WAVELET ANALYSIS OF CARBON DIOXIDE FLUXES IN THE SURFACE LAYER ABOVE A DENSELY BUILT CITY CENTRE AT NIGHT J.A. Salmond 1* , T.R. Oke 1 and C.S.B. Grimmond 2 1 University of British Columbia, Vancouver, British Columbia, 2 Indiana University, Bloomington, Indiana 1. INTRODUCTION The thermal inertia, dryness and roughness of urban areas, especially in densely-built districts, typically results in the development of a weakly convective nocturnal surface boundary layer, often capped by a nocturnal inversion (Uno et al. 1992). Although this has significant implications for the dispersion of atmospheric pollutants, particularly from street canyons, few studies have focused on the development and characteristics of turbulence under these conditions. Carbon dioxide (CO 2 ) is a comparatively inert gas which, in the absence of space heating sources and localized industrial emissions, is primarily emitted into the urban boundary layer at street level from vehicle exhausts. Here we use wavelet analysis to examine the hypothesis that CO 2 concentrations can be used as a tracer to identify characteristics of pollutant venting from street canyons into the nocturnal urban boundary layer (UBL). 2. EXPERIMENTAL AND WAVELET METHODS A Licor–7500 and a RM Young 3-axis sonic anemometer were mounted on a tower located 44 m above the city center of Marseille (France) during the ESCOMPTE field project (Grimmond et al. 2002; Cros et al. 2002). Preliminary examination of the 10 Hz nocturnal turbulence data revealed intermittent bursts of CO 2 , superimposed upon a more stable mean background concentration (Figure 1a). All times are given in Local Standard Time (LST). This study uses the symmetric ‘Mexican hat’ wavelet to objectively isolate, and quantitatively analyse these events in the turbulent time series. Symmetric wavelets are coherent at the edge of the transition and thus effectively define the start of a change in conditions (Hagelberg & Gamage 1994). In order to ascertain the statistical significance of the results, the wavelet coefficients are compared to a theoretical red noise spectrum calculated from the auto-correlation of the time series using the technique proposed by Torrence and Compo (1998). The scalogram shown in Figure 1b demonstrates that wavelet analysis effectively identifies the presence and temporal location of significant structures within a time series. White indicates a high degree of correlation between the wavelet and the time series. The ‘cone of influence’ (area below which data are considered unreliable) is also marked. The wavelet coefficients that are statistically significant are circled in black. Due to the high temporal _______________________________________ Corresponding author address: J.A. Salmond, Dept. of Geography, 1984 West Mall, Vancouver, B.C. V6T 1Z2; e-mail: jsalmond@geog.ubc.ca 0 10 20 30 40 50 60 13.5 14 14.5 Time (min) Period (s) 0 10 20 30 40 50 60 2 8 32 128 512 2048 0 10 20 30 40 50 60 0 0.5 1 Scale-averaged Wavelet coefficients Carbon dioxide (mmol m -3 ) a) c) b) Figure 1. a) CO 2 time series, b) scalogram and c) scale- averaged wavelet coefficients (and significance level) for scales < 8 s for 2300 – 0000 (LST) June 26 th 2001. resolution of the scalogram the circles appear as black lines at small scales. The data for different time series consistently show intermittent periods of significant activity at turbulent scales (< 8 s) that correspond well with bursts of CO 2 . The mean modulus of the wavelet coefficients at scales < 8 s (the scales at which the largest proportion of significant structures can be found) was then calculated. Three significant events are identified in Figure 1c. Averaging over this range of scales limits the effect of instrumental noise or shear-generated turbulence, which due to the low wind speeds, is expected to be limited to very small scales (~2 s). This process was repeated for the temperature (T) time series to ascertain the relationship between the two scalar variables at these scales. 3. RESULTS AND DISCUSSION The technique developed above was applied to 30- minute CO 2 and temperature time series recorded during a period of anti-cyclonic conditions observed between 2100 - 0700 June 23 rd – 27 th 2001. The mean wind speed in this period was typically < 2 m s -1 and the friction velocity was < 0.3 m s -1 . On all four nights the prevailing wind flow was initially NW followed by a marked shift to SE at 0230, that suggests the development of a local wind regime. Sunset occurs at approximately 2100 and sunrise at 0600. The sensible heat flux remained positive throughout each of the four nights. Under these conditions shear turbulence is expected to be minimal across the top of the urban canopy layer (UCL) providing good conditions to identify the penetration of convective plumes from the UCL into the UBL. 14.2