The combustion and flame spread behaviors of discrete fuels often depend on the distribution of the fuel arrays, including the fuel bed width and the spacing of the fuel arrays. In this study, scenarios with six different numbers of columns (n = 1-11 with an interval of 2) and nine different spacings (S , 1-9 mm) were designed. By varying n and S , 54 groups of experimental scenarios over thermally thick birch rod arrays were studied to determine the effects of fuel bed width and spacing. The experimental results showed that the critical spacing for multi-column arrays was larger than that for single-column arrays. Moreover, flame extinction was found when n > 1 and S = 1 mm . The value of S had a limited influence on the mass loss rate when S > 2 mm , but restrained the mass loss rate when S <= 2 mm . A piecewise correlation between a newly defined dimensionless mass loss rate and the exposed area ratio was proposed. Moreover, a prediction model of the mass loss rate was developed, which agreed reasonably well with the experimental data. With increasing spacing, the flame height showed three stages: increasing, relatively stable, and decreasing, which were dominated by the balance between air entrainment and radiant heat feedback. The three processes were further described by a piecewise correlation between the dimensionless flame height and fuel bed width. Based on the assumption of a constant incident heat flux, a prediction model of the global flame spread rate was built, which presented better predicted results than the previous models. In addition, it was found that the newly defined dimensionless global flame spread rate linearly increased with increasing porosity. (C) 2020 The Combustion Institute. Published by Elsevier Inc. All rights reserved.