Gap-filling of flux measurements over a heterogeneous urban landscape

A small, but growing, number of urban flux towers measure surface-atmospheric exchanges of energy, water, and greenhouse gases by the eddy covariance method. Imputation of gaps in these measurements caused by low turbulence conditions and system failures is essential for obtaining annual sums of CO2 exchange and evaporation. Yet most gap-filling methods were designed for natural measurement sites such as forests and grasslands. In the urban environment, however, the assumptions on which those approaches are based are violated and well known temperature or light response models are not applicable because of urban footprint heterogeneity and localized CO2 emissions. Observation-based methods of machine learning can reveal intrinsic mechanisms by using inputs such as wind direction, footprint size, and continuous traffic data, making gap-filling results more accurate. Here, we report preliminary gap-filling results using such empirical approaches for >3 years of flux measurements from the KUOM tall tower in a suburban neighborhood of Minneapolis, Minnesota, USA. We also ran one of the most common gap-filling methods that has been used for natural systems as a baseline or null model. We found that CO2 and water vapor fluxes from the urban landscape showed higher variability than those from a nearby turfgrass lawn, in which fluxes closely followed environmental drivers of light and temperature. Higher variability was found in NEE measurements as compared to LE, due to the relatively greater heterogeneity of sources and sinks that influenced CO2 exchange in the urban landscape.