The two-dimensional structure of thermospheric neutral composition, specifically, the atomic oxygen to molecular nitrogen column density ratio, 2>, is studied during the 17--24 April 2002 geomagnetic storms to understand the cause of ionospheric storms in regions equatorward of the auroral oval on an instantaneous large scale. The 2> ratio is derived from the dayglow emission ratio of O I 1356 ¿ to N2 Lyman-Birge-Hopfield (1600--1800 ¿) acquired from the Polar ultraviolet imager (UVI) and the total electron content (TEC), which is used to infer ionospheric storms, is derived from the phase delays of dual-band global positioning satellite (GPS) accumulated around the globe. It is found that the regions of decreasing 2> generally coincided with the regions of depleted TEC during and after the development of the storms. This is consistent with previous theoretical and experimental analysis in which composition changes play a major role in the negative ionospheric storm effects. At lower latitudes, long-lived positive storm effects predicted by empirical and general circulation models were not observed. In fact, there was no noticeable change in 2>. The TEC data also showed no noticeable change, except a few short-lived, localized positive TEC perturbations. For this particular event, the equatorward expansion of the decreased 2> at the onset of the first storms was estimated to be more than 600 m/s, much faster than the typical thermospheric wind, and reached ~30¿N in 2 hours much equatorward (>25¿ MLAT) beyond the auroral electrojets. This result suggests that midlatitude (negative) ionospheric storms are caused by direct equatorward penetration of a reduced thermospheric 2>, at least during the first few hours of storms.