滚动起源验证逆转多步PM10预测中的模型排名：XGBoost、SARIMA和持久性

Many air quality forecasting studies report gains from machine learning, but evaluations often rely on static chronological splits and omit persistence baselines, leaving the operational added value under routine updates unclear. Using 2,350 daily PM10 observations from 2017 to 2024 at an urban background monitoring station in southern Europe, a comparison is made between XGBoost and SARIMA against persistence under both static splits and a rolling-origin protocol with monthly updates. Horizon-specific skill and the predictability horizon, defined as the maximum horizon with positive persistence-relative skill, are reported. Static evaluation suggests that XGBoost performs well from one to seven days ahead; however, rolling-origin evaluation reverses these rankings, indicating that XGBoost does not perform as expected.