We investigate the compression of quantum information with respect to a given set \mathcal{M} of high-dimensional measurements. This leads to a notion of simulability, where we demand that the statistics obtained from \mathcal{M} and an arbitrary quantum state \rho are recovered exactly by first compressing \rho into a lower dimensional space, followed by some quantum measurements. A full quantum compression is possible, i.e., leaving only classical information, if and only if the set \mathcal{M} is jointly measurable. Our notion of simulability can thus be seen as a quantification of measurement incompatibility in terms of dimension. After defining these concepts, we provide an illustrative examples involving mutually unbiased basis, and develop a method based on semi-definite programming for constructing simulation models. In turn we analytically construct optimal simulation models for all projective measurements subjected to white noise or losses. Finally, we discuss how our approach connects with other concepts introduced in the context of quantum channels and quantum correlations.
