Data-driven surrogate models for computational fluid dynamics are often based on dimension reduction using convolutional autoencoders (CAE). While such models have shown great potential in various tasks, they usually rely on the same regular grid that is used for all inputs. On the contrary, coordinate-based models, i.e., neural fields, only process single coordinates at a time and are therefore independent of the discretization. This is a beneficial feature when data examples with different discretizations should be considered, as this does not require to map all examples to a meta mesh that is typically used for a CAE. Moreover, this enables super-resolution reconstructions as the output can be computed at arbitrary points. As part of the HYFLIP project, this work deals with the development of data-driven surrogate models for the prediction of aircraft ditching loads and deformations. In previous works, aircraft ditching loads were predicted using CAE-LSTM combinations [1] and the interpretability of the CAE-based model was improved by disentangling the latent space [2]. In this work, a neural field approach, where the latent variables are obtained with an auto-decoding strategy, is investigated. The ultimate goal of this approach is to consider different aircraft geometries with different discretizations within a single model. [1] H. Schwarz, M. Überrück, J.-P. M. Zemke, and T. Rung. Machine learning based prediction of ditching loads, AIAA Journal, 63(5):1835–1854, 2025. [2] H. Schwarz, P. P. Lin, J.-P. M. Zemke and T. Rung. Disentangled latent spaces for reduced order models using deterministic autoencoders, Computers & Fluids, 302:106837, 2025.