【作者】 李伟；

【导师】 姚霞；

【作者基本信息】 南京农业大学 ， 作物栽培学与耕作学， 2020， 硕士

【摘要】 叶面积指数(leafarea index,LAI)是作物群体生长状况的重要指标,传统的LAI获取方法费时耗力,而利用卫星遥感能够实现大范围小麦LAI的监测。随着秸秆还田在稻麦轮作区的推广与应用,小麦生长过程中的背景从传统的单一土壤背景变为“土壤-秸秆”混合背景,另外,当前一些发展中国家农田长势不均、地块破碎等因素增加卫星影像中的混合像元,均为大面积准确监测作物LAI增加了难度。前人研究发现红边信息在监测作物长势时能够缓解前期的土壤背景影响和后期高叶面积指数导致的饱和问题,但是如何消除“土壤-秸秆”混合背景影响的研究不足。传统红边区域的斜率在监测LAI时往往聚焦于植被光谱曲线红边特征的表达,并未考虑背景对该特征本身的影响。对此,本研究基于Sentinel-2与Planet影像数据,结合地面获取的小麦LAI数据、背景高光谱数据,开展影像融合与“秸秆-土壤”混合背景影响消除的研究,达到准确监测秸秆还田背景下的小麦LAI的目的:为解决目前国内外高分辨率影像缺少红边波段的问题,本研究综合应用加权解混(weight-and-unmixing,Wu)与多分辨率多光谱估计的超分辨率(super-resolution for multispectral Multiresolution Estimation,SupReME),即 Wu-SupReME,通过融合两个新数据源Sentinel-2与Planet的光谱和空间优势,首次生成带有红边波段的3米高分辨率融合影像产品。融合影像与Sentinel-2波段之间的相关性均大于0.98,表明所得融合影像较好地继承Sentinel-2的光谱优势。在估算长势不均的小麦LAI时,融合影像的所有模型构建结果(R2)与交叉验证结果(RRMSE)均优于原始影像(R2提高了 0.0215～0.3890,RRMSE降低了0.0018～0.1462),表明融合影像不仅继承Sentinel-2的光谱优势,同时保留了 Planet的空间优势。本研究为植被定量遥感中准确反演长势参数提供了高分辨率的红边卫星产品,并为多源数据融合提供技术参考。为解决秸秆还田管理技术中传统的土壤背景变为“土壤-秸秆”多端元的混合背景问题,本研究利用播种后、出苗前的融合影像数据获取“秸秆-土壤”背景的光谱曲线,利用各生育期融合影像获取大田小麦冠层的光谱信息,通过斜率的差值计算去除光谱信息中“土壤-秸秆”背景成分,确定由小麦引起的斜率增量,即斜率差。进一步利用斜率差优化传统植被指数,以减少混合背景的影响,提高植被指数估算LAI的精度。结果表明,斜率差、优化后的植被指数相对于先前未作改变的斜率、植被指数具有更高的精度,其中,斜率差ΔKwheat NIR-RE2(R2=0.79,RRMSE=19.68%)和优化后的植被指数 MEVI2(R2=0.80,RRMSE=19.25%)表现最佳。利用优化前后的MEVI2绘制LAI分布图,对比结果显示优化后的MEVI2能够缓解“秸秆-土壤”混合背景引起的高估现象,同时在后期能够缓解高LAI引起的饱和现象。本研究提出了一种利用两个生育期三类对象(包括“小麦-秸秆-土壤”和“秸秆-土壤”)的光谱曲线构建斜率差与植被指数的方法,去除了“秸秆-土壤”混合背景影响,提高了小麦LAI的估算精度,为多对象光谱信息共同作用改善植被监测提供了参考。更多还原

【Abstract】 Leaf area index(LAI)is an essential index of wheat population growth.The traditional method of LAI acquisition is time-consuming and labor-consuming,while satellite remote sensing can achieve large-scale monitoring of wheat LAI.Uneven growth of crops and fragmentations of fields in developing countries increase the mixed pixels in satellite images.At the same time,the red edge information can alleviate the saturation problem caused by the soil background and high LAI in the early stage of crop growth.With the popularization and application of straw returning to the field in the rice-wheat-rotation area,during the growth of wheat,the background has changed from the unique soil background to the "soil-straw" mixed background.However,there is a lack of researches on reducing the influence of mixed background and improving the accuracy of LAI estimation.The slope of the traditional red edge region often focuses on the expression of red edge feature of vegetation spectrum curve when monitoring LAI,without considering the influence of background on the feature itself.This study used Sentinel-2 and Planet image,together with wheat LAI and the spectrum of background acquired from wheat fields,to carry out the study about image fusion and elimination of "straw-soil" background:In order to solve the problem of lacking high-resolution red edge satellite products,this study applies weighted-and-unmixing and super-resolution for multispectral Multiresolution Estimation(SupReME),named Wu-SupReME.A high-resolution RE product was generated by fusing Sentinel-2 spectral advantage and Planet spatial advantage.The resultant fused image is highly correlated(R2>0.98)with the Sentinel2 image and clearly illustrates the persistent advantages of such products.This fused image was significantly more accurate than the originals when used to predict heterogeneous wheat LAI and,therefore,clearly illustrated the persistence of Sentinel2 spectral and Planet spatial advantage.This study provided method reference for multisource data fusion and image products for accurate parameter inversion in quantitative remote sensing of vegetation.In order to improve the LAI estimation accuracy,which is affected by the mixed"soil-straw" background,this study used the fused images between sowing and emergence to obtain the spectral curve of the "straw-soil" background.The spectral information of wheat canopy obtained by using the fusion image at each growth stage.To remove the influence of the "straw-soil" background,we calculated the slope difference for increment caused by wheat.Based on traditional VIs,using slope difference to optimize VIs can further reduce the influence of mixed background and improve the accuracy of LAI estimation.The results showed that the optimized slope and VIs have higher accuracy than that before optimization.The optimized slope△Kwheat NIR-RE2(R2=0.79,RRMSE=19.68%)and MEIV2(R2=0.80,RRMSE=19.25%)have the best performance separately.The comparison using MEVI2 before and after optimization to draw the LAI distribution map showed that the optimized MEIV2 not only reduces the influence of mixed background but also alleviates the saturation problem of red edge region slope with the increasing LAI.This study proposed a method to calculate slope difference and optimize VIs by using the spectral curves of two objects,including the "wheat-straw-soil" spectral curve and "straw-soil" spectral curve,to remove the influence of "straw-soil" mixture background and improve the accuracy of LAI estimation.This study provides a reference for multi-object spectral information to improve vegetation monitoring.更多还原