Abstract: 【Objective】This study aimed to reveal the diurnal and seasonal variation characteristics of airborne negative oxygen ion concentrations （NOIC） in urban forests，and to construct a high-precision prediction model. The findings were intended to deepen the understanding of the regulatory role of forest ecosystems on air quality， optimize forestry ecological management，and enhance the forest-based health and wellness functions.【Method】Based on continuous monito-ring data from the forest sample area in Chongming District of Shanghai during 2023-2024， the study analyzed the relationships between NOIC and factors such as the normalized difference vegetation index （NDVI），vegetation type，temperature， relative humidity，precipitation，wind speed，and visibility. Using random forest and a time series enhanced model （STXGB），NOIC prediction models were constructed. Hyperparameters were optimized via five-fold cross-validation and Bayesian optimization based on the Optuna framework. Model performance was evaluated by the fitting effects of coefficient of determination （R²），root mean square error （RMSE），and mean absolute error （MAE） on the model.【Result】The NOIC in evergreen broad-leaved forests was significantly higher than that in the other two vegetation types （P<0.05，the same below）. NOIC in salt marsh herbaceous vegetation was also significantly higher than that in deciduous broad-leaved forests. NDVI and NOIC exhibited a consistent seasonal trend throughout most of the year. In spring， NOIC showed an extremely significant positive correlation with temperature and ozone（O₃） concentration （P<0.01， the same below），with correlation coefficients of 0.256 and 0.175 respectively， and it was extremely significantly negatively correlated with particulate matter with diameter ≥2.5 μm （PM_2.5） concentration and relative humidity，with correlation coefficients both of -0.231. In summer，the correlation between NOIC and relative humidity shifted from an extremely significant negative correlation in spring to an extremely significant positive correlation. The seasonal variation of NOIC was as follows：autumn （2524 ion/cm³） > summer （2358 ion/cm³） > winter （1841 ion/cm³） > spring （1036 ion/cm³）. The proportion of NOIC rated as excellent or good exceeded 70% in all seasons. Particularly in summer and autumn，the proportion of NOIC reaching the excellent level was 100%. The STXGB model achieved a test set R² of 0.8468，an RMSE of 275.93 ion/cm³，and an MAE of 175.94 ion/cm³. Compared with the random forest model，the test set R² of the STXGB model nearly doubled，while RMSE and MAE were greatly reduced. When temperature and wind speed were both at high values，their interaction had a strong positive synergistic driving effect on the increase of NOIC. The combination with the most significant negative effect was visibility and wind speed. Models comparison results indicated that the STXGB model performed well，greatly outperforming traditional models.【Conclusion】The dynamic changes in NOIC are influenced by the synergistic effects of forest vegetation structure，meteorological factors，and regional pollutants. STXGB model demonstrates effectiveness and robustness in predicting NOIC under complex forest environments and can be applied to research on forest ecological monitoring and suitability assessment for forest-based health and wellness.