Friction velocity (u*) underpins estimates of dust emission, but profile-derived values may depend on measurement height, wind direction and site conditions. We analysed multilevel tower observations from Tazhong (TZ) and Xiaotang (XT) during March–July 2024, together with ERA5 data and convergent cross mapping (CCM). TZ had greater terrain ruggedness than XT (TRI, 3.0 versus 1.4) and a slightly higher mean daily u* (0.24 versus 0.21 m s−1). The monthly maximum frequency of hourly u* above 0.30 m s−1 was 57.23 % at TZ and 48.46 % at XT. Across seven height configurations, H01 (1, 2, 4, 10 and 20 m) gave the lowest RMSE and highest Pearson correlation for both fitting approaches. Against an ERA5-derived neutral-profile reference, finite-L H01 fits had RMSEs of 0.186 and 0.191 m s−1 at TZ and XT, respectively, with correlations of 0.195 and 0.261. Neutral-only H01 fits had RMSEs of 0.118 and 0.129 m s−1 and correlations of 0.396 and 0.417. Directional differences were more pronounced at TZ, where the high-level H05 configuration generally produced larger median u* than lower-level configurations across wind sectors. CCM cross-map skill for 10 m wind speed was similar at both sites (maximum ρ = 0.24), whereas net-radiation skill was higher at TZ (0.44) than XT (0.25). Surface–air temperature contrast showed a similar site difference (0.38 versus 0.22), while air-temperature contrast was comparable (0.40 at both sites). These results show that height configuration and wind direction shape profile-derived estimates, while thermal associations vary between sites.