大量中、高分辨率遥感图像是地球资源探测、地物变化监测、农林水利测绘勘探等研究的重要信息基础。而在遥感图像的应用中，由于云等大气不透明成分的遮挡，不良天气的影响，我们往往难以获取清晰、完整的高质量光学遥感影像，无法为各项应用提供详实的信息。同时又受限于单一星源传感器较低的时间分辨率，以陆地资源卫星Landsat-8的运行性陆地摄影仪（Operational Land Imager：OLI）图像为例，其时间分辨率为16天，这意味着我们在一个月内只能获取同一区域的2景图像，对于低纬度多云区域，这两景图像可能都被云层大量覆盖，难以满足遥感应用。这就需要我们利用多时相遥感图像序列、甚至来自于不同卫星的图像进行合成，以获得无云、清晰、时间和空间分辨率更高的遥感图像，从而满足大面积区域遥感应用的需求。 有效地使用计算机进行多时相遥感影像的合成，最基本的在于找到较好的算法，解决合成过程中所遇的有效数据源的获取、云像元的识别、干净像元选取方法的问题。近年来国际上对遥感卫星图像的合成进行了大量的研究，陆续提出了最大归一化植被指数合成法、时间序列的谐波分析法、多维均值法、基于像元的图像合成等各种算法，每种方法都有各自的优势，但由于有效数据源获取的困难、图像时间和空间分辨率有限、以及图像中云像元标识和识别不准确所带来的数据缺失带等问题仍然存在。所以，如何有效利用现有卫星数据进行无云数据合成获取高质量遥感影像是研究的重点和难点。 本文中我们首次将最新一颗陆地资源卫星Landsat-8数据与欧空局2015年和2017年陆续发射的哨兵2号A、B双星Sentinel-2A/B数据进行异源、不同时空分辨率遥感数据的合成，并获取了辐射测量一致性较高的无云合成影像。同时，本研究中选用中国北部黑龙江省孟家岗人工林区和南部广西南宁地区为试验区，结合了不同纬度地区生长季情况，充分利用了Landsat-8和Sentinel-2的数据特点，利用最佳像元合成算法分别对单颗卫星图像进行了季相、年度、多年合成，均获取了能够代表该时段地物特征的无云图像，且与源数据保持了较好的物候一致性，合成影像与参考影像地面反射值相关性高达0.94。

The acquisition of a large number of medium and high-resolution, multi-temporal remote sensing data provides an important information for research on resource detection, change monitoring, mapping and exploration, and agriculture, forestry, and water conservancy. In the application of remote sensing data, due to the contamination of atmospheric opaque components such as clouds and the influence of bad weather, it is often quite difficult to obtain clear and complete high-quality optical remote sensing images, which cannot provide detailed information for various applications. At the same time, it is limited by the low time resolution of a single sensor, taking the operational land imager (OLI) of Landsat-8 as an example. Its time resolution is 16 days, which means that we can only obtain two images of the same area within one month. For low-latitude and cloudy areas, these two images may be covered by clouds in large quantities and it is difficult to satisfy the needs of remote sensing applications. This requires us to synthesize multiple multi-temporal remote sensing images and even different satellite images to obtain groundless images that are cloudless, clear, and with higher time and spatial resolution, so as to explore the regional macro-rules. Therefore, the composite technology of multi-temporal remote sensing images has become one of the fundamental technologies for regional remote sensing monitoring and research. Effectively using computers for the composite of multi-temporal remote sensing images, the most basic is to find a better algorithm to solve the problem of the acquisition of effective data sources encountered in the compositing process, the identification of cloud pixels, and the selection method of clean pixels. In recent years, a large number of studies have been conducted on the composite of remote sensing satellite images in the world, and various algorithms such as maximum normalized vegetation index composite method, time series harmonic analysis method, multi-dimensional mean value method, and pixel-based image composite have been proposed one after another. Each method has its own advantages, but problems such as difficulties in obtaining valid data sources, limited temporal and spatial resolution of images, and missing data bands caused by inaccurate identification and recognition of cloud pixel in images still exist. Therefore, how to effectively use existing satellite data for cloudless data composite to obtain high-quality remote sensing images is the focus and difficulty of research. In this paper, we firstly combine images of Landsat-8 and sentinel-2 A/B which have different spatial resolution and come from different satellite together to generate composite images. The composite and acquisition of cloudless compositing images with high consistency in radiometric correspondence. At the same time, in this study, Mengjiagang artificial forest area in northern China's Heilongjiang Province and Nanning area in southern Guangxi were selected as the experimental area. Combined with the growing season in different latitudes, the data characteristics of Landsat-8 and Sentinel-2 were fully utilized, and the best image was utilized. We also produced quarter-, year-, and multi-year composite of individual satellite images, all acquired cloudless images that can represent the features of the time period, and maintained good phenological consistency with the source data. The correlation with the reference image surface reflectance is as high as 0.94.