Abstract:Cloud is one of the most important and active factors in weather and climate, and plays an important role in modulating the radiation-energy balance and water cycle of atmospheric system. The effective forecast of total cloud cover can lay a basis for better grasp of weather phenomena and prediction of new energy output such as photovoltaic power generation. The model CMA-BJ (Beijing Rapid Update Cycle System) can provide hourly high-resolution total cloud cover prediction products. In this paper, the prediction performance of CMA-BJ is systematically examined and evaluated by the time scale separation method, and the error sources are analyzed, so as to provide a reference for product interpretation and model improvement. The results show that the spatial distribution characteristics and diurnal variation intensity of total cloud cover can be well predicted by CMA-BJ. The pattern correlation coefficients between the CMA-BJ forecasted and observed total cloud cover with 1-24 h lead time are all greater than 0.6 in each month. However, the total cloud cover and diurnal variation intensity are significantly underestimated in winter (January), with the deviation of CMA-BJ reaching -0.133. As the forecasting time increases, the prediction ability of CMA-BJ decreases, with the averaged TCC skills being 0.470, 0.409, 0.355 and 0.315 for the 1-4 d forecast, which means the skillful prediction can be maintained up to 48-72 hours. The diagnostic analysis shows that the low relative humidity in the model may largely contribute to the negative deviation in total cloud cover prediction. Besides, the bias of vertical velocity prediction is also an important reason for the cloud cover prediction error.