Depth representation dominates methodological sensitivity in soil health assessment of semi-arid apple orchards on the Loess Plateau

Accurate soil health assessment in perennial orchards requires indicators that represent functionally relevant soil-profile conditions rather than topsoil properties alone. This is particularly important in semi-arid systems, where nutrient-enriched surface layers may be vertically decoupled from subsurface constraints affecting root growth, aeration, and water availability. We developed a minimum data set (MDS)-based soil health index (SHI) using 15 physical, chemical, and biological indicators from 27 Fuji apple orchards across three counties of the Weibei Upland on the Loess Plateau, China. We compared the sensitivity of SHI outcomes to depth representation, indicator weighting, and scoring functions. The reference SHI used profile-informed indicator values based on functionally relevant target-depth intervals, whereas alternative schemes included topsoil-only representation (0–20 cm), AHP- and CRITIC-based weighting, and fully linear or nonlinear scoring. Depth representation was the dominant source of methodological divergence. Compared with the profile-informed reference SHI, the topsoil-only scheme overestimated soil health by 6.08 points on average, equivalent to 10.9% of the mean reference SHI. This positive bias occurred in 26 of 27 orchards and was observed across all three counties. Bland–Altman analysis confirmed systematic overestimation by the topsoil-only scheme, although orchard rankings remained strongly correlated with the reference scheme (Spearman’s ρ = 0.82, p < 0.001). By contrast, weighting strategy had little effect on SHI values, while scoring strategy shifted absolute values but largely preserved orchard ranking. Total SHI was most strongly associated with the chemical sub-index, whereas management-category effects were not statistically detectable under the present observational design. These results indicate that, within the tested SHI framework, profile-informed depth representation was more important than weighting or scoring choices for reducing diagnostic bias. Topsoil-only assessment may support broad ranking, but it should not be treated as an equivalent substitute when the goal is to represent profile-related constraints and guide site-specific management.