结构函数法气溶胶光学厚度反演精度受像元间隔设置影响很大，且并非所有像元都能获得较好的反演结果，因而研究像元间隔的设置能够提高反演精度，研究反演误差小的像元能够提高算法效率。为了获得最佳的像元间隔设置，本文以胶州湾地区为例，利用250 m和500 m两种分辨率数据计算了不同像元间隔时的结构函数值，分别利用单一像元间隔法、均值法、坡高法以及线性区域均值法获得待反演像元最终的结构函数值反演550 nm处的气溶胶光学厚度，并依据CE318观测数据进行精度验证，通过分析点对点反演结果和光学厚度的空间分布，确定反演误差小、受分辨率影响小的像元间隔设置。实验发现线性区域均值法在一定程度上提高了反演精度和稳定性。此外，通过对反演结果可接受像元的地表反射率结构函数值的统计和分析，发现500 m分辨率时可接受像元比例优于250 m，当地表反射率结构函数值大于0.02时反演结果较好，而这些像元往往分布在山麓、山涧、海岸线、河流、城乡结合部等地理要素的突然改变的地区。

Retrieval of Aerosol Optical Depth(AOD)over land through satellite remote sensing is difficult. Traditional methods to retrieve aerosol over land involve the use of the Dense Dark Vegetation(DDV) algorithm. However, this algorithm requires the support of an infrared band and dark pixels. Compared with DDV, the structure function method can be utilized to retrieve AOD over bright areas with high surface reflectance. However, the precision of retrieving AOD through the structure function method can be significantly influenced by pixel distance. Not all pixels can be retrieved with high precision; this condition results in a large amount of unnecessary computations. As a result, determining the pixel distance settings and the pixels with a small retrieval error is important in improving the retrieval precision and efficiency of the structure function method.
To acquire the best pixel distance settings, we calculated the structure function values with different pixel distances and obtained the final structure function value of the target pixel through single-pixel distance, average, slopeheight, and average linear area methods to retrieve AOD at 550 nm over Kiaochow Bay. The first band of nine-scene MODIS L1B data with resolutions of 250 m and 500 m was used. Surface reflectance was the minimum value of the first band of MOD09Q1 in 2012. By validating the AOD data measured by CE318 and by analyzing the retrieval result and spatial distribution of AOD, we obtained the pixel distance settings with high precision and minimal effects of resolution.
The experiment indicated that the average linear area method has a small retrieval error whether the resolution is 250 m or 500 m. According to statistics and by analyzing the structure function value of the surface reflectance of pixels with an acceptable error (the absolute value of the absolute error is 0.1 for measured AOD less than 0.6 and 0.2 for measured AOD greater than 0.6), we found that the ratio of pixels with are solution of 500 m is obviously larger than that with 250 m resolution. Pixels with a surface reflectance structure function value larger than 0.02 have a good result and areaways distributed over areas with a sudden change in geographical elements, such as piedmonts, mountain streams, coasts, rivers, and rural-urban continua.
In this study, we utilized different methods and historical surface reflectance data with two resolutions to calculate the final structure function value. The retrieval precision and distribution of AOD were analyzed to acquire the best pixel distance settings, and its features were identified through the statistics of the structure function values with high retrieval precision. The average linear area method is stable, and its result has an insignificant relationship with data resolution. The retrieved AOD with high precision is concentrated on pixels with a surface reflectance structure function value larger than 0.02. The efficiency of this method increases when the measured AOD is greater than 0.5.