因子-蓄积量模型,评价遥感估算的林龄因子与遥感因子相结合提高森林蓄积量估算精度的有效性,为实现高效、快速、精准的大范围森林蓄积量估算提供一种新的思路和方法。【方法】以福建省将乐县为研究示范区,首先,基于 19872016 年时序 Landsat 影像,采用 LandTrendr 森林干扰与恢复监测算法监测年度林分更替干扰并估算干扰区林龄;然后,基于 GF-1 号影像光谱、纹理、地形等特征,采用递归特征消除的随机森林算法(RFE-RF)估算非干扰区林龄;在此基础上,结合 GF-1 影像光谱、纹理因子和森林资源二类调查小班实测蓄积量数据,采用极端梯度提升算法估算研究区森林蓄积量。对比有无林龄因子的森林蓄积量估算精度,进一步验证遥感林龄因子对提高森林蓄积量估算精度的重要性。【结果】采用 LandTrendr 森林干扰与恢复监测算法获得的干扰区林分林龄误差仅 12 年,林龄估算精度明显优于传统利用遥感因子估算的林龄精度(误差 412 年)。仅采用常规遥感因子估算森林蓄积量时,XGBoost 模型决定系数(R2)为 0.59,平均均方根误差(RMSE)为 30.72 m3hm2,相对均方根误差(rRMSE)为 16.46%;加入林龄因子后,模型 R2提高至 0.73,平均 RMSE 减少至 23.73 m3hm2,rRMSE 为 13.26%,森林蓄积量估算平均总体精度约提高 10.4%,达 84.4%。【结论】相比仅采用常规遥感因子估算森林蓄积量,应用XGBoost 算法建立包含林龄的遥感因子-蓄积量模型,其估算精度接近森林资源调查相关规定要求,可为大范围亚热带森林资源快速调查评估提供重要技术支持。关键词:森林蓄积量;林龄;时序遥感;递归特征消除的随机森林;极端梯度提升算法中图分类号:Q958.15;S154.5文献标识码:A文章编号:10017488(2023)04008812A Method for Estimating Subtropical Forest Stock by Combining RemotelySensed Forest Age FactorsZhou Xiaocheng1Huang Tingting1Li Yuan1Xiao Xiangxi2Zhu Hongru3Chen Yunzhi1Feng Zhiqing4(1.Local Joint Engineering Research Center of Satellite Geospatial Information TechnologyKey Laboratory of Spatial Data Mining and Information Sharing ofMinistry of Education,Fuzhou UniversityFuzhou 350108;2.Fujian Academy of ForestryFuzhou 350012;3.Fujian Forest Inventory and Planning InstituteFuzhou 350003;4.Fujian Jinsen Forestry Co.Ltd.Jiangle 353300)Abstract:【Objective】The XGBoost algorithm was applied to establish a remote sensing factor-stock volume modelcontaining forest age,to evaluate the effectiveness of combining the remote sensing estimated forest age factor with the remotesensing factor to improve the accuracy of forest volume estimation,and to provide a new idea and method to achieve efficient,fastand accurate forest volume estimation on a large scale.【Method】Taking Jiangle County,Fujian Province as a case,firstly,basedon the time-series Landsat images from 19872016,combined with the measured stock volume data of subcompartment of forestresource inventory and planning,the LandTrendr forest disturbance and restoration monitoring algorithm was used to monitor theannual stand turnover disturbance and estimate the forest age in the disturbance area;Second,based on the GF-1 image spectral,texture,and topography features,the recursive feature elimination random forest algorithm(RFE-RF)to estimate the forest age inthe non-disturbed area;Finally,the GF-1 image spectral and texture factors were combined with the forest age factor by theextreme gradient boosting algorithm (XGBoost)to estimate the forest stock of the study area.The accuracy of forest stockestimation with and without the forest age factor was compared to further verify the importance of remote sensing forest age factorto improve the accuracy of forest stock estimation.【Result】The error of forest age obtained by using LandTrendr algorithm in 收稿日期:20210917;修回日期:20220310。基金项目:福 建 省 科 技 厅 对 外 合 作 项 目(2022I0007);福 建 省 科 技 厅 高 校 产 学 合 作 项 目(2022N5008);福 建 省 林 业 科 技 攻 关 项 目(2021FKJ01);“十二五”科技支撑计划项目专题“华北土石山区森林可持续经营技术研究与示范”(2012BAD22B0304);国家林业局林业公益性行业科研专项(20100400205)。第 59 卷 第 4 期林业科学 Vol.59,No.42 0 2 3 年 4 月SCIENTIA SILVAE SINICAEApr.,2 0 2 3the forest disturbance area was only 1-2 years,and the accuracy of forest age estimation was significantly better than that of thetraditional estimation of forest age using remote sensing factors(error of 4-12 years).When only conventional remote sensingfactors were used to estimate the volume,the model R2 of XGBoost was 0.59 and the average RMSE was 30.72 m3hm2,therRMSE was 16.46%;after adding the forest age factor,the model R2 increased to 0.73,the average RMSE decreased to23.73 m3hm2,the rRMSE was 13.26%,and the average overall accuracy of the stock volume estimation improved by about 10.4%to 84.4%.【Conclusion】The accuracy of the XGBoost algorithm combined with the forest age parameter for estimating the stockvolume is close to the requirements of the relevant regulations of forest resources survey,which can provide important technicalsupport for the rapid survey and assessment of forest resources on a large scale.Key words:forest stock volume;forest age;time series remote sensing;RFE-RF;XGBoost 林龄和蓄积量是森林资源调查中重要的林分调查因子,林龄一般指林木自种子萌发后生长的年数(孟宪宇,2006),蓄积量通常指森林中所有林木材积的总和(Nilsson,1996)。森林蓄积量是反映一个国家或地区森林资源总规模和水平的基本指标之一,也是反映森林资源丰富程度、衡量森林生态环境优劣的重要依据,大范围准确估算森林蓄积量对于表征林业经济和评判森林质量至关重要。传统的森林蓄积量调查以森林资源一、二类调查为主,不仅调查周期长,耗费大量人力、物力和财力等(罗凯健等,2021),而且无法获得特殊地形地势的森林蓄积量。卫星遥感是一种动态性较强、视野较为广阔、信息获取便捷的新兴技术,其可克服人工调查的局限性,为大范围森林蓄积量估算提供了可能,对森林资源开发和有效利用具有极其重要的意义和作用(杨军,2020)。研究发现,林龄与卫星影像的光谱反射率和纹理等信息有关,通过实测林龄与影像光谱值、纹理等特征及其衍生植被指数之间的回归关系(Li et al.,2014),可推算林龄的空间分布。所用卫星影像包括 MODIS影像(Loboda et al.,2017)、QuickBird 影像(Dye et al.,2012)、Landsat 影 像(Kou et al.,2015)、SPOT 影 像(Wunderle et al.,2007)、Sentinel-2 影 像(唐 少 飞 等,2020)以及高分辨率影像(Cong et al.,2018;