A feasibility study of computational fluid dynamics (CFD)-based AI-surrogates based on data-driven neural networks for learning flow parameters is conducted to reflect their use in the field of site and wind resource assessment. The AI-surrogates are developed for varying forest canopy heights for complex terrain(s). The model inputs are the forest canopy height and corresponding CFD simulated velocity and pressure values at different locations. The model output is a set of flow parameters such as velocity and pressure. The data-driven network is trained on CFD simulation results for Reynolds-Averaged Navier-Stokes (RANS). The network architecture is a feed-forward convolutional neural network with loss functions that incorporate residuals from the computed results. The L-BFGS optimization algorithm is used to minimize this loss and tune the network hyperparameters. A total of 180 CFD cases are simulated to study the impact of wind directions (ranging from 0-360 degrees) and forest height variations (ranging from 5-25 m) on wind speed at a complex site in Baden-Württemberg, Germany. A set of 160 of these simulations is used to train the network, and the network estimates are validated against CFD simulations for the 20 cases excluded from the training set [1]. The network is systematically extended by incorporating CFD simulations together with geometrical features of several sites. The results are presented together with a note on the strengths and shortcomings of regression-based neural network training. REFERENCES [1] Z Lakdawala, M.W. Nadeem, H. Kassem, M. Doerenkaemper, Investigating the usability of physics informed machine learning approaches for wind farm planning, 9th ECCOMAS Congress (2024) [forthcoming].