Abstract
This study develops and validates a shape-aware hybrid adaptive kernel density estimation (SHA-KDE) method for probabilistic modeling of wind speed and solar irradiance. The method combines curvature-based and k-nearest-neighbor-based bandwidth components through a location-dependent weight, thereby preserving local distributional features while maintaining stable estimation in sparse and boundary regions. Meteorological data from NASA POWER for 2022–2024 were examined across 16 cities in Kazakhstan. Compared with the Weibull, Beta, standard KDE, and AKDE models, SHA-KDE generally achieved the best or near-best city-level goodness-of-fit performance for wind speed. For solar irradiance, it achieved the lowest CvM and A–D statistics and the highest log-likelihood in all surveyed cities. When the fitted densities were propagated through standardized wind-turbine and photovoltaic models, differences among the probability models produced city-level absolute AEP deviations of up to 4.83%. These results demonstrate the practical importance of accurate probability density estimation in renewable-energy applications.