Constructing high-resolution wind fields is vital for wind energy evaluation. This study develops an integrated framework that combines turbulence parameterization sensitivity analysis with a deep-learning-based super-resolution model. Using Weather Research and Forecasting (WRF) simulations and Doppler wind lidar observations, turbulence schemes are evaluated under four representative wind conditions to ensure physically consistent and reliable inputs. In addition, a deep learning model that integrates the Beluga Whale Optimization (BWO) algorithm with an enhanced deep super-resolution (EDSR) network incorporating channel attention (CA) is further proposed, enabling efficient reconstruction of wind fields at the 100-m scale. Results show that the Mellor–Yamada Nakanishi and Niino Level 3 (MYNN3. 0) adding scale-dependent functions (MYNN3. 0₂) scheme performs best under low-level jets, reducing average normalized root mean square error (NRMSE) by 30%. For the other conditions, the MYNN3. 0 scheme yields higher accuracy, with average NRMSE reductions of 35%, 35%, and 40. 6%, respectively. Moreover, the BWO-EDSR-CA super-resolution model outperforms cubic spline interpolation (CSI), EDSR, and EDSR-CA in reconstructing wind speed and turbulent kinetic energy. Compared with CSI, the root mean square error is reduced by 13. 8% and 63. 2%, respectively. These results confirm both the methodological innovation and the practical utility of the approach, offering a reliable pathway for future high-resolution wind resource assessments.
Wang et al. (Thu,) studied this question.