基于卫星云图的大区域云层预测方法

云层覆盖是影响对地观测卫星成像的一个重要问题，如果遥感图像中云层比例太高，或者特定目标不可见，则遥感图像就会失效。对地观测卫星能够根据云层预测信息，在多个观测目标之间进行选择。面向对地观测卫星任务规划的应用，设计了大区域范围的短期云层预测方法，首先通过光流法获取云运动矢量，然后依据云运动矢量外推获得预测的云层图像，同时引入拉普拉斯算子刻画云层运动过程中的扩散现象，利用风云二号卫星的真实云图序列数据，通过神经网络的反向传播算法优化扩散因子，以提升云层预测的效果。通过对结果进行分析，引入的拉普拉斯算子方法能够提高云层预测的精度，80%分位数的云层覆盖率误差约为11.7%，该精度的云层预测可以用于指导对地观测卫星任务规划。

Research on large area cloud forecasting method based on satellite cloud images

Cloud cover is an important issue for Earth observation satellite imaging.When the proportion of clouds in the remote sensing image is too high，or the specific target is not visible，the remote sensing image will be invalid.Earth observation satellites can choose among multiple observation targets based on cloud forecast information.For the application of Earth observation satellite mission planning，a large-area short-term cloud forecasting method was proposed.First，the cloud motion vector was obtained by the Farneback optical flow method.Subsequently，the predicted cloud image can be obtained by extrapolating the cloud motion vector.Meanwhile，the Laplace operator was introduced to depict the characterization of the diffusion phenomenon in the process of cloud movement.Utilizing the real cloud image sequence data of the FY-2 satellite，the diffusion factor can be optimized through the backpropagation algorithm of the neural network to improve the effect of cloud forecasting.Through the analysis of forecasting results，the introduced Laplace operator method can improve the accuracy of cloud forecasting.The error of the 80% quantile of cloud coverage is about 11.7%.The cloud cover forecasting with this level of accuracy can be used to guide the mission planning of Earth observation satellites.