Terrain-aided navigation is often used for unmanned aerial vehicles. This method consists in estimating the current dynamics and position of a vehicle by matching terrain profiles obtained by exteroceptive sensors with onboard maps. Whereas the resolution of the map plays a crucial role in minimising positioning errors, accurate reference maps are difficult to upload when transmission restrictions are experienced. In this context of sparse communications, we propose to model the maps by Gaussian Processes completely characterised by a mean function that can be interpreted as a low resolution approximation of the map and a covariance kernel to account for the spatial correlations. The objective of this paper is then to perform simultaneous localisation and mapping by leveraging radio-altimetric data. The inference is carried out by a non-linear Bayesian filter that takes advantage of the conditional linear Gaussianity of the state space model: the Rao-Blackwellised particle filter.