基于青铜峡灌区的土地利用数据源适用性研究

Study on the Applicability of Land Use Data Source Based on Qingtongxia Irrigation District

摘要: 快速准确地获取灌区土地利用信息是进行农作物精细分类、土壤墒情监测、作物长势监测以及作物估产的前提。本文以我国宁夏回族自治区青铜峡市的青铜峡灌区为研究对象，以高分一号(GF1-WFV)影像、高分六号(GF6-WFV)影像以及Sentinel-2数据为数据源，对比分析三类影像数据的建模精度，选取精度最高的一类影像并基于决策树方法进行土地利用信息提取，将土地利用类别划分为作物覆盖农田、裸露农田、水体、田间道路和建筑物五类。结果显示：高分六号(GF6-WFV)数据建模精度最高，三层时模型精度98.98%，验证精度100%；高分一号(GF1-WFV)数据建模精度最低，三层时模型精度95.92%，验证精度92.86%；高分六号(GF6-WFV)与Sentinel-2对比，波段数量以及红边波段的设置对信息的提取起到的作用强于分辨率，在波段信息相同的情况下，分辨率对提取的精度影响明显。

Abstract: Fast and accurate acquisition of land use information in irrigation area is the premise of fine classification of crops, soil moisture monitoring, crop growth monitoring and crop yield estimation. In this paper, Qingtongxia Irrigation District in Qingtongxia City, Ningxia Hui Autonomous Region is taken as the research object, and GF1-WFV, GF6-WFV and Sentinel-2 data are taken as data sources, and the modeling accuracy of three types of image data is comparatively analyzed. The image with the highest accuracy is selected and the land use information is extracted based on the decision tree method, and the land use categories are divided into five categories: crop-covered farmland, bare farmland, water body, field roads and buildings. The results show that the modeling accuracy of GF6-WFV is the highest, with the model accuracy of 98.98% and the verification accuracy of 100% in three layers. The modeling accuracy of GF1-WFV is the lowest, with the model accuracy of 95.92% and the verification accuracy of 92.86% in three layers. Compared with Sentinel-2, the number of bands and the setting of red band play a stronger role in information extraction than resolution, and the resolution has obvious influence on the extraction accuracy under the same band information.