Despite the occurrence of riffle±pool sequences in many rivers there are few data concerning riffle±pool unit morphology. Of many criteria proposed to identify riffle±pool units, only three methods can be regarded as objective and robust. These are the zero-crossing', the spectral analysis' and the `control-point' methods. In this paper statistics are developed using the first two of these methods to describe the streamwise morphology of 275 riffles and 285 pools which form a continuous 32Á1 km reach of the bed of the River Severn in Shropshire, England.
Yalin's theoretical relationship between the average riffle:pool unit length (! p ) and channel width (W), ! p = 3W, applies to the River Severn. Reach-average riffle height (H) is a constant proportion of bankfull depth (h); typically H 0Á16h. Riffle height is a positive function of riffle length. Pool depth is a positive function of pool length. However, both riffle length and pool length increase more rapidly than the bed-level amplitude, such that long riffles or pools are relatively `flat'. As channel gradient reduces, bedforms flatten and become more asymmetric as riffle stoss sides and the proximal slope of pools lengthen at the expense of riffle lee sides and pool distal slopes. The statistical relationships between riffle steepness (H/L) and water depth are similar to those for equilibrium subaqueous dunes.
The Severn data are consistent with Yalin's theoretical analysis relating riffle bedform length (L r ) to water depth, i.e. L r = 2% h, wherein 1 for steep near-equilibrium bedforms but 2 to 3 as the relative depth decreases and riffles become long, low features. Theoretical consideration and turbulence data indicate that the frequency of coherent turbulent-flow structures associated with the riffle±pool mixing length in the Severn should be of the order of 50 to 100 s. The morphological similarity of the steepest River Severn riffles with dunes raises intriguing questions with respect to self-similar, convergent organization of periodic alluvial bedforms and to bedform dynamic classification particularly.