Evaluating the monitoring capability of uncrewed air vehicle (UAV) remote sensing imagery for spring maize growth status provides critical guidance for agricultural field management and yield increase. This study developed a composite growth index (CGI) for spring maize under drip irrigation by integrating five growth parameters (leaf area index, chlorophyll content, etc.) through a combined coefficient of variation-entropy weighting approach. A growth status diagnostic model for spring maize was developed by integrating UAV multispectral data with a Recursive Feature Elimination based extreme gradient boosting (RFE-XGBoost) algorithm. Compare the effect of feature variable optimization on model robustness and conduct interpretability analysis of the model using Shapley values. The results demonstrate that CGI calculated using the entropy weight method accurately reflects the growth status of spring maize. The optimal features selected differ across the four growth stages of maize. The optimized feature variables significantly enhanced the robustness of the XGBoost model on the test set, with R-2 increasing by 1.89-20.83 %, MAE decreasing by 1.23-10.28 %, and RMSE reducing by 1.71-10.20 %. The optimized feature parameters, when used as inputs for the XGBoost model, showed strong performance on the test sets from the V6 to R6 stages (R-2 = 0.602-0.791). The model's CGI prediction capability during the Total stage was significantly superior to that of the V6-R6 stage models (R(2)Total = 0.863). Moreover, the XGBoost model outperformed other benchmark models, including random forest (RF), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM), thereby establishing a new benchmark for UAV-based monitoring of spring maize growth status. Shapley analysis precisely quantified the impact of characteristic parameters from different growth stages on model performance. The generated plot-scale CGI spatiotemporal distribution maps for drip-irrigated spring maize showed strong agreement with ground observations, further confirming the robust capabilities of the modeling framework developed in this study. This study establishes a theoretical foundation for constructing integrated growth inversion models in cold-arid regions while offering a practical pathway for large-scale, low-cost agricultural management.