Machine Learning (ML) is becoming an increasingly popular and important tool for predicting ocean wave conditions. Here it is applied to downscale offshore conditions to a nearshore location utilising more detailed representations of the offshore wave field using 1D wave spectra. Our aim is to identify some of the sensitivities in input data when using machine learning to conduct downscaling (a common application) and present results from different approaches. The results demonstrate that downscaling wave conditions using ML can be enhanced using 1D wave spectra to improve performance. Here, we obtained a 27 % reduction in root mean squared error in significant wave height when compared to integrated parameter only machine learning approach with performance improved when using 1D wave spectra. Though we identified that the Long-Term Short-Term Memory approach applied here improved performance overall, it also appears there is not a one-size fits-all approach for all wave parameters. Careful feature selection (which features to include or exclude when training a model), feature engineering (such as feature encoding and sequence selection) and model configuration continue to be key factors in achieving accurate wave conditions.