This is a quite complete paper about tropospheric ozone in East and Southeast Asia in the past 25 years. Many authors are involved, and the study considers an extensive station network and modelling studies where 80% of data are for training and 20% of data are for testing. The analysis is focused on the summertime and surface concentrations, profiles and trends are investigated. Finally, the contribution of anthropogenic emissions and meteorology are quantified. Due to the extension and intensity of this analysis, it merits to be published in Atmospheric Chemistry and Physics, although the following minor issues should be answered by the authors.

Some results could be mixed with the discussion in the current paper, due to their comparison with other studies. Perhaps the authors could indicate if both sections could be more separated.

Limitations of this study could be highlighted. Moreover, a comment about results extrapolation to the future would be acknowledged by the readers.

Minor remarks.

L. 367. “aaply” or “apply”?

L. 621. “Aisa” or “Asia”?

This study investigates tropospheric ozone trends and attributions over East and Southeast Asia from 1995 to 2019 using an integrated approach that includes statistical methods, machine learning models, and multiple chemical transport models (CTMs). The paper provides valuable insights into the key drivers of ozone changes, including anthropogenic emissions and meteorological influences. The methodology is sound.

However, the manuscript has several major issues, particularly regarding the uncertainty quantification of models, statistical method validation, the role of meteorological factors, and the clarity of explanations. Below, I provide detailed comments with specific suggestions for revision. Once these revisions are addressed, the manuscript will provide a stronger contribution to understanding ozone trends in East and Southeast Asia.

Page 4, Line 60: "Ozone is growing especially fast over the densely populated regions of East and Southeast Asia."

"Ozone concentrations are increasing rapidly over..."

Page 7, Lines 170-180: "We adopt one conventional statistical method, i.e., the multiple linear regression (MLR) method, and two machine learning models, i.e., the ridge regression (RR) and random forest regression (RFR) methods."

Comment: These methods are useful, but have their assumptions been checked? For example:

MLR: Have you tested for multicollinearity e.g. Variance Inflation Factor, VIF?

RR: Why was Ridge Regression chosen over Lasso Regression, which can also reduce collinearity?

RFR: Decision tree-based models can be prone to overfitting, especially when trained on high-dimensional meteorological variables. 

Page 8, Lines 200-210: "We follow standard machine learning practices by splitting the training and testing sets for RR and RFR."

Comment: It is unclear whether time-series dependencies were considered when splitting the dataset. Standard random splitting may not be appropriate for time-dependent ozone trends. 

Page 8, Line 200: "We follow standard machine learning practices by splitting the training and testing sets for RR and RFR." – misleading

"We follow standard machine learning practices by splitting the dataset into training and testing sets for RR and RFR."

Page 13, Lines 395-400: "A notable upward trend in surface downward solar radiation is discernible across Southeast Asia, whereas a decline is evident in most parts of China."

Comment: This statement is correct but lacks explanation. The decrease in solar radiation over China is likely linked to aerosol reductions, while increases in Southeast Asia may be due to decreasing cloud cover. Were aerosol-cloud interactions important here?

Page 13, Lines 405-415: "Significant positive trends in BVOC emissions are shown in eastern China, Southeast Asia, and parts of India, in contrast to a significant decline in Myanmar."

Comment: Why Myanmar is different?

Page 14, Lines 450-460: "We find that, overall, all models applied in this study capture the observed ozone vertical profiles over East Asia and Southeast Asia."

Comment: From my perspective, the large-scale vertical distribution of simulated ozone should not vary significantly across different models. Therefore, simply providing an average bias difference of a few ppb is clearly insufficient to indicate good model performance. Moreover, the discrepancies at the surface are excessively large in Southeast Asia.

Page 15, Lines 470-475: "Overall, all models capture the spatial distributions of surface ozone over ESEA, as indicated by the high spatial correlation coefficients between the observed and simulated values ranging from 0.50-0.78."

Comment: While spatial correlation coefficients provide some measure of agreement, they do not reveal the absolute differences in ozone levels. I suggest including scatter plots to compare observed vs. modeled values more rigorously. 

The y-axis values in Figure 6 are too large; a range from 0 to 100 is not appropriate. In most regions, ozone changes are relatively flat.

In Figure 7, the term "other anthropogenic" is unclear; it actually refers to emissions from outside ESEA. Additionally, regarding the impact of climate change, the two models exhibit different spatial differences, including for surface ozone in Figure 11. The regional discrepancies should be further explained by identifying the specific climate changes responsible for these variations.

Another point regarding the tropospheric ozone column changes in Figure 7: GEOS-Chem does not exhibit strong sensitivity to individual variables. Why does its change in the Base experiment appear comparable to that of WRF-CMAQ? This needs further clarification.

The description of OPE (Ozone Production Efficiency) is too brief. Would it be worthwhile to include a separate figure for it? The authors have already used H₂O₂/HNO₃ as an indicator, which should be sufficient. However, in Figure 12, the color bar is unclear, making it difficult to interpret how this indicator changes. 

Figure 13 presents the predictive performance of different statistical learning models. However, R² alone is not sufficient for evaluation. The key aspect here is the model error, i.e., how much of the ozone variation can actually be explained by the model, rather than just the correlation.

From Figure 14, it is evident that the influence of meteorology in all models appears relatively flat compared to the observed values. This suggests that the observed ozone variations cannot be fully explained by meteorology alone. Here, the emission impact is calculated as the observed values minus the meteorology-driven component, i.e., the residual. How exactly are the emission-driven values calculated? This should be clearly explained in the figure description.

However, if we instead subtract the emission-driven values from the observed values, would the resulting values be exactly equal to the meteorology-driven component shown in Figure 14?

It would be helpful to include an ensemble mean of all statistical and physical models in Figure 16. Since all models inherently have biases and uncertainties, an average would help mitigate individual model errors.