Amplified by global climate change, the crisis of agricultural water resources has intensified. The estimation of water status in the deep root zone of woody crops using canopy spectra is fundamentally constrained by the progressive attenuation of root-shoot hydraulic coupling signals. To address this mechanistic bottleneck, a multilevel irrigation gradient experiment was conducted in a walnut orchard in Aksu, Xinjiang, to systematically elucidate the cross-scale transfer mechanisms linking root-zone water conditions to canopy spectral responses. The results identified the 20-60 cm soil layer as the key functional zone for root-shoot hydraulic coupling, wherein water availability mediates canopy spectral reflectance by modulating leaf structural parameters and photosynthetic pigment status. Based on this, a three-band optimized spectral index-termed the Ratio and Division Index of Variability (DIV)-was developed using full-band combinations, enabling the synergy of direct water absorption bands and indirect stress-responsive bands. This approach significantly improved the sensitivity for detecting water content in the middle soil layer (40-60 cm), with an R2 increase of 10.4%. Furthermore, Bayesian optimization was employed to fuse spectral indices with band reflectance information, and machine learning models were established to estimate water content across multiple soil layers, achieving high estimation accuracy for the 20-40 cm and 40-60 cm layers (R2 = 0.7878 and 0.7585, respectively). SHapley Additive exPlanations (SHAP) analysis revealed a systematic shift in driving mechanisms with increasing soil depth, and coupled with t-SNE visualization, further elucidated the physiological boundary underlying the unreliable estimation accuracy below 80 cm. Collectively, this study defines a theoretical depth limit of 80 cm for retrieving root-zone water status in walnut trees from canopy spectra, providing cross-scale mechanistic insights and a technological pathway for precision water monitoring and smart irrigation in woody crops cultivated in arid regions.