Abstract:In order to study the extrapolation predictability of cloud images or cloud clusters with different scales, two things were done. First, a cloud detection method suitable for FY-4 satellite infrared cloud images (10.8 μm) was designed, and the cloud clusters with different scales were separated by the regional recognition algorithm. Second, the optical flow method of HS global constraint scheme was used to carry out extrapolation test. The statistical results from 12 examples in 2020 show that in the extrapolation forecasting of brightness temperature, the available optical flow information is close to 6 h. The RMSE of bright temperatures in 0.5, 1 and 6 hours are about 4.4, 7.1 and 16.7 K respectively. The accuracy of extrapolation prediction decreases with the increasing of time length in forecasting. In addition, because of diurnal variations in brightness temperature, the cloud detection results are used in the following extrapolation tests. In the short-time (0-1 h) extrapolation forecast, the error increase is mainly caused by the deviations from cloud location. During the first to sixth hours, the main forecast error is caused by the prediction errors in cloud location and cloud area. The accuracy of extrapolation forecasting decreases as the spatial scale of the cloud cluster decreases. The useable time lengths of extrapolating cloud clusters in scales >2000 km, 200-2000 km, 20-200 km, and <20 km are shorter than 6 h, 1.5 h, 1 h and 15 min, respectively. The extrapolated results in all scales are similar with that in the scale >2000 km. The research results can give a significant guide in extrapolation forecasting of infrared cloud image in operational applications.