There are several lines of evidence that the atmosphere of Mars has significantly evolved over the history of the planet. Because Mars does not have a strong intrinsic magnetic field, the atmosphere is eroded by interactions with the solar wind. Conditions in the early solar-system significantly enhanced this loss, notably the component due to solar-wind induced sputtering away. Modeling indicates that, integrated over the last 3.5 billion years, 0.8 bars of CO_2 have been sputtered. This is accompanied by the loss of 50 m of water. The loss of CO_2 is a significant loss when compared to the estimates of the thickness of the early atmosphere. A simple model of the behavior of the atmospheric δ^(13)C, based on the expected Martian carbon "cycle" constrains the size the current CO_2 reservoirs by putting the sputtering loss into the context of the evolution of the atmosphere. In order to balance the isotopic effects of the sputtering loss, it is necessary for there to be ~ 100 mbars of CO_2 trapped in the planet. This is quite reasonable given the ability of the regolith to hold adsorbed CO_2. Using a modified form of Optimal Interpolation, it is possible to assimilate Thermal Emission Spectrometer (TES) observations from the Mars Global Surveyor (MGS) spacecraft into the Ames Mars General Circulation Model (MGCM). The method is optimized for the assimilation of the irregular data obtained during the aerobraking phase of the mission. Based on 25 sols of data at L_s ≈ 200, the assimilation process reveals several interesting features of the Martian atmosphere. The assimilation indicates that the lower atmosphere (up to ~ 0.1 mbar) in the northern polar regions is very cold-probably at or close to the CO_2 condensation temperature. Furthermore, the data imply that the midlatitudinal westerly jets extend poleward of the indicated MGCM locations. In addition to correcting the phasing of the Northern baroclinic storm belt, the data indicate that the amplitude of the waves are stronger than expected and possibly with a lower zonal wavenumber. Thus the data and assimilation allows the MGCM to create the 1997 Martian Fall instead of a random Fall.