Optical Flow Based Solar Irradiance Forecasting in Satellite Images

Abstract

Prediction of cloud movement is essential for solar irradiance forecasting. However, most of the cloud tracking and predicting methods at present refer to ground-based images. This paper provides a new viewpoint from satellite images to analyze cloud movement. We choose Shi-Tomasi's method to pick out feature points, filter the image with Gaussian kernel and track the cloud by Lucas-Kanade optical flow methods. We also evaluate detailed implementations of different filters and sampling methods with our own metrics. The result indicates that the satellite image is high-resolution enough for the task of cloud tracking and irradiance forecasting and optical flow is a suitable method for this task.