Accurate humidity forecasts are crucial for environmental and operational applications, yet Numerical Weather Prediction systems frequently exhibit systematic and random errors. To address this problem, this study introduces a modified hybrid post-processing approach that extends a previously developed methodology, enabling a direct comparison of computational efficiency and predictive capacity. The proposed framework integrates a quadratic Kalman Filter with a Radial Basis Function Neural Network trained via the Orthogonal Least Squares algorithm and updated online through Recursive Least Squares. This modified method was evaluated via a time-window process, using forecasts from the Weather Research and Forecasting model and recorded observations from stations in northern Greece. The results show substantial improvements in forecast accuracy, as the Bias was reduced by over 85%, and the MAE and RMSE decreased by approximately 65% and 58%, respectively, compared with the baseline model. Furthermore, the proposed framework also demonstrates enhanced computational efficiency, reducing processing time by more than 95% relative to the initial methodology.
Donas et al. (Sat,) studied this question.