Existence of universal features, in the inertial sublayer and the outer region of turbulent boundary layer, has been suggested by Raupach (1981). Over thirty years of studies on smooth-wall boundary layers have resulted in a consensus based on a vortex organization model in the outer region summarized by Adrian et al. (2000). Coherent structures from this model, such as low-speed streaks (or low momentum region, Tomkins and Adrian, 2003), seem to exist over rough-wall independently of its nature. Wind tunnel (Cheng and Castro 2002) and LES (Kanda et al., 2004, Kanda, 2006) studies over cube arrangements showed that the low speed streaks are similar to those over a smooth wall. However, in the roughness sublayer (RSL) of a rough-wall boundary layer, turbulence mechanisms depend on the nature of the roughness elements (Krogstad and Antonia 1999). Castro et al. (2006) confirmed the influence of the roughness elements. A two-scale behaviour has been evidenced: small-scale eddies, less than the mean height h of the buildings in size, generated by the separated shear layers and larger-scale structures from the boundary layer flow (Castro et al. 2006). Focusing on the flow dynamics within the building canopy, Coceal et al. (2007) found from their DNS results that eddies shed off the vertical edges of the buildings are rotated by mean shear. Furthermore, strong canopy-top shear layers intermittently penetrate into the canopy, impacting upon buildings downstream and driving a recirculation in front of the buildings. Takimoto et al. (2011) described existence of two characteristic modes from instantaneous canopy flow: flushing and cavity-eddy modes. Christen et al. (2007) have indicated the existence of very large roller structures. In their study of a high Reynolds number boundary layer over a smooth wall, Marusic and Hutchins (2007) showed the existence of very-large-scale motions consisting of meandering elongated low- and high-speed regions. These structures were found to play a significant role in redistributing small-scale turbulent motions throughout the boundary layer. Recently, Inagaki et al. (2012) suggested that the turbulent organized structures represented by low and high speed streaks, control canopy flow events, as flushing and cavity eddies. This relationship confirms a top-down mechanism (Hunt and Morrison, 2000, Marusic et al., 2010).
Following Townsend's original idea (Townsend, 1976), Inagaki and Kanda (2010) and Castillo et al. (2011) used a spatial decomposition method to separate inner and outer layer turbulence. Building on these previous results, the aim of the present study is to use the same type of approach to decompose the flow developing above the canopy in a large-scale motions part originating from the boundary layer and a small-scale motions part corresponding to the emerging canopy flow. To perform this study, an atmospheric boundary layer is simulated over a staggered cube arrangement in a wind tunnel. SPIV is used to measure 3-component velocity fields in two vertical planes, aligned and normal to the mean flow, and two horizontal planes at z/h = 1.5 and 3. Analysis of the one-point statistics confirms the existence of different regions above the canopy. Spatial information provided by SPIV clearly shows the influence of the roughness elements on the lower part of the flow. In the normal plane where mean velocities, standard deviations and turbulent fluxes are heterogeneous in spanwise direction with height, between z/h = 1 and 2. Above the RSL, a constant flux region is observed. Observation of instantaneous velocity fields show existence of structures associated with the roughness elements in the RSL, and large-scale structures above. Combining a two-point correlations analysis in each plane and the hairpin packet model of Adrian et al. (2000), the coherent structures (quasi-streamwise streak and hairpin vortex) are identified and, streamwise and spanwise length scales are extracted. Then, a spatial filtering of the velocity fields is used to separate the scales from the canopy flow from those from the boundary layer flow. In the final paper, spatio-temporal interactions between the two flows will be discussed using conditional averages and inter-scale two-point correlations.