基于Landsat-8和Sentinel-2A多光谱影像的陆良县日光温室面积估算

Estimation of Solar Greenhouse Area in Luliang County Based on Landsat-8 and Sentinel-2A Multispectral Images

【目的】选取云南省陆良县,开展基于Landsat-8 OLI和Sentinel-2A MSI影像的设施种植区域监测研究,及时为农业生产管理提供准确的设施农业空间信息。【方法】首先分析日光温室及其他典型地物的遥感特征,然后使用目视解译以及最小距离法、马氏距离法、最大似然法、支持向量机等4种监督分类方法提取设施种植空间信息,再使用亚米级Google earth历史影像为底图建立验证样区。通过目视解译提取样区设施种植区域信息,以此验证Landsat-8和Sentinel-2A影像的监测精度,选择最优监测方法。【结果】使用Landsat-8影像,基于目视解译及4种监督分类方法的精度依次为97. 54%、77. 31%、91. 35%、91. 89%和83. 15%;使用Sentinel-2A影像,基于目视解译及4种监督分类方法的精度依次98. 82%、80. 80%、87. 01%、93. 89%和85. 41%。【结论】①2种影像的日光温室监测均适于使用最大似然法,其估算精度与目视解译精度误差分别为5. 65%和4. 93%;②遥感监测显示,2016年陆良县日光温室主要集中于该县湖积平原区。基于Landsat-8影像监测的面积为5381. 62 hm2,基于Sentinel-2A影像监测的面积为5347. 84 hm2。

【Objective】This paper aimed to provide timely and accurate agricultural spatial information for agricultural production management,Luliang county in Yunnan province was selected to carry out surveillance research facilities planting area based on the Landsat-8 OLI and Sentinel-2A MSI images.【Method】Firstly,the remote sensing features of solar greenhouse and other typical landforms in the region were analyzed,secondly,visual interpretation and four kinds of supervised classification methods such as minimum distance method,Mahalanobis distance method,maximum likelihood method and support vector machine were used to extract facility planting spatial informations of facility planting,then sub-level Google earth historical images were used to create verification sample area for the basemap.Planting area information of sampling facilities by visual interpretation was extracted,which was used to verify the monitoring accuracy of the Landsat-8 and Sentinel-2A images and choose the optimal monitoring method.【Result】According to the above methods,the visual interpretation area accuracy based on OLI images was 97.54%,and the accuracy of the four kinds of supervised classification methods was 77.31%,91.35%,91.89%and 83.15%,respectively;The precision of using Sentinel-2A image based on visual interpretation and four supervised classification methods was 98.82%,80.80%,87.01%,93.89%and 85.41%,respectively.【Conclusion】(i)Two kinds of images of greenhouse monitoring were suitable for the maximum likelihood method,and the error of estimation accuracy and visual interpretation accuracy were 5.65%and 4.93%.(ii)Remote sensing monitoring showed that the solar greenhouse in Luliang county mainly concentrated in the lacustrine plains area in 2016.The area based on Landsat-8 image surveillance was 5381.62 hm^2,and the area based on Sentinel-2A image surveillance was 5347.8 4 hm^2.