In this work, a hybrid solar irradiance now-casting mechanism is proposed. The proposed hybrid predictor fuses the results from both Kalman filter predictor and regressor predictor to benefit from the advantages of both techniques. A time-varying adaptive system function for Kalman filter is designed to deal with ramp-down events for more accurate prediction. Three fusion alternatives based on local root mean square error computation are proposed and compared. The experimental results have validated the effectiveness of the proposed method on a challenging dataset. (C) 2016 Elsevier Ltd. All rights reserved.