摘要
柑橘普遍种植于中国南方地区,受天气多云多雨、种植类型复杂等因素影响,利用光谱信息直接识别柑橘果园信息存在一定困难。该研究根据柑橘特有的物候特征,提出了"柑橘在其果实生长膨大过程中柑橘果园的植被信息可能减弱"的假设。根据此特征提出柑橘果园信息识别方法,确定关键时间窗口的阈值,并以广西壮族自治区南宁市武鸣区为研究区,开展柑橘果园信息遥感识别实证研究。首先,获取2018年研究区多时相Sentinel-2遥感影像,构建归一化植被指数(Normalized Difference Vegetation Index,NDVI)、绿色归一化植被指数(Green Normalized Difference Vegetation Index,GNDVI)、差值植被指数(Difference Vegetation Index,DVI)和红边波段指数(Sentinel-derived Red-edge Spectral Indices,RESI)等多个植被光谱指数;其次,根据地面样本点信息,对比不同植被类型在不同时期的遥感植被信息差异,进而确定柑橘果园识别的最优特征。研究结果表明,柑橘果园与研究区其他主要作物类型(如甘蔗、香蕉、玉米、水稻等)没有明显的光谱特征差异,但研究区多时相遥感植被指数显示10月的柑橘果园NDVI相比11月出现明显低值0.47,且明显低于其他作物类型;10月的柑橘果园GNDVI也出现了低值0.43,但与其他月份相比差异不明显;而柑橘果园DVI的离散程度低,分离性不强。根据作物物候历,9—10月为柑橘果实迅速膨大期,这验证了该研究提出的科学假设,即该时期柑橘果园的植被信息会减弱。柑橘果实膨大期不同植被指数的离散程度差异明显,NDVI离散程度最高,差异性最强。根据10月柑橘果园NDVI的物候特征,进一步构建归一化指数,通过阈值法识别研究区柑橘果园空间分布,该识别方法的总体精度达到82.75%,优于其他植被指数的识别结果,研究结果可为柑橘果园信息遥感识别研究提供较好的理论与实践支撑。
Citrus has been widely planted in the south of China in recent years.However,the multiple crop types and frequent cloud cover locally have posed a great challenge to the direct identification of the large-scale citrus orchard using the spectral information.In this study,a systematic approach was developed to identify the citrus orchard using the phenological characteristics of citrus during the fruit expansion stage,where the threshold value was determined to the key time window.Taking the Wuming District of Nanning Guangxi Zhuang Autonomous Region in southwest China as the study area,an empirical investigation was carried out using the Google Earth Engine platform.1751 ground samples were also collected in the field for validation.Meanwhile,the cloud coverage assessment was performed on the Sentinel-2 images over the whole year of 2018.According to the citrus phenology in the study area,the characteristics of flowering were not outstanding in the first half of 2018(the flowering period of citrus),while the key identification features were found in the second half of 2018(the peak stage of citrus fruit growth and expansion).As such,the second half of 2018 was determined as the study period,considering the data availability and citrus phenological stage.Then,a multi-temporal image dataset was obtained from August to December.The specific procedure was as follows.Firstly,some indices were calculated using the time series Sentinel-2 data in 2018,including the multiple vegetation indices(e.g.Normalized Difference Vegetation Index(NDVI),Green Normalized Difference Vegetation Index(GNDVI),Difference Vegetation Index(DVI)),and Sentinel-derived Red-Edge Spectral Indices(RESI).The vegetation information of multi-temporal remote sensing was then obtained for the entire study area.Secondly,the vegetation information of various planting types was compared in different growing periods,according to the measured ground data.The results showed that there was no outstanding spectral difference between the citrus orchards and other vegetation types(e.g.sugarcane,banana,maize,and rice)in the study area.However,the vegetation indices of multi-temporal remote sensing presented that the NDVI of 0.47 for the citrus orchards was distinctly lower than that for the other vegetation types in October.There was also a lower GNDVI of citrus orchards in October and November.But there was no difference in the DVI of citrus orchards from other vegetation.Furthermore,the fruit expansion stage of citrus was located from September to October,indicating weak vegetation information of citrus.Nevertheless,there were significant differences among different vegetation indices,where the NDVI presented the highest.In addition,a renormalization of vegetation indices was further constructed to identify the spatial distribution of citrus in the study area by a threshold,according to the NDVI in October.The overall accuracy of citrus orchards reached 82.75%using the renormalization of NDVI,where the Kappa coefficient was 0.66,indicating a better identification,compared with the rest of renormalized vegetation indices(GNDVI,DVI,and RESI).The NDVI presented more complete identification for the citrus orchards,whereas,the GNDVI performed more fragmented identification.Consequently,the total area of citrus planting was calculated as 3.42×10^(4) hm^(2) using the phenological parameters of citrus orchards.This finding can provide strong theoretical and practical support to timely mapping the citrus orchards using remote sensing.
作者
梁晨欣
黄启厅
王思
王聪
余强毅
吴文斌
Liang Chenxin;Huang Qiting;Wang Si;Wang Cong;Yu Qiangyi;Wu Wenbin(Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences/Key Laboratory of Agricultural Remote Sensing,Ministry of Agriculture and Rural Affairs,Beijing 100081,China;Institute of Agricultural Science and Technology,Guangxi Academy of Agricultural Sciences,Nanning 530000,China;Institute of Remote Sensing Application,Sichuan Academy of Agricultural Sciences,Chengdu 610066,China)
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2021年第24期168-176,共9页
Transactions of the Chinese Society of Agricultural Engineering
基金
中国农科院-广西农科院协同创新项目课题“广西智慧耕地关键技术研发与集成示范”(XTCX2019026-2)。