Supervised machine learning to constrain the CO2 background for city studies.

Determining an accurate and representative CO₂ mole fraction background for air entering a city is a fundamental component for greenhouse gas (GHG) emissions estimates using in-situ observations and atmospheric inverse modeling. However, there is currently no consistent, reliable method to determine CO₂ background that can be applied to cities. In Auckland, Aotearoa New Zealand, continuous in-situ observations and weekly flask measurements of CO₂ are made at four sites around the city. The upwind site is proposed as a background, and we subtract its CO₂ mole fractions observations from the urban sites’ mole fractions, to obtain the urban contribution, from observations and model. The primary background site is isolated from urban emissions, located close to the Tasman Sea at Manukau Heads, about 30km SW of Auckland city center. The prevailing wind brings oceanic air from SW, yet from other wind directions, the atmospheric mole fractions at this site can be influenced by anthropogenic and biogenic fluxes and are frequently affected by sea breezes, not necessarily representing the urban background. The lack of additional background sites covering other wind directions leads us to exclude such observations from the dataset, removing a significant amount of data. To optimize observations available for GHG emissions estimates, we propose a method to create a continuous (e.g., hourly) and accurate background for Auckland. Using over 2 years of in-situ observations, we will show the results of applying supervised machine learning for data gap-filling of missing CO₂ observations. Additionally, we use this method to help determine whether continuous atmospheric CO₂ measurements are necessary, or if sparse measurements (e.g., weekly air samples) would suffice to describe a background in cities like Auckland where incoming air is strongly oceanic. Furthermore, deeply investigating CO₂ background and implementing statistical methods to support in-situ observations will help guide experimental design for cities.