Owing to its 1200 yr lifetime, atmospheric oxygen (O₂) is a global tracer of biological and hydrological processes. The dominant source of O₂ is located in the the low latitudes, where most of the O₂ production/uptake occurs. Atmospheric O₂ can therefore provide valuable information on the tropics, a region of the world which still lacks of climatic reconstructions and whose role is widely debated in the context of millennial-scale climate variations. Atmospheric O₂ is enriched in heavy isotopologues (δ¹⁷O , δ¹⁸O ) relative to O₂ in ocean water, the ultimate source of O₂ for photosynthesis. The processes causing enrichment of ¹⁷O/¹⁶O and ¹⁸O/¹⁶O isotope ratios involve the biological cycle, the water cycle, global ice volume/sea level variations, climatic conditions and stratospheric photochemistry. It is thus essential to estimate the relative importance of these processes to unleash the potential of O₂ isotopologues as global tracers of past changes in the climate system. The evolution of the past atmosphere can be retrieved from the air bubbles occluded n polar ice cores back to 800 kyr. However, elemental and isotopic fractionation processes alter O₂ isotope ratios during the transport and entrapment of air in the porous layer (firn) on top of the ice sheets, during storage and during the experimental analysis. An understanding of these non-climatic mechanisms is aprerequisite for a correct interpretation of gases preserved in ice cores.
The work presented in this thesis focuses on the past evolution of stable isotopes of atmospheric oxygen from the technical aspects of the measurements of O₂ isotope ratios to the interpretation of their past variations. First, we present the O₂ cycle and describe the results of process-based modeling studies aiming  at reproducing the observed enrichment in atmospheric δ¹⁸O and ¹⁷Δ (¹⁷Δ_atm= ln(δ¹⁷O_atm)−0.516 · ln(δ¹⁸O_atm). We review the current understanding of pastorbital and millennial time-scale variations of atmospheric O₂ isotopes. We also give a description of air transport and associated processes in the firn, which alte rthe climatic signal preserved in ice core bubbles.
Second, a very high analytical precision and accuracy is required to measure the past variations of δ¹⁸O_atm and especially ¹⁷Δ_atm preserved in ice core bubbles. One must primarily have the ability to measure variations as small as 10 permeg (0.01 ‰), corresponding to the millennial-scale changes observed in ¹⁷Δ_atm . O₂ needs to be separated from other atmospheric constituents to achieve such a level of precision. This motivated us to develop a new method of O₂ separation, based on membrane technology. We verify its 100 % selectivity to O₂ , and estimate its O₂ permeability. This method is currently not applicable to ¹⁷Δ_atm measurements due to sealing issues and variable isotope fractionation during O₂ permeation across the membrane.  
Third, a semi-automated, offline experimental setup for δ¹⁸O_atm and ¹⁷Δ_atm measurements was build up from scratch as an alternative, based on the conventional method relying on gas chromatograph (GC) separation of O₂ and nitrogen (N₂ ). It includes air extraction from ice, standard introduction and cryo-collection at 12 K (with a closed Helium cooler) of an O₂ /Argon (Ar) mixture, after separation fromwater (H₂O ), carbon dioxide (CO₂ ) and N₂ . The GC unit can be bypassed for δ¹⁸O_atm measurements in a dried and CO₂ -free air mixture. We give an overview of the units and controls of the experimental setup, and detail the developed procedure to extract, purify and collect atmospheric O₂ from ice core samples.
Fourth, the external precision of the setup, or the reproducibility of ice core δ¹⁸O_atm measurements is estimated with 21 Late Holocene Neem (Greenland) icecore samples from the same depth. A melt-extraction method is applied on these large samples (⋍ 30 g) and δ¹⁸O_atm , δO₂/N₂ and δ¹⁵N are measured in an O₂ /N₂/Ar mixture by isotope ratio mass spectrometry in dual inlet mode. We describethe automation of a measurement sequence of up to 10 ice core samples. The scatter observed in the raw δ¹⁸O_atm and O₂/N₂ underlines the occurrence of gas loss fractionation processes in ice core samples. We detail the method of data-processing,its associated uncertainty and the strategy employed to correct for non climatic effects. Based on zero-enrichment tests, the internal precision of DI measurementsof δ¹⁸O and δ¹⁵N is 0.008‰ (1 σ) and 0.005‰ (1 σ). A similar precision is reached with individual ice core sample measurements. Based on the 21 Neem ice core samples, the ability of the analytical system to reproduce δ1⁸O_atm and δO₂/N₂ is estimated as 0.028‰ (1 σ) and 0.021‰ (1 σ), respectively.
Fifth, thanks to improving isotope measurement techniques, millennial scale variations of δ¹⁸O_atm and ¹⁷Δ_atm preserved in polar ice cores have been revealed. In particular, a systematic δ¹⁸O_atm increase is recorded during Heinrich stadials (Greenland stadials during which a Heinrich event occurs). Because of its global character, δ¹⁸O_atm provides added value compared to the different local records ofhydrological cycle variations in different continental and marine archives. However, until now, no quantitative, robust interpretation of past variations in 18Oatmhas been established, which limits the use of δ¹⁸O_atm as a quantitative indicator for past biospheric production or variations of the hydrological cycle. Here, we quantifythe response of δ¹⁸O_atm to such millennial events using a freshwater hosing simulation performed under glacial boundary conditions. Our O₂ isotope mass balance model takes into account the latest estimates of isotope fractionation factors for respiratory and photosynthetic processes, and makes use of atmospheric water isotope and vegetation changes obtained with the general circulation model IPSL-CM₄. The atmospheric component of IPSL-CM₄ is fitted with a water isotope module (LMDZ₄), and its land component, the dynamic global vegetation model ORCHIDEE, is run offline. Our modeling approach reproduces the main observed features of a Heinrich stadial in terms of climatic conditions, vegetation distribution and δ¹⁸O of precipitation. We use it to decipher the relative importance of the different processes behind the observed changes in δ¹⁸O_atm . Our results highlight the dominant role of hydrology on δ¹⁸O_atm and confirm that δ¹⁸O_atm can be seen as a global integrator of hydrological changes over vegetated areas. This work has been published in Climate of the Past in 2015 under the title Quantifyingmolecular oxygen isotope variations during a Heinrich stadial.