Intercomparison of biogenic CO2 flux models in four urban parks in the city of Zurich

Abstract. Quantifying the capacity and dynamics of urban carbon dioxide (CO2) emissions and carbon sequestration is becoming increasingly relevant in the development of integrated monitoring systems for urban greenhouse gas emissions. There are multiple challenges towards these goals, such as the partitioning of atmospheric measurements of CO2 fluxes to anthropogenic and biospheric processes, the insufficient understanding of urban biospheric processes, and the applicability of existing biosphere models to urban systems. In this study we applied four biosphere models of varying complexity (diFUME, JSBACH, SUEWS, VPRM) in four urban parks in the city of Zurich and evaluated their performance against in-situ measurements collected over almost two years on park trees and lawns. In addition, we performed an uncertainty analysis of gross primary productivity (GPP), ecosystem respiration (Reco), and net ecosystem exchange (NEE) of CO2 based on the differences between the estimates of the four models and compared the estimated uncertainties and biospheric fluxes with the monthly anthropogenic CO2 emissions of a wide urban area surrounding the four parks. The results showed that despite the large differences in model architecture, there was considerable agreement in the seasonal and diurnal GPP, Reco and NEE estimates. Larger discrepancies between the four models were found for lawn GPP compared to tree GPP, while for Reco the differences between lawns and tree areas were similar. On an annual scale, all models agreed, on average, that lawns acted as CO2 sources and tree-covered areas as CO2 sinks during the simulation period, with the exception of diFUME which simulated both tree and lawn areas as CO2 sources. diFUME and VPRM were more accurate in capturing the onset of the tree leaf growth in spring compared to JSBACH and SUEWS. On the other hand, JSBACH and SUEWS simulated soil water availability more accurately than the satellite-derived water index used by VPRM. The in-situ observations revealed a very high spatial variability in lawn Reco across the park areas. All models underestimated the lawn Reco during spring in sunny mowed locations, whereas the model simulations were closer to the observed Reco at un-mowed, partially shaded locations. The mean monthly uncertainties of biogenic NEE reached 1.1 μmol m-2 s-1, which is 13.4 % of the magnitude of the total CO2 balance over the studied area during the month of June. This balance was composed of a mean anthropogenic flux of 8.7 μmol m-2 s-1 and a mean biospheric flux of -0.5 μmol m-2 s-1. Overall, this study highlights the importance of properly accounting for the biogenic CO2 fluxes and their uncertainties in urban CO2 balance studies, especially during the vegetation growing season, and shows that even simple models, such as VPRM, can adequately simulate the urban biospheric fluxes when appropriately parameterized.