STAIR: A generic and fully-automated method to fuse multiple sources of optical satellite data to generate a high-resolution, daily and cloud-/gap-free surface reflectance product
STAIR: A generic and fully-automated method to fuse multiple sources of optical satellite data to generate a high-resolution, daily and cloud-/gap-free surface reflectance product
复制标题
DOI：
10.1016/j.rse.2018.04.042
复制
发表时间：
2018-09
影响因子：
13.5
通讯作者：
Yunan Luo;K. Guan;Jian Peng
中科院分区：
文献类型：
--
作者：
Yunan Luo;K. Guan;Jian Peng
高分辨率的地表反射率数据是科学研究和社会应用的迫切需求。标准的卫星任务无法提供这两种高分辨率的数据。利用各种卫星来源的互补优势（例如，中分辨率成像分光仪/VIIRS/GOES-R的近每日重访频率和大地卫星/哨兵-2的高空间分辨率）的融合方法提供了一种可行的手段，可以同时实现融合数据的高分辨率。在本文中，我们提出了一种新的，通用的和全自动的方法，STAIR，融合多光谱卫星数据，以产生一个高频率，高分辨率和云/间隙的数据。基于多个卫星数据源的时间序列，STAIR首先使用自适应平均校正过程来估算卫星图像中的缺失值像素（由于云层覆盖或传感器机械问题），该过程通过自动分割来考虑不同的土地覆盖和缺失值像素的邻域信息。为了融合卫星图像，它采用局部插值模型来捕获由高空间分辨率数据提供的最具信息性的空间信息（例如，Landsat），然后执行调整步骤以合并由高频数据提供的时间模式（例如，中分辨率成像分光仪）。由此产生的融合产品包含每日、高空间分辨率和无云/无间隙的融合图像。我们测试了我们的算法，融合表面反射率数据的MODIS和Landsat在伊利诺伊州的尚潘县，并生成每日时间序列的所有生长季节（4月1日至11月1日）从2000年到2015年在30米的分辨率。大量的实验表明，STAIR不仅捕捉正确的纹理模式，但也预测准确的反射率值在生成的图像，与经典的STARFM算法的性能显着提高。这种方法计算效率高，可以扩大到大陆尺度。它还具有足够的通用性，可以很容易地将各种光学卫星数据包括在内，以便进行融合。我们设想这种新的算法可以提供有效的手段，利用历史的光学卫星数据，建立长期的日常，30米的表面反射率记录（例如，从2000年到现在）在大陆尺度的各种应用，以及产生业务的近实时的日常和高分辨率的数据，为未来的地球观测应用。
Surface reflectance data with high resolutions in both space and time have been desired and demanded by scientific research and societal applications. Standard satellite missions could not provide such data at both high resolutions. Fusion approaches that leverage the complementary strengths in various satellite sources (e.g. MODIS/VIIRS/GOES-R's sub-daily revisiting frequency and Landsat/Sentinel-2's high spatial resolution) provide a viable means to simultaneously achieve both high resolutions in the fusion data. In this paper, we presented a novel, generic and fully-automated method,STAIR, for fusing multi-spectral satellite data to generate a high-frequency, high-resolution and cloud-/gap-free data. Building on the time series of multiple sources of satellite data, STAIR first imputes the missing-value pixels (due to cloud cover or sensor mechanical issues) in satellite images using an adaptive-average correction process, which takes into account different land covers and neighborhood information of miss-value pixels through an automatic segmentation. To fuse satellite images, it employs a local interpolation model to capture the most informative spatial information provided by the high spatial resolution data (e.g., Landsat) and then performs an adjustment step to incorporate the temporal patterns provided by the high-frequency data (e.g., MODIS). The resulting fused products contain daily, high spatial resolution and cloud-/gap-free fused images. We tested our algorithm to fuse surface reflectance data of MODIS and Landsat in Champaign County at Illinois and generated daily time series for all the growing seasons (Apr 1 to Nov 1) from 2000 to 2015 at 30 m resolution. Extensive experiments demonstrated that STAIR not only captures correct texture patterns but also predicts accurate reflectance values in the generated images, with a significant performance improvement over the classic STARFM algorithm. This method is computationally efficient and ready to be scaled up to continental scales. It is also sufficiently generic to easily include various optical satellite data for fusion. We envision this novel algorithm can provide effective means to leverage historical optical satellite data to build long-term daily, 30 m surface reflectance record (e.g. from 2000 to present) at continental scales for various applications, as well as produce operational near-realtime daily and high-resolution data for future earth observation applications.