基于Sentinel-2时间序列数据及物候特征的棉花种植区提取

doi:10.12118/j.issn.1000-6060.2022.119

干旱区地理 ›› 2022, Vol. 45 ›› Issue (6): 1847-1859.doi: 10.12118/j.issn.1000-6060.2022.119

基于Sentinel-2时间序列数据及物候特征的棉花种植区提取

美合日阿依·莫一丁¹(),买买提·沙吾提^1,^2,³(),李金朝¹

1.新疆大学地理与遥感科学学院，新疆乌鲁木齐 830046
2.新疆绿洲生态重点实验室，新疆乌鲁木齐 840046
3.智慧城市与环境建模自治区普通高校重点实验室，新疆乌鲁木齐 840046

收稿日期:2022-03-23 修回日期:2022-05-16 出版日期:2022-11-25 发布日期:2023-02-01
通讯作者: 买买提·沙吾提（1976-），男，博士，副教授，主要从事多源遥感信息提取及农业遥感等方面的研究. E-mail: korxat@xju.edu.cn
作者简介:美合日阿依·莫一丁（1997-），女，硕士研究生，主要从事卫星遥感和植被监测等方面的研究. E-mail: mihray_m@163.com
基金资助:
新疆自然科学计划（自然科学基金）联合基金项目(2021D01C055)

Extraction of cotton planting area based on Sentinel-2 time series data and phenological characteristics

Mihray MOYIDIN¹(),Mamat SAWUT^1,^2,³(),LI Jinzhao¹

1. College of Geography and Remote Sensing Science, Xinjiang University, Urumqi 830046, Xinjiang, China
2. Xinjiang Key Laboratory of Oasis Ecology, Xinjiang University, Urumqi 830046, Xinjiang, China
3. Key Laboratory of Smart City and Environment Modelling of Higher Education Institute, Xinjiang University, Urumqi 830046, Xinjiang, China

Received:2022-03-23 Revised:2022-05-16 Online:2022-11-25 Published:2023-02-01
Contact: SAWUT Mamat

摘要/Abstract

摘要：

棉花是中国重要的经济作物，在新疆大面积种植。及时、准确获取棉花种植面积，对农业政策制定与农业经济发展有重要意义。以渭干河—库车河三角洲绿洲棉花为主要研究对象，利用2018—2020年（1景/1月）36景哨兵2号（Sentinel-2）数据，构建归一化植被指数（Normalize difference vegetation index，NDVI）和红边归一化植被指数（Red edge normalize difference vegetation index，RENDVI₇₈₃）时序数据；采用Savitzky-Golay（S-G）滤波法对时序数据进行平滑、重构并提取11个物候特征；利用袋外误差法对11个物候特征进行特征优选；在此基础上利用重构后的时序数据（NDVI Fit）、（RENDVI₇₈₃ Fit）、物候特征（RENDVI₇₈₃ Ph）、物候特征优选组合构建6种不同的特征数据集，利用随机森林分类（RFC）方法分别进行分类和提取，并采用最大似然分类方法和支持向量机分类方法对分类效果进行验证。结果表明：（1） NDVI和RENDVI₇₈₃时序数据变化趋势较为一致，棉花在5月（苗期）到8月初（开花盛期）有明显的上升趋势，在8月末至9月（花铃期）达到峰值。相比NDVI，红边波段构成的RENDVI₇₈₃时序曲线峰值从0.7提高到0.9，棉花区分效果更佳。（2） 11个物候特征中拟合函数最大值、生长季长度、生长季振幅、生长季结束、生长季大积分和生长季小积分对分类的贡献性最大，重要性得分分别为1.43、1.40、1.23、1.16、1.02和1.01。（3） RFC方法对特征数据集（RENDVI₇₈₃ Fit+物候特征优选组合）分类精度最佳。总体精度和Kappa系数分别为92.20%和0.92。（4）研究区内棉花分类精度达到了91.02%，种植面积约为3424 km²，占研究区总面积的24.67%。

关键词: NDVI时间序列, RENDVI时间序列, 物候特征, 袋外误差, 随机森林分类

Abstract:

Cotton is an important economic crop in China and is cultivated on a large scale in Xinjiang. The timely and accurate acquisition of cotton planting areas is significant to agricultural policy formulation and economic development. This study takes the Ugan River-Kuqa River delta oasis cotton as the main research object and uses the Sentinel-2 data of 36 scenes from 2018 to 2020 (1 scene/month) to construct a normalized vegetation index (NDVI) and red-edge normalized vegetation index (RENDVI) time-series data; The Savitzky-Golay filtering method is used to smooth and reconstruct the time-series data, and 11 phenological features are extracted. The out-of-bag method is used to optimize the 11 phenological features. Time-series data NDVI Fit and RENDVI₇₈₃ Fit, phenological feature RENDVI₇₈₃ Ph, and phenological feature optimization were combined to construct six different feature datasets, which were classified and extracted by random forest classification (RFC). The maximum likelihood and support vector machine were used to verify the classification effect. The results are as follows: (1) The changing trends of NDVI and RENDVI₇₈₃ time-series data are relatively consistent. Cotton has an obvious upward trend from May (seedling stage) to early August (blooming stage) and reaches a peak from the end of August to September (flowering and boll stage). Compared with NDVI, the peak value of the RENDVI₇₈₃ time-series curve composed of the red-edge band is increased from 0.7 to 0.9, and the cotton discrimination effect is better. (2) Among the 11 phenological features, the maximum value of the fitting function, the length of the growing season, the amplitude of the growing season, the end of the growing season, the large integral of the growing season, and the small integral of the growing season contributed the most to the classification effect, with importance scores of 1.43, 1.40, 1.23, 1.16, 1.02, and 1.01, respectively. (3) The RFC method has the best classification accuracy for the feature dataset (RENDVI₇₈₃ Fit + phenological feature optimal combination). The overall accuracy and Kappa coefficient are 92.20% and 0.92, respectively. (4) The cotton classification accuracy in the study area reached 91.02%, and the planting area was approximately 3424 km², accounting for 24.67% of the total area of the study area.

Key words: NDVI time-series, RENDVI time-series, phenological characteristics, out-of-bag error method, random forest classification

美合日阿依·莫一丁, 买买提·沙吾提, 李金朝. 基于Sentinel-2时间序列数据及物候特征的棉花种植区提取[J]. 干旱区地理, 2022, 45(6): 1847-1859.

Mihray MOYIDIN, Mamat SAWUT, LI Jinzhao. Extraction of cotton planting area based on Sentinel-2 time series data and phenological characteristics[J]. Arid Land Geography, 2022, 45(6): 1847-1859.

图/表 14

图1

表1

表2

表3

表4

图2

图3

图4

图5

图6

图7

表5

图8

图9

参考文献 33

[1]	王晓伟, 苗红萍, 刘卫东, 等. 目标价格对新疆地方棉区棉花规模的调控效应研究[J]. 中国农业资源与区划, 2017, 38(6): 150-156.
	[Wang Xiaowei, Miao Hongping, Liu Weidong, et al. The regulatory role of target price on cotton scale in Xinjiang local cotton region[J]. Chinese Journal of Agricultural Resources and Regional Planning, 2017, 38(6): 150-156.]
[2]	王琼, 王克如, 李少昆. HJ卫星数据在棉花种植面积提取中的应用研究[J]. 棉花学报, 2012, 24(6): 503-510.
	[Wang Qiong, Wang Keru, Li Shaokun. Cotton planting area extraction based on multi-temporal HJ remote sensing[J]. Cotton Science, 2012, 24(6): 503-510.]
[3]	平跃鹏, 臧淑英. 基于MODIS时间序列及物候特征的农作物分类[J]. 自然资源学报, 2016, 31(3): 503-513.
	[Ping Yuepeng, Zang Shuying. Crop identification based on MODIS NDVI time-series data and phenological characteristics[J]. Journal of Natural Resources, 2016, 31(3): 503-513.]
[4]	熊元康, 张清凌. 基于NDVI时间序列影像的天山北坡经济带农业种植结构提取[J]. 干旱区地理, 2019, 42(5): 1105-1114.
	[Xiong Yuankang, Zhang Qingling. Cropping structure extraction with NDVI time-series images in the northern Tianshan economic belt[J]. Arid Land Geography, 2019, 42(5): 1105-1114.]
[5]	黄青, 王利民, 滕飞. 利用MODIS_NDVI数据提取新疆棉花播种面积信息及长势监测方法研究[J]. 干旱地区农业研究, 2011, 29(2): 213-217.
	[Huang Qing, Wang Limin, Teng Fei. MODIS-NDVI-based monitoring of cotton planting areas and growth conditions in Xinjiang[J]. Agricultural Research in the Arid Areas, 2011, 29(2): 213-217.]
[6]	白燕英, 高聚林, 张宝林. 基于Landsat8影像时间序列NDVI的作物种植结构提取[J]. 干旱区地理, 2019, 42(4): 893-901.
	[Bai Yanying, Gao Julin, Zhang Baolin. Extraction of crop planting structure based on time-series NDVI of Landsat8 images[J]. Arid Land Geography, 2019, 42(4): 893-901.]
[7]	李美丽, 尹礼昌, 张园, 等. 基于MODIS-EVI的西南地区植被覆盖时空变化及驱动因素研究[J]. 生态学报, 2021, 41(3): 1138-1147.
	[Li Meili, Yin Lichang, Zhang Yuan, et al. Spatio-temporal dynamics of fractional vegetation coverage based on MODIS-EVI and its driving factors in southwest China[J]. Acta Ecologica Sinica, 2021, 41(3): 1138-1147.]
[8]	谢登峰, 张锦水, 潘耀忠, 等. Landsat 8和MODIS融合构建高时空分辨率数据识别秋粮作物[J] .遥感学报, 2015, 19(5): 791-805.
	[Xie Dengfeng, Zhang Jinshui, Pan Yaozhong, et al. Fusion of MODIS and Landsat 8 images to generate high spatial-temporal resolution data for mapping autumn crop distribution[J]. Journal of Remote Sensing, 2015, 19(5): 791-805.]
[9]	柳文杰, 曾永年, 张猛. 融合时间序列环境卫星数据与物候特征的水稻种植区提取[J]. 遥感学报, 2018, 22(3): 381-391.
	[Liu Wenjie, Zeng Yongnian, Zhang Meng. Mapping rice paddy distribution by using time series HJ blend data and phenological parameters[J]. Journal of Remote Sensing, 2018, 22(3): 381-391.]
[10]	黄林生, 江静, 黄文江, 等. Sentinel-2影像和BP神经网络结合的小麦条锈病监测方法[J]. 农业工程学报, 2019, 35(17): 178-185.
	[Huang Linsheng, Jiang Jing, Huang Wenjiang, et al. Wheat yellow rust monitoring method based on Sentinel-2 image and BPNN model[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(17): 178-185.]
[11]	Sonobe R, Yamaya Y, Tani H, et al. Assessing the suitability of data from Sentinel-1A and 2A for crop classification[J]. Giscience & Remote Sensing, 2017, 54(6): 918-938.
[12]	易秋香. 棉花不同生长阶段Sentinel-2卫星植被指数变化特征[J]. 中国棉花, 2019, 46(8): 13-18.
	[Yi Qiuxiang. The change features of Sentinel-2 vegetation indices with cotton growth[J]. China Cotton, 2019, 46(8): 13-18.]
[13]	张磊, 宫兆宁, 王启为, 等. Sentinel-2影像多特征优选的黄河三角洲湿地信息提取[J]. 遥感学报, 2019, 23(2): 313-326.
	[Zhang Lei, Gong Zhaoning, Wang Qiwei, et al. Wetland mapping of Yellow River Delta wetlands based on multi-feature optimization of Sentinel-2 images[J]. Journal of Remote Sensing, 2019, 23(2): 313-326.]
[14]	何云, 黄翀, 李贺, 等. 基于Sentinel-2A影像特征优选的随机森林土地覆盖分类[J]. 资源科学, 2019, 41(5): 992-1001. doi: 10.18402/resci.2019.05.15
	[He Yun, Huang Chong, Li He, et al. Land-cover classification of random forest based on Sentinel-2A image feature optimization[J]. Resources Science, 2019, 41(5): 992-1001.] doi: 10.18402/resci.2019.05.15
[15]	马依拉·热合曼, 买买提·沙吾提, 尼格拉·塔什甫拉提, 等. 基于遥感与GIS的渭—库绿洲生态系统服务价值时空变化研究[J]. 生态学报, 2018, 38(16): 5938-5951.
	[Reheman Mayila, Sawuti Mamat, Taxipulati Nigela, et al. The ecosystem service value spatial-temporal changes in the Ugan-Kuqa River Delta Oasis based on RS and GIS[J]. Acta Ecologica Sinica, 2018, 38(16): 5938-5951.]
[16]	安赛日丁·艾尔肯, 买买提·沙吾提, 李金朝. 基于冠层光谱特征的长绒棉叶绿素含量估算[J]. 激光与光电子学进展, 2022, 59(5): 1-11.
	[Arkin Ansardin, Sawut Mamat, Li Jinzhao. Estimation of chlorophyll content of long-staple cotton based on canopy spectrum characteristics[J]. Laser & Optoelectronics Progress, 2022, 59(5): 1-11.]
[17]	曹霖. 基于Sentinel-2影像的延庆区森林蓄积量遥感估测研究[D]. 北京: 北京林业大学, 2019.
	[Cao Lin. Estimation of forest stock volume in Yanqing District based on Sentinel-2 image[D]. Beijing: Beijing Forestry University, 2019.]
[18]	Sharma L K, Bu H G, Denton A, et al. Active-optical sensors using red NDVI compared to red edge NDVI for prediction of corn grain yield in north Dakota, U.S.A[J]. Sensors, 2015, 15(11): 27832-27853. doi: 10.3390/s151127832 pmid: 26540057
[19]	刘德强. 基于MODIS时间序列数据的广西壮族自治区作物分类研究[D]. 南宁: 广西师范学院, 2017.
	[Liu Deqiang. Study on crop classification of Guangxi Zhuang Autonomous Region on MODIS time series data[D]. Nanning: Guangxi Teacher Education University, 2017.]
[20]	宋春桥, 柯灵红, 游松财, 等. 基于TIMESAT的3种时序NDVI拟合方法比较研究——以藏北草地为例[J]. 遥感技术与应用, 2011, 26(2): 147-155.
	[Song Chunqiao, Ke Linghong, You Songcai, et al. Comparison of three NDVI time-series fitting methods based on TIMESAT: Taking the grass land in northern Tibet as case[J]. Remote Sensing Technology and Application, 2011, 26(2): 147-155.]
[21]	孟祥珍, 刘丹, 黄可, 等. 基于植被物候特征的互花米草提取方法研究——以长三角湿地为例[J]. 海洋通报, 2021, 40(5): 591-600.
	[Meng Xiangzhen, Liu Dan, Huang Ke, et al. Extraction method of Spartina alterniflora based on vegetation phenology characteristics: A case study of wetlands in the Changjiang River Delta[J]. Marine Science Bulletin, 2021, 40(5): 591-600.]
[22]	张磊, 宫兆宁, 王启为, 等. Sentinel-2影像多特征优选的黄河三角洲湿地信息提取[J]. 遥感学报, 2019, 23(2): 313-326.
	[Zhang Lei, Gong Zhaoning, Wang Qiwei, et al. Wetland mapping of Yellow River Delta wetlands based on multi-feature optimization of Sentinel-2 images[J]. Journal of Remote Sensing, 2019, 23(2): 313-326.]
[23]	孙琳, 宋爱红, 高文秀. 基于最大似然法和SVM法的太湖流域HJ-1B影像分类[J]. 测绘信息与工程, 2012, 37(3): 30-33.
	[Sun Lin, Song Aihong, Gao Wenxiu. Land use/cover classification in Taihu Lake Basin with HJ-1B image: Comparing maximum likelihood classification and support vector machine methods[J]. Journal of Geomatics, 2012, 37(3): 30-33.]
[24]	Pal M, Foody G M. Feature selection for classification of hyperspectral data by SVM[J]. IEEE Transactions on Geoscience & Remote Sensing, 2010, 48(5): 2297-2307.
[25]	王利民, 刘佳, 杨玲波, 等. 随机森林方法在玉米-大豆精细识别中的应用[J]. 作物学报, 2018, 44(4): 569-580. doi: 10.3724/SP.J.1006.2018.00569
	[Wang Limin, Liu Jia, Yang Lingbo, et al. Application of random forest method in maize-soybean accurate identification[J]. Acta Agronomica Sinica, 2018, 44(4): 569-580.] doi: 10.3724/SP.J.1006.2018.00569
[26]	郝鹏宇. 基于多时相遥感数据的农作物分类研究[D]. 北京: 中国科学院大学, 2017.
	[Hao Pengyu. Crop classification using time-series remote sensing data[D]. Beijing: University of Chinese Academy of Sciences, 2017.]
[27]	伊尔潘·安尼瓦尔, 买买提·沙吾提, 买合木提·巴拉提. 基于GF-2影像和Unet模型的棉花分布识别[J/OL]. 自然资源遥感.[2022-04-26]. http://kns.cnki.net/kcms/detail/10.1759.P.20211109.1111.006.html.
	[Anwar Erpan, Sawut Mamat, Balati Maihemuti. Cotton distribution recognition based on GF-2 image and Unet medel[J/OL]. Remote Sensing for Natural Resources.[2022-04-26]. http://kns.cnki.net/kcms/detail/10.1759.P.20211109.1111.006.html. ]
[28]	韩振锋. 融合环境卫星和MODIS数据的甘蔗种植区提取[D]. 南宁: 南宁师范大学, 2019.
	[Han Zhenfeng. Sugarcane extraction based on fused HJ satellite and MODIS data[D]. Nanning: Nanning Normal University, 2019.]
[29]	陈波, 王加胜. 基于Landsat影像的马龙区水稻种植区遥感提取与分析[J]. 江苏科技信息, 2021, 38(5): 56-58.
	[Chen Bo, Wang Jiasheng. Remote sensing extraction and analysis of rice planting areas in Malong District based on Landsat images[J]. 2021, 38(5): 56-58.]
[30]	彭壮壮, 武永斌, 朱梦豪, 等. 基于时序Sentinel-2影像的农作物分类研究[J]. 测绘与空间地理信息, 2021, 44(12): 81-84.
	[Peng Zhuangzhuang, Wu Yongbin, Zhu Menghao, et al. Research on crop classification based on time series Sentinel-2 images[J]. Geomatics & Spatial Information Technology, 2021, 44(12): 81-84.]
[31]	周增光, 唐娉. 基于质量权重的Savitzky-Golay时间序列滤波方法[J]. 遥感技术与应用, 2013, 28(2): 232-239.
	[Zhou Zengguang, Tang Ping. VI-quality-based Savitzky-Golay method for time series data[J]. Remote Sensing Technology and Application, 2013, 28(2): 232-239.]
[32]	姚茂林, 江洪, 张丽玉. 基于Sentinel-2影像红边光谱指数与特征优选的竹林提取研究[J/OL]. 海南大学学报(自然科学版).[2022-04-25]. http://kns.cnki.net/kcms/detail/46.1013.N.20220421.1203.002.html.
	[Yao Maolin, Jiang Hong, Zhang Liyu. Bamboo extraction using improved spectral indices and random forest gini index[J/OL]. Natural Science Journal of Hainan University.[2022-04-25]. http://kns.cnki.net/kcms/detail/46.1013.N.20220421.1203.002.html. ]
[33]	古丽努尔·依沙克, 买买提·沙吾提, 马春玥. 基于多时相双极化SAR数据的作物种植面积提取[J]. 作物学报, 2020, 46(7): 1099-1111. doi: 10.3724/SP.J.1006.2020.94134
	[Ishake Gulnur, Shawuti Mamat, Ma Chunyue. Extraction of crop acreage based on multi-temporal and dual-polarization SAR data[J]. Acta Agronomica Sinica, 2020, 46(7): 1099-1111.] doi: 10.3724/SP.J.1006.2020.94134

特征	指数全称	计算公式
光谱特征	归一化植被指数（NDVI）	(B8-B4)/(B8+B4)
红边特征	红边1归一化植被指数（RENDVI₇₀₅）	(B8-B5)/(B8+B5)
	红边2归一化植被指数（RENDVI₇₄₀）	(B8-B6)/(B8+B6)
	红边3归一化植被指数（RENDVI₇₈₃）	(B8-B7)/(B8+B7)

物候特征	特征含义
生长季开始	左边NDVI增加到用户定义的拟合函数值的时刻
生长季结束	右边NDVI减少到用户定义的拟合函数值的时刻
生长季长度	生长结束与生长季开始的时间差值
生长季基值	拟合函数左右2部分NDVI最小值的平均值
生长季中期	左边增加到80％和右边降低到80％的平均时间
拟合函数最大值	NDVI最大值
生长季振幅	最大值和基值之间均值的差
生长季左导数	NDVI增加到左边振幅80％的时间
生长季右导数	NDVI减少到右边振幅80%的时间
生长季大积分	从生长季开始到生长季结束的区域面积
生长季小积分	NDVI拟合曲线与生长季开始到结束的分量

方案	特征集	波段数目
A	重构后的NDVI时序数据（NDVI Fit）	36
B	重构后的RENDVI₇₈₃时序数据（RENDVI₇₈₃ Fit）	36
C	重构后的RENDVI₇₈₃ 11个物候特征（RENDVI₇₈₃ Ph）	11
D	物候特征优选组合	6
E	重构后的RENDVI₇₈₃时序数据+11个物候特征（RENDVI₇₈₃ Fit+RENDVI₇₈₃ Ph）	47
F	重构后的RENDVI₇₈₃时序数据+物候特征优选组合	42

分类方法	分类精度	A方案	B方案	C方案	D方案	E方案	F方案
MLC	制图精度	79.92	82.06	79.36	82.89	85.05	86.66
MLC	用户精度	71.70	78.39	76.54	78.51	81.92	80.59
SVM	制图精度	83.01	85.03	86.11	90.09	89.33	90.88
SVM	用户精度	76.22	78.21	79.55	82.56	84.17	88.44
RFC	制图精度	87.14	89.07	90.54	91.07	87.09	88.89
RFC	用户精度	79.66	82.32	83.38	85.41	89.36	94.12

基于Sentinel-2时间序列数据及物候特征的棉花种植区提取

Extraction of cotton planting area based on Sentinel-2 time series data and phenological characteristics

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 33

相关文章 1

Metrics

本文评价

推荐阅读 0

卫星名称	获取时间	影像数量	波段数	空间分辨率/m
Sentinel-2	2018年1—12月	12	12	10
	2019年1—12月	12	12	10
	2020年1—12月	12	12	10