The quadrant analysis of the intense tangential Reynolds stress in plane turbulent channels is generalized to three-dimensional structures (Qs), with special emphasis on the logarithmic and outer layers. Wall-detached Qs are background stress fluctuations. They are small and isotropically oriented, and their contributions to the mean stress cancel. Wall-attached Qs are larger, and carry most of the mean Reynolds stresses. They form a family of roughly self-similar objects that become increasingly complex away from the wall, resembling the vortex clusters in del Álamo et al. (J. Fluid Mech., vol. 561, 2006, pp. 329–358). Individual Qs have fractal dimensions of the order of $D= 2$, slightly fuller than the clusters. They can be described as ‘sponges of flakes’, while vortex clusters are ‘sponges of strings’. The number of attached Qs decays away from the wall, but the fraction of the stress that they carry is independent of their sizes. A substantial fraction of the stress resides in a few large objects extending beyond the centreline, reminiscent of the very large structures of several authors. The predominant logarithmic-layer structure is a side-by-side pair of a sweep (Q4) and an ejection (Q2), with an associated cluster, and shares dimensions and stresses with the conjectured attached eddies of Townsend (J. Fluid Mech., vol. 11, 1961, pp. 97–120). Those attached eddies tend to be aligned streamwise from each other, located near the side walls between the low- and high-velocity large-scale streaks, but that organization does not extend far enough to explain the very long structures in the centre of the channel.

The minimal simulation boxes of the buffer layer of turbulent channels can be extended to the logarithmic and outer regions, where they contain a segment of streamwise velocity streak, and a vortex cluster. Smaller boxes restrict "healthy" turbulence closer to the wall, to a layer whose thickness scales with the spanwise size of the box. These minimal boxes burst quasiperiodically, and the bursting period for a band of wall distances grows linearly away from the wall, independently of the box size within the limits within which turbulence is well represented.

The nocturnal atmospheric boundary layer (ABL) poses several challenges to standard turbulence and dispersion models, since the stable stratification imposed by the radiative cooling of the ground modifies the flow turbulence in ways that are not yet completely understood. In the present work we perform direct numerical simulation of a turbulent open channel flow with a constant (cooling) heat flux imposed at the ground. This configuration provides a very simplified model for the surface layer at night. As a result of the ground cooling, the Reynolds stresses and the turbulent fluctuations near the ground re-adjust on times of the order of L/u τ , where L is the Obukhov length scale and u τ is the friction velocity. For relatively weak cooling turbulence survives, but when Re L = Lu τ /ν 100 turbulence collapses, a situation that is also observed in the ABL. This criterion, which can be locally measured in the field, is justified in terms of the scale separation between the largest and smallest structures of the dynamic sublayer.

The dynamics of the sublayer and buffer regions of wall-bounded turbulent flows are analysed using autonomous numerical simulations in which the outer flow, and on some occasions specific wavelengths, are masked. The results are compared with a turbulent channel flow at moderate Reynolds number. Special emphasis is put on the largest flow scales. It is argued that in this region there are two kinds of large structures: long and narrow ones which are endogenous to the wall, in the sense of being only slightly modified by the presence or absence of an outer flow, and long and wide structures which extend to the outer flow and which are very different in the two cases. The latter carry little Reynolds stress near the wall in full simulations, and are largely absent from the autonomous ones. The former carry a large fraction of the stresses in the two cases, but are shown to be quasi-linear passive wakes of smaller structures, and they can be damped without modifying the dynamics of other spectral ranges. They can be modelled fairly accurately as being infinitely long, and it is argued that this is why good statistics are obtained in short or even in minimal simulation boxes. It is shown that this organization implies that the scaling of the near-wall streamwise fluctuations is anomalous.