顾及邻近关系的NPP/VIIRS序列影像时空插值方法

针对可见红外成像辐射仪(visible infrared imaging radiometer suite，VIIRS)月度夜光遥感影像的数据缺失问题，提出一种利用地物邻近关系相关性的像元时空插值方法，以时、空关系互相作为约束条件，将时序变化一致性较好的像元数据作为空间插值的参考，将空间关系一致性较好的月度数据作为时序插值的参考，通过构建不同的卷积核, 在时序和空间维度分别对初步插值结果进行卷积运算，求得待插值像元的时空插值。以2015年江苏省月度夜光遥感影像修复为例，对不同维度时空插值方法进行对比分析，结果表明, 空间维度插值虽然顾及到像元的空间关联性，仍无法满足数据大范围缺失的插值要求，插值结果整体偏低；时间维度插值考虑到像元的时间趋势性，插值精度较空间维度插值有一定提高，但部分月份插值结果有较大偏差；相对于三次Hermit插值，时空插值方法获得的月度影像灯光亮度总和的最大相对误差、年度影像灯光亮度总和相对误差以及逐像元差值均显著降低。总的来看，所提时空插值方法在插值过程中同时顾及到VIIRS数据的时间趋势平稳性和空间结构稳定性，影像插值精度提高明显，且对待插值月份前后时序数据没有严格要求，更具有广泛性。

Spatiotemporal Interpolation Method of NPP/VIIRS Sequence Images Considering Neighbor Relationships

  Objectives  The problem of missing data in the monthly nocturnal visible infrared imaging radiometer suite (VIIRS) remote sensing images has become one of the limitations for application. Exploring practical data processing methods to obtain high-quality spatiotemporal continuous VIIRS monthly image data has become necessary for night time light remote sensing research.  Methods  A spatiotemporal interpolation method of image elements using the correlation of feature proximity relationship is proposed. The image element data with good consistency of temporal variation is used as the reference for spatial interpolation. The monthly data with good consistency of spatial relationship is used as the reference for temporal interpolation, and the spatiotemporal interpolation of the image element to be interpolated is obtained by constructing different convolution kernels to convolve the preliminary interpolation results in temporal and spatial dimensions, respectively.  Results  Taking the monthly night time light remote sensing image restoration of Jiangsu Province in 2015 as an example, we compare and analyze the spatiotemporal interpolation methods in different dimensions. Compared with the three Hermit interpolations, the maximum relative error of the monthly image light brightness sum, the relative error of the annual image light brightness sum, and the image-by-image difference obtained by the spatiotemporal interpolation method are significantly reduced.  Conclusions  The proposed spatiotemporal interpolation method takes both the temporal trend smoothness and spatial structure stability of VIIRS data into account in the interpolation process, and the image interpolation accuracy is significantly improved. The proposed method is more extensive because there is no strict requirement for the time series data before and after the month to be interpolated.