Remote sensing has assisted precision fertilization by mapping crop and soil nutrient status, but these layers are often assessed separately. We developed a matched-unit diagnosis framework for 110 soybean and maize units at Heshan Farm, China. Linear spectral models estimated leaf nitrogen (N) and phosphorus (P) from ZiYuan 1-02D imagery and soil alkali-hydrolyzable nitrogen (SAN) and available phosphorus (SAP) from multi-source bare soil imagery. Soil sampling occurred 27–32 days after foliar sampling, supporting the retrospective comparison of post-harvest residual soil and earlier canopy nutrient status. Canopy models achieved moderate validation performance (R² = 0.424–0.571), while SAN and SAP models yielded R² values of 0.656 and 0.583, respectively. Matched-unit analyses revealed positive SAN–leaf N and SAP–leaf P associations, whereas cross-nutrient relationships were weak. The soil and canopy N-P status indices (SNPSI and CNPSI) identified four relative status classes, with correspondence quantified by the N-P coupling index (NPCI). Spatial block validation of the downstream CNPSI regression yielded a higher root mean square error than random cross-validation (1.712 versus 0.561), indicating limited spatial transferability. The framework supports targeted field verification; application elsewhere requires local recalibration and independent spatial validation.