Table 1.

Details of the various datasets used

Case	Method	Acronym: Authors	H / h	$\mathit{\Delta }/h$
Staggered cubes	DNS	LC: Leonardi and Castro (2010)	8.0	1/100
Staggered and aligned cubes	DNS	CTCB: Coceal et al. (2006)	8.0	1/32
Staggered and aligned cubes	DNS	BCTB: Branford et al. (2011)	8.0	1/32
Staggered cubes, various angles	LES	CCTBBC: Claus et al. (2012)	4.0	1/25
Staggered, random height	LES	XCC: Xie et al. (2008)	10.0	1/16
Aligned cubes (b.layer)	LES	CP-A: Cheng and Porte-Agel (2015)	$\approx 12$	1/16
Staggered and aligned cubes (b.layer)	LES	YSMM: Yang et al. (2016)	$\approx 24$	1/8
Aligned $h\times 2h\times h$ blocks	DNS, LES	CXFRCHHC: Castro et al. (2016)	12	1/12

Except for CP-A’s and YSMM’s LES, where H refers to the approximate boundary-layer depth, all cases were channel-flow computations, with H the channel half-height. All cases except CCTBBC and CXFRCHHC considered only arrays that were flow-aligned—i.e. wind direction normal to the faces of the obstacles. For the LES cases, the subgrid models used were either the standard Smagorinsky model (CCTBBC, XCC, CXFRCHHC), the modulated gradient model of Lu and Porte-Agel (2010) (CP-A), or the Vreman (2010) model (YSSM)