The manuscript presents a multi-layered perceptron (MLP) model designed for continuous learning to enhance the three-dimensional reconstruction of nitrate concentrations in oceanic environments. This approach improves the model's generalization capabilities by incorporating the nonlinear relationships among various marine surface environmental variables and surface nitrate levels. The model has been successfully applied in the Mediterranean Sea and Northeast Atlantic. Additionally, the study provides a reconstructed nitrate dataset spanning from 2010 to 2023, which holds significant value for marine ecology and environmental research. Below are my comments and suggestions regarding the manuscript:

1. The methodology for cross-validation should be elaborated upon. The current description does not adequately clarify how cross-validation is conducted, nor does it explain the rationale for utilizing all Argo data in the validation process.

1. The rationale for the inclusion of numerous variables needs to be articulated. A robust artificial intelligence model should be grounded in physical principles; achieving favorable results without this foundation may hinder the accurate representation of dynamic processes. It is essential to conduct experiments to determine whether the exclusion of certain variables could enhance the simulation outcomes. Furthermore, the absence of consideration for influential factors such as precipitation and river runoff warrants explanation.

1. The role of upper ocean phytoplankton should be critically evaluated, as they are significant sinks for nitrate. The analysis suggests that phytoplankton are not influential, which raises concerns regarding the validity of this finding. It is imperative to provide a detailed explanation of the model's capacity to capture the underlying physical and biological processes in the ocean. Is it possible that the inclusion of numerous other variables has overshadowed the impact of chlorophyll a (Chla)?

1. A more comprehensive discussion of the model's limitations is necessary, particularly regarding the consistency and discrepancies between the model's results and existing literature, including its performance in data-sparse regions.

Minor Comments：

Line 5: Provide the full name for “CMEMS”.

Line 36: Provide the full name for “SSN”

Line 45: Several studies, for example, Liu et al. (2022, http://doi.org/10.3390/rs14195021) have considered other factors besides SST.

Line 81-82: Need more explanation.

Line 94: “Pan-European” or “pan-European”? should be identical.

Line 95: Please check the definition of “shelf sea”. Apparently, the study area is beyond the shelf sea.

Section 2.2: There is no introduction to the CMEMS simulated nitrate data, which should be introduced detailedly.

Figure 3: It looks familiar to me, and although the author may have redrawn it from a certain document, the original source should be noted.

Line 274: “. the” should be “. The”.

“Figure” or “Fig. ” should be identical throughout the text.

Line 317-318: More introductions need to explain why and how.

The manuscript titled " A continual learning-based multilayer perceptron for improved reconstruction of three-dimensional nitrate concentration" presents a model to estimate the three-dimensional nitrate concentration in the Pan-European Ocean. Overall, this study is interesting, and the research and reconstruction data could contribute to a deeper understanding of regional ocean nutrient dynamics. The manuscript demonstrates good levels of research significance, innovation, and expression. However, before accept this work, there are some concerns need to be addressed.

Major Comments:

1. Sections 2.1: The study area is divided into two parts: MED and NEA. The observations in NEA are sparse. What is the rationale for selecting this region? How can the model's performance be ensured to remain stable in this area?
2. In Figure 6, the coupling of the continuous learning model improves overall performance. However, in terms of spatial distribution, errors increase at the locations of a few observation points. How should this phenomenon be understood, and is there potential for improvement? An additional discussion of the model's strengths and weaknesses is needed.
3. The simulated nitrate distribution pattern in Figure 6e appears somewhat unusual. How was the performance comparison of this widely used three-dimensional nitrate data product implemented here?
4. Regarding the interannual variation trend in Figure 11 and the abnormal increase after 2022, are there any studies or data that support similar conclusions? Caution is required when evaluating and interpreting these phenomena.
5. The feature importance calculated through SHAP includes both positive and negative values. The manuscript evaluates these values based on absolute values, which may overlook the sign of the contributions. Consider whether a deeper assessment of feature contributions using the positive and negative relationships of SHAP is possible.

Minor Comments:

1. The data processing steps and validation methods need more detailed explanation, such as interpolation for missing values and the division of the test set.
2. The manuscript's descriptions should be consistent, such as "Figure" or "Fig," and "pan-Europe" or "Pan-Europe."
3. The titles of Figures 6 and 7 are not sufficiently clear and would benefit from refinement and reorganization.