The search for high-temperature superconductors has advanced significantly in the last ten years, with many hydrogen-rich compounds exhibiting superconductivity above 200 K. However, their stability at extreme pressures limits practical applications. Theoretical prediction of new compounds is hindered by expensive calculations of critical temperatures (Tc) combined with a vast chemical space. To address this, we propose some inexpensive descriptors that can provide a simple characterization of superconducting hydrides, and can be used to do estimations of Tc using machine learning. We divide electron-phonon coupling into electronic and lattice contributions. Using the electron localization function and hydrogen density of states, we can classify hydrides according to their type of bonding, and do first estimations of Tc. It is seen that covalent systems and clathrate-like hydrides are beneficial for superconductivity, as well as a high delocalization of electrons in the hydrogen sublattice. However, systems like Im- 3m-H3S and Im-3m-H3Se, with similar bonding but vastly different Tc, remain challenging to distinguish. To improve the description, we introduce a real-space function based on the derivatives of the Kohn-Sham potential upon lattice fluctuations, which correctly distinguishes the different magnitude of electron-phonon coupling in those compounds. This approach also provides insights into electron-phonon coupling mechanisms, where our observations suggest that Tc is enhanced when the regions of high electron localization coincide with those of large fluctuations of the potential. These real-space descriptors offer a powerful tool for advancing the search for hydride superconductors stable at ambient conditions.
