• A Composite Growth Index (CGI) for apple trees was developed by dynamically weighting SPAD, LAI, and tree height using CV. • XGBoost performed best during the flower falling and fruit setting stages (R² up to 0.556). • RF excelled in the fruit enlargement stage (R² = 0.844), while MLR was optimal for the coloring and ripening stages. • The CV-based CGI model achieved high accuracy (R² > 0.8) during fruit enlargement and coloring stages. • Stage-specific CGI models significantly improved growth monitoring accuracy, supporting precision orchard management. Apple trees exhibit a complex canopy architecture, a long phenological cycle, and strong spatial heterogeneity, which challenges timely and accurate growth regulation using conventional orchard management. To provide an interpretable vigor metric for precision management, we constructed a coefficient-of-variation–weighted comprehensive growth indicator (CGIcv) by integrating three complementary field traits: SPAD, Leaf Area Index (LAI), and tree height, and linked CGIcv to UAV multispectral observations across five key growth stages. Multispectral imagery was radiometrically calibrated, canopy and soil separation was performed, and vegetation indices were extracted as predictors. Five algorithms (MLR, PLSR, SVR, RF, and XGBoost) were evaluated to invert CGIcv and to identify stage-appropriate modeling strategies. Model performance was stage-dependent: RF achieved the best monitoring accuracy at the fruit enlargement stage ( R 2 = 0.84; RMSE = 0.15), whereas MLR performed best at the fruit coloring stage ( R 2 = 0.82; RMSE = 0.14). In contrast, early stages (flower falling and fruit setting) showed lower accuracy ( R 2 < 0.60), likely due to sparse canopy coverage and stronger soil-background interference in spectral signals. These results demonstrate that a CV-weighted CGI combined with stage-adaptive modeling can provide practical technical support for UAV-based orchard growth monitoring and precision management.
Wang et al. (2026) studied this question.