基于多时相HJ-1B CCD影像的玉米倒伏灾情遥感监测

Remote Sensing Monitoring of Maize Lodging Disaster with Multi-Temporal HJ-1B CCD Image

【目的】通过对倒伏发生前后的多时相HJ-1B卫星CCD多光谱影像植被指数的变化分析,实现区域尺度下的玉米倒伏受灾范围监测和灾情评估。【方法】以2015年8月3日因大风暴雨导致大面积玉米倒伏的河北省藁城市为研究区,提取HJ-1B CCD多光谱影像的多种植被指数,根据多时相植被指数变化量与实测倒伏样本的关联分析,筛选玉米倒伏敏感植被指数,采用正态(偏正态)统计理论的双阈值划分策略划定阈值,构建基于植被指数变化分析的玉米倒伏灾情遥感监测模型,评价玉米倒伏灾情严重程度,并基于野外实测样本进行精度评价。【结果】基于城镇、水体、倒伏与未倒伏玉米的HJ-1B CCD2多光谱反射率曲线可得,玉米与城镇建筑物和水体的光谱反射率存在较大差异,在可见光波段,倒伏玉米要高于未倒伏玉米的反射率,然而在近红外波段(830 nm),未倒伏玉米略高于倒伏玉米的反射率。由已提取出玉米种植区域后的前后两期HJ星的差值影像分析可得其像元整体分布趋势基本符合正态(偏正态)分布。由相关性分析可得,倒伏前后的比值植被指数(RVI)差值与玉米倒伏比例具有最高的相关性(R=0.9377),而且倒伏越严重,其RVI差值越大;通过野外实测样本与模型验证结果进行混淆矩阵分析,总体分类精度达到85.7%,Kappa系数为0.804;研究区玉米倒伏遥感空间制图结果与当地农业技术推广站监测结果基本一致。【结论】倒伏发生后,其玉米长势以及灾后恢复程度均存在较大差异,因此,基于玉米倒伏前后RVI差值的遥感监测模型能有效反映不同倒伏程度的冠层群体变化信息以及长势恢复情况,能在区域尺度下实现玉米倒伏受灾范围监测和灾情等级评估。

【Objective】The aim of this study is to analyze the variance of a variety of vegetation indices derived from HJ-1B（One of the Small Satellite Constellations for Environment and Disaster Monitoring and Forecasting, HJ-1A/1B/1C） CCD multi-temporal images in pre- and post-lodging, in order to achieve the objective for measuring the area of maize lodging and evaluation of disaster rank. 【Method】This paper took Gaocheng, Hebei Province as a research area where suffered a heavy rainstorm which led to a large-scale maize lodging on August 3rd, 2015. A variety of vegetation indices derived from HJ-1B CCD multi-temporal images in pre- and post-lodging period were adopted to conduct the sensitivity analysis among various vegetation index variations（？VI_i, the value of vegetation index before lodging minus that of after lodging） and the lodging characteristic values measured in field（Lodging proportion for research areas） to seek the preferable vegetation indices for maize lodging monitoring. The double threshold partition strategy of normal statistical theory was adopted to determine the thresholds and a remote sensing monitoring model for maize lodging was built based on the analysis of the difference of vegetation indices pre- and post-lodging to evaluate the disaster degree of the maize lodging. The accurate assessment of the model was conducted by comparing the in-field measured data and predicted results from the built model at last. 【Result】Based on the features of HJ–1B CCD2 multispectral reflectance curve of the town, water, lodging and no lodging maize, there was a difference between maize and building and water in special reflectance. In the visible spectrum, spectral reflectance of lodging maize was higher than that of no lodging maize, therefore, it was opposite in near infrared wave. From the analysis of RVI difference image, it nearly conform to normal distribution（skewed normal distribution）. The result of correlation analysis suggested that RVI（ratio vegetation index） had the best correlation with the proportion of maize lodging（R=0.9377）. It means that the more serious of the lodging, the higher the RVI difference. It showed that the maize growth status was obviously affected by lodging, the more severe of the lodging, and it was more difficult for maize to restore to its normal state before lodging and thus causing bigger values of the RVI differences. Generally, RVI differences for different small areas showed a normal（or skew） distribution in general from the statistical analysis of the image of RVI difference. The data measured infield about lodging proportion and the values predicted from the model were adopted to carry out confusion matrix analysis, it showed that a good overall classification accuracy up to 85.7% while the Kappa coefficient was 0.804. The spatial distribution of maize lodging area and the degree of disaster could be mapped via the constructed maize lodging remote sensing model based on the differences of vegetable index derived from the double threshold partition strategy which based on normal（skew） statistical theory. The result from remote sensing mapping was basically consistent with the monitoring data from the local agricultural technical extension station. 【Conclusion】After occurrence of lodging of maize, there was a big difference between the maize growth and the resumption of maize after disaster. Therefore, it was concluded that the remote sensing monitoring model based on the differences of RVI measured in pre- and post-lodging period of maize is a feasible method for reflecting the information which indicate the changes of canopy groups described by different lodging degrees and can satisfy the need for measuring the area of maize lodging and evaluation of disaster rank.