UC Berkeley

This dissertation explores the utility of high-resolution satellite carbon dioxide (CO2) and water vapor measurements for advancing climate treaty verification, for improving numerical weather prediction (NWP), and for understanding natural carbon cycling in the terrestrial biosphere. We present a series of Observing System Simulation Experiments (OSSEs) using a carbon-weather data assimilation (DA) system, where the state vector comprises weather variables (wind, temperature, humidity and pressure) and atmospheric CO2 mixing ratios. The system seeks the optimal fit between a suite of synthetic meteorological and satellite-based total column CO2 (XCO2) observations with forecasts from a global Earth system model. Given the incomplete observations and imperfect model, the simultaneous assimilation of weather and CO2 observations into our system yields the best approximation of atmospheric transport as well as its uncertainty, something not captured by other community carbon data assimilation and surface flux inversion systems which use a single realization of atmospheric transport. Our assimilation window is six hours, meaning that we have a time-evolving estimate of the atmospheric state, and its uncertainty (represented by the spread in the ensemble) at the resolution of six hours.

In Chapter 2, we employ this machinery to assess the capability of our carbon-weather DA system, along with satellite-borne XCO2 observations, to detect underreporting of CO2 emissions at the scale of a large country. In a series of OSSEs, we assimilate synthetic observations of XCO2 at the locations of (1) the Orbiting Carbon Observatory 2 (OCO-2) soundings and (2) a hypothetical observing system which observes globally at 1pm local time. Fossil fuel CO2 emissions are modified to have a -50% bias over China, but the observations are pulled from a model run where this bias is not present. We test whether the data assimilation system can detect the imposed bias by examining the near-surface innovation in CO2 mass in a method similar to the mass-balance inversion. We find that with the hypothetical observation strategy, we can recover half of the imposed bias, and that the ensemble mean of the near-surface CO2 tracks the truth during the daytime, but underestimates the truth during the unconstrained nighttime hours over the region of the imposed bias. For the OCO-2 strategy, we detect a signal at the location of the imposed bias that is obscured by problems such as observation coverage. We discuss potential additions to the observing system which could optimize the detection of biased emissions with our data assimilation machinery.

Chapter 3 presents results from OSSEs aimed at understanding the potential of OCO-2 total precipitable water (TPW) and XCO2 to improve weather forecasting capabilities. The hypothesis is that the time- and space-varying correlation between the satellite observable and wind in the Earth System Model could be used to improve the weather forecast where wind observations are sparse. We find that the TPW observations impact all meteorological state variables in the experiment, and that the XCO2 observations reduce weather forecast errors globally, and most significantly in the southern extratropics, in all meteorological fields except humidity. We conclude that both of these observation types from OCO-2 could serve as useful additions to the suite of observations assimilated by national weather forecasting centers.

In Chapter 4, we calculate global CO2 surface fluxes as a residual in the vertically-integrated CO2 tracer transport equation, using time-varying 3D-CO2 and meteorology reanalysis fields from a carbon-weather DA system that assimilates weather and XCO2 from the Atmospheric Infrared Sounder (AIRS). As AIRS XCO2 is weighted in the mid-troposphere, we find that the most significant impact on the surface flux calculation is in the tropics, especially over the Amazon and in the tropical Pacific, where intense convection mixes CO2 through the entire tropospheric column. We compare our posterior flux estimates to those made by CarbonTracker and find general sign agreement except in the Amazon region. Here we estimate a net annual sink of -0.26 PgC whereas CarbonTracker, which uses only surface observations, estimates a net annual source of about the same magnitude.