Tropospheric 18O18O is an emerging proxy for past tropospheric ozone and free-tropospheric temperatures. The basis of these applications is the idea that isotope-exchange reactions in the atmosphere drive 18O18O abundances toward isotopic equilibrium. However, previous work used an offline box-model framework to explain the 18O18O budget, approximating the interplay of atmospheric chemistry and transport. This approach, while convenient, has poorly characterized uncertainties. To investigate these uncertainties, and to broaden the applicability of the 18O18O proxy, we developed a scheme to simulate atmospheric 18O18O abundances (quantified as ∆36 values) online within the GEOS-Chem chemical transport model. These results are compared to both new and previously published atmospheric observations from the surface to 33 km. Simulations using a simplified O2 isotopic equilibration scheme within GEOS-Chem show quantitative agreement with measurements only in the middle stratosphere; modeled ∆36 values are too high elsewhere. Investigations using a comprehensive model of the O-O2-O3 isotopic photochemical system and proof-of-principle experiments suggest that the simple equilibration scheme omits an important pressure dependence to ∆36 values: the anomalously efficient titration of 18O18O to form ozone. Incorporating these effects into the online ∆36 calculation scheme in GEOS-Chem yields quantitative agreement for all available observations. While this previously unidentified bias affects the atmospheric budget of 18O18O in O2, the modeled change in the mean tropospheric ∆36 value since 1850 C.E. is only slightly altered; it is still quantitatively consistent with the ice-core ∆36 record, implying that the tropospheric ozone burden increased less than ~40% over the twentieth century.

Ice cores and other paleotemperature proxies, together with general circulation models, have provided information on past surface temperatures and the atmosphere’s composition in different climates. Little is known, however, about past temperatures at high altitudes, which play a crucial role in Earth’s radiative energy budget. Paleoclimate records at high-altitude sites are sparse, and the few that are available show poor agreement with climate model predictions. These disagreements could be due to insufficient spatial coverage, spatiotemporal biases, or model physics; new records that can mitigate or avoid these uncertainties are needed. Here, we constrain the change in upper-tropospheric temperature at the global scale during the Last Glacial Maximum (LGM) using the clumped-isotope composition of molecular oxygen trapped in polar ice cores. Aided by global three-dimensional chemical transport modeling, we exploit the intrinsic temperature sensitivity of the clumped-isotope composition of atmospheric oxygen to infer that the upper troposphere (5 – 15 km altitude, effective mean 10 – 11 km) was 4 – 10°C cooler during the LGM than during the late preindustrial Holocene. These results support a minor or negligible steepening of atmospheric lapse rates during the LGM, which is consistent with a range of climate model simulations. Proxy-model disagreements with other high-altitude records may stem from inaccuracies in regional hydroclimate simulation, possibly related to land-atmosphere feedbacks.