We defined f micro as the liquid–solid contact fraction of the microstructure and f nano as the liquid–solid contact fraction of the nanostructures (Fig. 3a, Supplementary Discussion section 2.3). The values of both θ* and θroll-off show that these surfaces can maintain high levels of water repellency after abrasion if f micro is lower than 8% (Fig. 3d, e). These results are consistent with the ideal Cassie–Baxter model, providing further evidence that non-wettability is independent of the size of inverted-pyramidal structures. However, the smaller the microstructures,the more extensive are the changes to the liquid–solid contact fraction Δf micro after the same abrasion fracture (Supplementary Fig. 17). As such, the size of the microstructure armour could be tailored for various practical applications.To further understand the effect of the liquid–solid contact fraction on controlling the non-wettability after abrasion, pull-off force maps of the superhydrophobic surfaces were obtained using scanning droplet adhesion microscopy. Surfaces with two different values of f micro (~2% and ~7.8%) were measured before and after abrasion. As shown in Extended Data Fig. 1 and Fig. 3f, the pull-off forces before abrasion were similar for both surfaces. After abrasion, the damage to the hydrophobic layer on the top of the armour resulted in an increase in pull-off force at the same f micro. However, the pull-off forces on the high-f micro surface increased more rapidly than those on the low-f micro surface, consistent with the trend of θroll-off (Fig. 3e). Extended Data Fig. 2 illustrates a water jet impinging onto the surface before and after abrasion, with an incidence angle of 28° and a volume velocity of 6 ml min−1. These results agree with our previous wettability measurements, showing that for lower values of f micro the deflected angles are higher—that is, less energy dissipation occurs when water bounces way from the surface. Similar results were obtained when analysing the impact of water droplets, further supporting this principle (Extended Data Fig. 3, Supplementary Video 4).