冬小麦面积遥感识别精度与空间分辨率的关系

不同空间分辨率农作物面积识别精度是农情遥感监测数据源选择的依据。该文采用WFV(wide field view)、MODIS(moderate-resolution imaging spectroradiometer)、OLI(operational land imager)、Google Earth影像,在天津市武清区选择了12 km×14 km的冬小麦种植区作为研究区域,采用目视识别的方法,分析了2、5、10、15、30、100、250 m共7个空间分辨率尺度下冬小麦面积识别精度与遥感数据分辨率、农田景观破碎度之间的关系。结果表明,随着空间分辨率由2 m变化到250 m,冬小麦面积识别的总体精度逐步由98.6%降低到70.1%,精度降低28.5%;面积数量比例由5.5%扩大到110.6%,误差增加105.1个百分点;面积精度呈明显下降趋势,数量误差呈明显增加趋势,数量误差的增加速度高于精度下降的趋势。高、中、低3个景观破碎度条件下,随着分辨率由2 m降低到250 m,作物识别精度分别降低了72.8、63.2和47.0个百分点,破碎度的增加导致面积识别精度下降速度更快;同等分辨率下,破碎度越高的地区面积识别精度越低。像元内冬小麦占比与可识别能力密切相关,像元占比达到45.0%以上时才能够被正确识别为冬小麦类型,像元尺度降低导致细小斑块丢失是造成面积识别与数量精度降低的主要原因。像元空间分辨率越高,冬小麦像元的光谱一致性越强,越有利于冬小麦分类精度的提高。针对农情遥感监测业务运行的需要,上述研究结果可以作为区域范围不同用户精度要求前提下遥感数据源选择的依据。

Relationship between accuracy of winter wheat area remote sensing identification and spatial resolution

Abstract: Intensive study on the relationship among remote sensing data source spatial resolution, farmland landscape fragmentation index, and crop area identification accuracy is the basis for the selection and optimization of crop area monitoring data source. A 12 km × 14 km winter wheat planting area in Wuqing District, Tianjin City was taken as the study area. The visual interpretation method was adopted to compare the crop area estimation accuracies of the data with 8 spatial resolution levels of 0.3, 2, 5, 10, 15, 30, 100 and 250 m, and the Google Earth images were taken as the major data sources which were supported with 16 m GF-1/WFV (wide field view), 30 m LandSat-8/OLI (operational land imager), and 250 m EOS/MODIS (moderate-resolution imaging spectroradiometer) data. Meanwhile, the paper also analyzed the impact of fragmentation indices on area accuracy under the scale effect, the change of the proportion of winter wheat area, the change of crop patches, and the change laws of classification figure DN (digital number) standard deviation. The result shows that, the visual interpreting result of remote-sensing images with the resolution of 0.3 m is a "real value image". Along with the change of spatial resolution from 2 to 250 m, the winter wheat identification accuracies gradually decrease from 98.6% to 70.1%, showing a gradually downward trend. Meanwhile, Kappa coefficient also gradually decreases from 0.96 to 0.39, indicating that the winter wheat classification accuracy is closely related to the resolutions of remote sensing images. Purely from the perspective of winter wheat area estimation accuracy, the decrease speed of resolution is faster than the decrease speed of classification identification accuracy. When the image resolution decreases from 100 to 250 m, even though the slight decrease of classification identification accuracy from 70.3% to 70.1%, the error of winter wheat area estimation increases significantly from 86.0% to 110.6%. The main reason is that relatively low spatial resolution causes the winter wheat patches with small area to be missed. The study area is divided into 3 areas with high, medium and low fragmentation indices. Along with the increase of fragmentation index of farmland landscape, the crop area identification accuracy decreases, and the image accuracy with spatial resolution of 2 m decreases from 98.8% to 94.2% and then to 70.7%. Along with the decrease of spatial resolution, the decrease speed of identification accuracy of winter wheat area in the regions with higher fragmentation index is faster than that of the regions with lower fragmentation index. With the spatial resolution decreasing from 2 to 250 m, the decreased magnitude in the regions with higher fragmentation index is 51.5%, and that in the regions with lower fragmentation index is 46.1%. The main reason is that under the condition of higher fragmentation index, along with the decrease of resolution, the number of mixed pixels is higher than that under the condition of lower fragmentation index, more winter wheat pixels are missed, and the speed of accuracy decrease is also higher. Winter wheat identification capacity is closely associated with its area proportion within the pixels. Along with the decrease of resolution from 2 to 250 m, the average value of the winter wheat area pixel proportion decreases from 0.94 to 0.45. It can be seen from the patch scale analysis that the size of missed patches also gradually increases from 0.13 to 0.57 hm2. It is also found that long and narrow crop classification patches are likely to be missed along with the decrease of resolution, because they easily generate mixed pixels, which leads to the convergence between spectrum of winter wheat areas and that of background and thus lowers the identification capacity. The gray standard deviation of winter wheat pixel constantly decreases along with the increase of resolution, indicating that the higher the resolution, the stronger the spectrum consistency of winter wheat pixels, which is more conductive to the classification of winter wheat. The above studies show that in the regions with complicated planting conditions in China, considering both image expenses and computation efficiency, improving image resolution is a precondition for improving winter wheat identification accuracy. Meanwhile, in the regions with relatively high cropland fragmentation index, the same identification accuracy can be achieved by using the images with higher resolution.