The switching impulse discharge characteristics of rod-plane gaps are widely recognized as one of the key fundamentals underpinning the external insulation design of ultra-high voltage (UHV) systems, serving as critical guidance for UHV engineering insulation design. Utilizing 360 sets of experimental data collected under varying altitudes and meteorological conditions, an eXtreme gradient boosting (XGBoost) prediction model was developed for the positive-polarity switching impulse discharge voltage of rod-plane gaps, with gap distance, air pressure, dry-bulb temperature, and absolute humidity selected as feature parameters. The test results demonstrated that an average relative error of 3.41% is achieved by the XGBoost model when comparing predicted voltage with experimental results on the test set, indicating superior predictive performance relative to other commonly used intelligent prediction methods. Notably, higher accuracy is observed under extremely cold conditions and short gap distances (≤1.5 m). Furthermore, the applicability of existing engineering methods is evaluated based on the predictions, which confirmed that greater adaptability is exhibited by the XGBoost algorithm in addressing discharge voltage prediction under complex environmental conditions.