With the continuous advancement of atmospheric environmental governance, particulate matter pollution has significantly decreased, but at the same time, ozone pollution has become increasingly severe. Therefore, constructing a long-term ground-level ozone dataset for China is essential for understanding the distribution and transmission of ground-level ozone and promoting the coordinated control of fine particulate matter and ozone. In this study, by combining the advantages of two machine learning algorithms, extreme random tree and extreme gradient boosting, we use ozone monitoring data, remote sensing products, and atmospheric reanalysis data to construct a daily maximum eight-hour average ozone (MDA8 O₃) concentration estimation model for China's surface. The model accuracy is validated from sample, space, and time perspectives, and its spatiotemporal applicability is verified at annual, quarterly, historical, and regional scales. And at last, a China-wide ozone data product covering the period of 2001 to 2020 is derived. The results show that: (1) the ozone estimation model, which combines the advantages of the two algorithms, exhibits excellent accuracy, with R² ranging from 0.89 to 0.95 and root mean square error (RMSE) ranging from 10.73 to 15.56 μg/m³; (2) The multiple spatiotemporal verifications indicate that the model constructed in this study can be applied to large-scale, long-term ozone estimation work in the China region; (3) The ozone data product constructed in this study can well reflect the spatiotemporal differentiation of ground-level ozone at the monthly and annual scales, and display the spatiotemporal changes of ozone concentration intuitively.