This thesis investigates the potential of data-driven modelling to enhance wind farm monitoring, control, and operations and maintenance (O&M) strategies by leveraging the extensive SCADA data generated by offshore wind farms. Focusing on the Lillgrund offshore wind farm, the study develops regression-based surrogate models using XGBoost, artificial neural networks, and Gaussian process regression to predict two critical targets: the blade root flapwise and tower bottom fore-aft damage equivalent loads. All models are found to effectively capture the underlying patterns from the input features. Among them, XGBoost consistently outperforms the others in terms of prediction accuracy, computational efficiency, and robustness. Its superior performance is further validated across multiple preprocessing settings and operational scenarios, including its capacity to generalise across turbines and exploit spatial information. Finally, the applicability of the models is demonstrated through a simplified use case that estimates fatigue damage over a specific period. The findings underline the value of integrating machine learning-based surrogate models into operational workflows to reduce O&M costs and support decision-making in wind farm management.