We compare global magnetohydrodynamic (MHD) simulation results with an empirical model and observations to understand the magnetic field configuration and plasma distribution in the inner magnetosphere, especially during geomagnetic storms. The physics-based Lyon-Fedder-Mobarry (LFM) code simulates Earth's magnetospheric topology and dynamics by solving the equations of ideal MHD. Quantitative comparisons of simulated events with observations reveal strengths and possible limitations and suggest ways to improve the LFM code. Here we present a case study that compares the LFM code to both a semiempirical magnetic field model and to geosynchronous measurements from GOES satellites. During a magnetic cloud event, the simulation and model predictions compare well qualitatively with observations, except during storm main phase. Quantitative statistical studies of the MHD simulation shows that MHD field lines are consistently under-stretched, especially during storm time (Dst < -20 nT) on the nightside, a likely consequence of an insufficient representation of the inner magnetosphere current systems in ideal MHD. We discuss two approaches for improving the LFM result: increasing the simulation spatial resolution and coupling LFM with a ring current model based on drift physics (i.e., the Rice Convection Model (RCM)). We show that a higher spatial resolution LFM code better predicts geosynchronous magnetic fields (not only the average Bz component but also higher-frequency fluctuations driven by the solar wind). An early version of the LFM/RCM coupled code, which runs so far only for idealized events, yields a much-improved ring current, quantifiable by decreased field strengths at all local times compared to the LFM-only code.