基于随机森林的绿洲典型湿地信息提取
|
摘要
以塔里木河边缘的渭-库绿洲(渭干河-库车河绿洲)为研究区,采用在特征选择和分类提取方面具有明显优势的随机森林算法,对研究区内的湿地信息进行提取。基于多时相、光谱信息丰富的Landsat8 OLI数据生成包括光谱特征、植被和水体指数、盐分指数、纹理信息在内的4种特征变量;根据以上特征设计6种不同的提取方案,对绿洲内部的干旱区湿地信息进行提取并验证不同方案的提取精度,旨在选取最佳方案提高湿地信息提取的精度。结果表明:①多种特征变量的有效组合是提高湿地信息提取精度的关键,就不同特征对湿地信息提取的贡献度而言,光谱特征>植被和水体指数>纹理特征>盐分指数;②基于随机森林算法优选的特征变量提取速度最快,效果最佳,总体精度为90.09%,Kappa系数为0.882 5。提取方法在挖掘特征变量的同时,保证了湿地信息提取的准确性,提高了运行效率。湿地提取结果对当地绿洲制定科学的水肥管理措施及进行干旱状况评估具有一定的现实意义。
In this study,the Wetland in Werigan-Kuqia River Delta Oasis( WKRDO),as a typical wetland on the banks of the Tarim River,is considered as a study area. The method of Random Forests,which has obvious advantages in feature selection and classification,is chosen to extract wetland information from the study area. First of all,four different characteristic variables,including spectral features,vegetation indices,water indices,salinity indices and texture features,are generated based on landsat 8 OLI data with rich multi-temporal and spectral information,and then six different classification schemes are constructed based on the above characteristic information. At last,the random forest classifier is used to extract the wetland information of the WKRDO,and to verify the extraction accuracy of different results.The purpose is to select the best plan to improve the effect of wetland information extraction. The results show that: ① The effective use of multiple feature variables is the key to improving the information wetland of extraction. For the contribution of different characteristics to the wetland information extraction,spectral features > vegetation indices and water indices > texture feature > salinity indices; ② the preferred features based on the random forest algorithm are significant to extraction accuracy,with the overall accuracy up to 90.09%,and the Kappa coefficient is 0.882 5. The extracted types reflect the differences in soil properties. It shows that the random forest algorithm can effectively process the feature selection. The result has important practical significance for formulating scientific water and fertilizer management measures and evaluating drought situation in the local oasis.
In this study,the Wetland in Werigan-Kuqia River Delta Oasis( WKRDO),as a typical wetland on the banks of the Tarim River,is considered as a study area. The method of Random Forests,which has obvious advantages in feature selection and classification,is chosen to extract wetland information from the study area. First of all,four different characteristic variables,including spectral features,vegetation indices,water indices,salinity indices and texture features,are generated based on landsat 8 OLI data with rich multi-temporal and spectral information,and then six different classification schemes are constructed based on the above characteristic information. At last,the random forest classifier is used to extract the wetland information of the WKRDO,and to verify the extraction accuracy of different results.The purpose is to select the best plan to improve the effect of wetland information extraction. The results show that: ① The effective use of multiple feature variables is the key to improving the information wetland of extraction. For the contribution of different characteristics to the wetland information extraction,spectral features > vegetation indices and water indices > texture feature > salinity indices; ② the preferred features based on the random forest algorithm are significant to extraction accuracy,with the overall accuracy up to 90.09%,and the Kappa coefficient is 0.882 5. The extracted types reflect the differences in soil properties. It shows that the random forest algorithm can effectively process the feature selection. The result has important practical significance for formulating scientific water and fertilizer management measures and evaluating drought situation in the local oasis.
引文
[1]陈宜瑜,吕宪国.湿地功能与湿地科学的研究方向[J].湿地科学,2003,1(1):7-11.
[2]缑倩倩,屈建军,王国华,等.中国干旱半干旱地区湿地研究进展[J].干旱区研究,2015,32(2):213-220.
[3]Finlayson C M,Davidson N C,Spiers A G,et al.Global wetland inventory:status and priorities[R].Marine and Freshwater Re-search,1999.
[4]牛振国,张海英,王显威,等.1978-2008年中国湿地类型变化[J].科学通报,2012,57(16):1 400.
[5]刘红玉,吕宪国,张世奎.三江平原流域湿地景观多样性及其50年变化研究[J].生态学报,2004,24(7):1 472-1 479.
[6]Medjani F,Aissani B,Labar S,et al.Identifying saline wetlands in an arid desert climate using Landsat remote sensing imagery.Application on Ouargla Basin,southeastern Algeria[J].Arabian Journal of Geosciences,2017,10(7):176.
[7]Halabisky M,Moskal L M,Gillespie A,et al.Reconstructing semiarid wetland surface water dynamics through spectral mixture analysis of a time series of Landsat satellite images(1984-2011)[J].Remote sensing of environment,2016,177:171-183.
[8]Khatami R,Mountrakis G,Stehman S V.A meta-analysis of remote sensing research on supervised pixel-based land-cover image classification processes:General guidelines for practitioners and future research[J].Remote Sensing of Environment,2016,177:89-100.
[9]张树文,颜凤芹,于灵雪,等.湿地遥感研究进展[J].地理科学,2013,33(11):1 406-1 412.
[10]Tian S,Zhang X,Tian J,et al.Random forest classification of wetland landcovers from multi-sensor data in the arid region of Xinjiang,China[J].Remote Sensing,2016,8(11):954.
[11]Belgiu M,Drgut L.Random forest in remote sensing:a review of applications and future directions[J].ISPRS Journal of Photogrammetry and Remote Sensing,2016,114:24-31.
[12]Dronova I.Object-based image analysis in wetland research:a review[J].Remote Sensing,2015,7(5):6 380-6 413.
[13]丁建丽,瞿娟,孙永猛,等.基于MSAVI-WI特征空间的新疆渭干河-库车河流域绿洲土壤盐渍化研究[J].地理研究,2013,32(2):223-232.
[14]刘平,关蕾,吕偲,等.中国第二次湿地资源调查的技术特点和成果应用前景[J].湿地科学,2011,9(3):284-289.
[15]王飞,丁建丽,魏阳,等.基于Landsat系列数据的盐分指数和植被指数对土壤盐度变异性的响应分析---以新疆天山南北典型绿洲为例[J].生态学报,2017,37(15):5 007-5 022.
[16]郑淑丹,郑江华,石明辉,等.基于分形和灰度共生矩阵纹理特征的种植型药用植物遥感分类[J].遥感学报,2014,18(4):868-886.
[17]Breiman L.Random forests[J].Machine learning,2001,45(1):5-32.
[18]葛翔宇,丁建丽,王敬哲,等.基于竞争适应重加权采样算法耦合机器学习的土壤含水量估算[J].光学学报,2018,38(10):1-8.
[19]Mutanga O,Adam E,Cho M A.High density biomass estimation for wetland vegetation using WorldView-2 imagery and random forest regression algorithm[J].International Journal of Applied Earth Observation and Geoinformation,2012,18:399-406.
[20]Lewis H G,Brown M.A generalized confusion matrix for assessing area estimates from remotely sensed data[J].International Journal of Remote Sensing,2001,22(16):3 223-3 235.
[21]Townsend J T.Theoretical analysis of an alphabetic confusion matrix[J].Perception&Psychophysics,1971,9(1):40-50.
[22]马晓梅,尹林克,陈理.塔里木河干流胡杨和柽柳根际土壤微生物及其垂直分布[J].干旱区研究,2008,25(2):183-189.
[2]缑倩倩,屈建军,王国华,等.中国干旱半干旱地区湿地研究进展[J].干旱区研究,2015,32(2):213-220.
[3]Finlayson C M,Davidson N C,Spiers A G,et al.Global wetland inventory:status and priorities[R].Marine and Freshwater Re-search,1999.
[4]牛振国,张海英,王显威,等.1978-2008年中国湿地类型变化[J].科学通报,2012,57(16):1 400.
[5]刘红玉,吕宪国,张世奎.三江平原流域湿地景观多样性及其50年变化研究[J].生态学报,2004,24(7):1 472-1 479.
[6]Medjani F,Aissani B,Labar S,et al.Identifying saline wetlands in an arid desert climate using Landsat remote sensing imagery.Application on Ouargla Basin,southeastern Algeria[J].Arabian Journal of Geosciences,2017,10(7):176.
[7]Halabisky M,Moskal L M,Gillespie A,et al.Reconstructing semiarid wetland surface water dynamics through spectral mixture analysis of a time series of Landsat satellite images(1984-2011)[J].Remote sensing of environment,2016,177:171-183.
[8]Khatami R,Mountrakis G,Stehman S V.A meta-analysis of remote sensing research on supervised pixel-based land-cover image classification processes:General guidelines for practitioners and future research[J].Remote Sensing of Environment,2016,177:89-100.
[9]张树文,颜凤芹,于灵雪,等.湿地遥感研究进展[J].地理科学,2013,33(11):1 406-1 412.
[10]Tian S,Zhang X,Tian J,et al.Random forest classification of wetland landcovers from multi-sensor data in the arid region of Xinjiang,China[J].Remote Sensing,2016,8(11):954.
[11]Belgiu M,Drgut L.Random forest in remote sensing:a review of applications and future directions[J].ISPRS Journal of Photogrammetry and Remote Sensing,2016,114:24-31.
[12]Dronova I.Object-based image analysis in wetland research:a review[J].Remote Sensing,2015,7(5):6 380-6 413.
[13]丁建丽,瞿娟,孙永猛,等.基于MSAVI-WI特征空间的新疆渭干河-库车河流域绿洲土壤盐渍化研究[J].地理研究,2013,32(2):223-232.
[14]刘平,关蕾,吕偲,等.中国第二次湿地资源调查的技术特点和成果应用前景[J].湿地科学,2011,9(3):284-289.
[15]王飞,丁建丽,魏阳,等.基于Landsat系列数据的盐分指数和植被指数对土壤盐度变异性的响应分析---以新疆天山南北典型绿洲为例[J].生态学报,2017,37(15):5 007-5 022.
[16]郑淑丹,郑江华,石明辉,等.基于分形和灰度共生矩阵纹理特征的种植型药用植物遥感分类[J].遥感学报,2014,18(4):868-886.
[17]Breiman L.Random forests[J].Machine learning,2001,45(1):5-32.
[18]葛翔宇,丁建丽,王敬哲,等.基于竞争适应重加权采样算法耦合机器学习的土壤含水量估算[J].光学学报,2018,38(10):1-8.
[19]Mutanga O,Adam E,Cho M A.High density biomass estimation for wetland vegetation using WorldView-2 imagery and random forest regression algorithm[J].International Journal of Applied Earth Observation and Geoinformation,2012,18:399-406.
[20]Lewis H G,Brown M.A generalized confusion matrix for assessing area estimates from remotely sensed data[J].International Journal of Remote Sensing,2001,22(16):3 223-3 235.
[21]Townsend J T.Theoretical analysis of an alphabetic confusion matrix[J].Perception&Psychophysics,1971,9(1):40-50.
[22]马晓梅,尹林克,陈理.塔里木河干流胡杨和柽柳根际土壤微生物及其垂直分布[J].干旱区研究,2008,25(2):183-189.