Review Reports - Research on the Prediction of Liquid Injection Volume and Leaching Rate for In Situ Leaching Uranium Mining Using the CNN–LSTM–LightGBM Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

My reviewer report is in the attached file.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

Thank you for your thoughtful comments.

Comment 1: " It was stated that one of the article's objectives is to increase uranium recovery rates. In demonstrating the relevance of uranium recovery to precisely estimating uranium leaching rates in ISL operations, it would be beneficial to discuss not only primary uranium mining using ISL techniques but also uranium mineral recovery from uranium-bearing waste. This will further demonstrate the article's concrete academic contribution to its readers. "

Response: After revision, this paragraph highlights the model's accurate prediction of leaching rate. This capability, in turn, facilitates the adjustment of core operations (such as injection strategies) and contributes positively to enhancing uranium recovery rate—thereby underscoring the connection between the model and the uranium ore recovery process.

Revised version: “Despite progress in applying machine learning to ISL uranium mining, significant challenges remain. A critical technical priority is optimizing injection strategies to improve uranium leaching rates and recovery rates—an effort closely tied to the overarching goal of efficient resource utilization. To address this, we propose an attention-based CNN-LSTM-LightGBM fusion model for precise prediction of uranium leaching rates in ISL operations. Building on these predictions, the model provides targeted support for optimizing uranium recovery processes. By clarifying the relationships between key parameters in the leaching process—such as daily lixiviant volume and sulfuric acid concentration—and leaching efficiency, it helps adjust core operations like injection strategies. This, in turn, indirectly enhances uranium recovery rates while meeting the current demand for efficiency in ISL resource recovery. The proposed approach offers actionable guidance for improving leaching performance and ensuring consistency with engineering expectations.”

Comment 2: " Based on the results of your study, to what extent can environmental pollution be reduced? Can an estimate be made of the level of environmental pollution reduction achieved by the method used in these mining operations, taking into account environmental pollution parameters and citing other references in literature? Using data from these uranium mining operations, how much CO2 emissions will be reduced in the environment thanks to your model? Or how much harmful gases will be reduced? Providing these estimated calculations would be beneficial to encourage the use of the model you analyzed. This could be the subject of a separate article. However, including calculable estimated data in this study, if possible, would improve the quality of the current article."

“The proposed CNN-LSTM-LightGBM fusion model with attention mechanisms provides reliable predictions of uranium leaching rates during in-situ uranium leaching processes. Comparative experiments demonstrate that the proposed model achieves MAE of 0.085%, MAPE of 0.833%, and RMSE of 0.201% on the validation set, outperforming six heterogeneous and extensively utilized methods including SVR, MLP, and KNN across three metrics. Ablation experiments quantifying component contributions through systematic exclusion reveal the fusion model's advantages in MAE, MAPE, and RMSE on the validation set over standalone CNN, LSTM, and CNN LSTM counterparts. The research outcomes, by enabling accurate prediction of uranium leaching rates and thereby optimizing lixiviant injection strategies, not only enhance the efficient utilization of uranium resources but also provide actionable insights for reducing environmental pollution.”

Revised version（add in the conclusion）: “There is still room for expansion in the dimensionality of data indicators for prediction and the coverage of mining area samples in this study’s model, and the depth of integration between the model and professional mechanistic knowledge such as geological fluid dynamics remains insufficient. Future work should focus on integrating the model with the full-process environmental impact assessment of in-situ leaching uranium mining, quantifying the long-term effects of injection strategies on ecosystems such as groundwater, and exploring interdisciplinary integration with hydrological simulation and ecological risk assessment, so as to provide more systematic guarantees for in-situ leaching uranium mining technology in moving toward a high-quality development direction featuring low environmental impact and high resource utilization.”

Response: “Unfortunately, during the research process, there were no quantitative indicators for the experimental evaluation of environmental impacts, which is a limitation. Therefore, we have revised the conclusions: at the end, we added some limitations of this study as well as an outlook on future research directions, and we hope to have the opportunity to conduct research in this area in the future.”

Comment 3: " It would be beneficial to further elaborate on the theoretical and practical contributions of your study. - The following should also be explained: Under what conditions might this model yield erroneous results when used in mines? Or could similar limitations arise in the application of this model? - It would be beneficial to discuss the limitations of the study and what future studies could address these shortcomings. "

“In conclusion, the proposed fusion model presents a deep learning-integrated approach for in-situ uranium leaching technology, while offering a methodological reference for related industrial domains. This innovation holds significant potential to propel associated industrial applications toward more sophisticated development paradigms”

Response: “Due to the limited dimensions of data collection, the model finds it difficult to take all factors into account and thus inevitably cannot make accurate judgments in some cases. Future research can focus on integrating knowledge from other fields, analyzing data from more perspectives, and providing more systematic support for the prediction of uranium leaching rate.”

Thank you very much for your valuable time, careful review, and constructive comments on our manuscript. We will earnestly address each of your suggestions to further refine the work, and we greatly appreciate your guidance in improving the quality of this paper.

Sincerely,

Dr. Zhi-feng Liu

Sept. 16, 2025

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript entitled “Research on Prediction of Liquid Injection Volume and Leaching Rate for In-situ Leaching Uranium Mining Using CNN-LSTM-LightGBM Model” presents an innovative study aimed at improving prediction accuracy in ISL uranium mining through the integration of CNN, LSTM, LightGBM, and an attention mechanism. The authors show that their hybrid model outperforms both its component models and other benchmark ML approaches, demonstrating strong results with very low MAE, MAPE, and RMSE values. The topic is highly relevant for process optimization in mining and resource recovery, and the proposed approach can contribute significantly to the intelligentization of uranium mining operations. The work is of interest and, after major revisions, may be suitable for publication in Processes.

The introduction provides a solid background on ISL uranium mining and the application of machine learning. However, it would benefit from a broader contextualization within the field of process intensification and hydrometallurgical advancements. I recommend citing the following paper to strengthen the connection: Chemical Engineering and Processing-Process Intensification, 155, 108015. This reference will help link your work to the broader literature on process intensification and resource recovery.
Please expand on the novelty of your proposed model in comparison to other recent hybrid ML approaches beyond those cited. A critical discussion of why CNN-LSTM-LightGBM with attention is particularly suited for ISL would improve clarity.
Discuss the practical implications of the prediction improvements: for example, how much production efficiency or environmental risk reduction could realistically be achieved in operational ISL systems.
While the manuscript is generally understandable, there are grammatical and typographical errors that should be corrected. Professional English editing is strongly recommended to improve clarity and readability.
The conclusions currently restate results. They should also reflect on limitations (e.g., data availability, generalizability across different geological conditions) and outline future work directions (e.g., integration with real-time monitoring, transferability to other leaching systems). Please avoid overly broad statements such as “propel industrial applications toward more sophisticated paradigms” without concrete examples. A more specific and balanced conclusion will strengthen the manuscript.
Some Minor Comments
1. Line 10-15: Please clarify the description of “13 technicians’ understanding of the mining area”, this statement needs refinement.
2. Line 91-93: The description of CNNs is somewhat generic. Consider shortening and focusing more directly on how CNN features are applied in this specific context (time-series features for uranium leaching prediction).
3. References: Ensure consistency in formatting (journal names should be italicized and follow Processes reference style).

Author Response

Dear Reviewer,

Thank you for your thoughtful comments.

Comment 1（introduction）: "The introduction provides a solid background on ISL uranium mining and the application of machine learning. However, it would benefit from a broader contextualization within the field of process intensification and hydrometallurgical advancements. I recommend citing the following paper to strengthen the connection: Chemical Engineering and Processing-Process Intensification, 155, 108015. This reference will help link your work to the broader literature on process intensification and resource recovery. "

Response 1: The article from Chemical Engineering and Processing-Process Intensification are valuable; we have read and cited them in the introduction.

Comment 2（introduction）: " Please expand on the novelty of your proposed model in comparison to other recent hybrid ML approaches beyond those cited. A critical discussion of why CNN-LSTM-LightGBM with attention is particularly suited for ISL would improve clarity."

Revised version 2: “Despite progress in applying machine learning to ISL uranium mining, significant challenges remain. A critical technical priority is optimizing injection strategies to improve uranium leaching rates and recovery rates—an effort closely tied to the overarching goal of efficient resource utilization. To address this, we propose an attention-based CNN-LSTM-LightGBM fusion model for precise prediction of uranium leaching rates in ISL operations. Building on these predictions, the model provides targeted support for optimizing uranium recovery processes. By clarifying the relationships between key parameters in the leaching process—such as daily lixiviant volume and sulfuric acid concentration—and leaching efficiency, it helps adjust core operations like injection strategies. This, in turn, indirectly enhances uranium recovery rates while meeting the current demand for efficiency in ISL resource recovery. The proposed approach offers actionable guidance for improving leaching performance and ensuring consistency with engineering expectations.”

Response 2: After revision, this paragraph highlights the model's accurate prediction of leaching rate. This capability, in turn, facilitates the adjustment of core operations (such as injection strategies) and contributes positively to enhancing uranium recovery rate—thereby underscoring the connection between the model and the uranium ore recovery process. Meanwhile, in Section 2.1.4, several paragraphs have been added to describe the advantages of the model mentioned in this paper for this task (CNNs excel at capturing local features yet struggle to model long-term dependencies, while LSTMs hold an advantage in long-term time-series analysis but lack the ability to focus on key local features. The proposed coupled architecture effectively addresses these aforementioned limitations: it integrates the spatial feature extraction capabilities of CNNs, the time-series analysis capabilities of LSTMs, and the selective focusing capability of the attention mechanism.).

Comment 3: " Discuss the practical implications of the prediction improvements: for example, how much production efficiency or environmental risk reduction could realistically be achieved in operational ISL systems."

Revised version 3（add in the conclusion）: “There is still room for expansion in the dimensionality of data indicators for prediction and the coverage of mining area samples in this study’s model, and the depth of integration between the model and professional mechanistic knowledge such as geological fluid dynamics remains insufficient. Future work should focus on integrating the model with the full-process environmental impact assessment of in-situ leaching uranium mining, quantifying the long-term effects of injection strategies on ecosystems such as groundwater, and exploring interdisciplinary integration with hydrological simulation and ecological risk assessment, so as to provide more systematic guarantees for in-situ leaching uranium mining technology in moving toward a high-quality development direction featuring low environmental impact and high resource utilization.”

Response 3: “Unfortunately, during the research process, there were no quantitative indicators for the experimental evaluation of environmental impacts, which is a limitation. Therefore, we have revised the conclusions: at the end, we added some limitations of this study as well as an outlook on future research directions, and we hope to have the opportunity to conduct research in this area in the future.”

Comment 4: " The conclusions currently restate results. They should also reflect on limitations (e.g., data availability, generalizability across different geological conditions) and outline future work directions (e.g., integration with real-time monitoring, transferability to other leaching systems). Please avoid overly broad statements such as “propel industrial applications toward more sophisticated paradigms” without concrete examples. A more specific and balanced conclusion will strengthen the manuscript. "

Revised version 4 ：“This innovation holds significant potential for promoting the transformation of in-situ leaching uranium mining technology toward a data-driven and optimization-oriented intelligent transformation.”

Response 4: “Similar to Comment 3, in future research directions, emphasis can be placed on integrating knowledge from other fields; the phrase "propel industrial applications toward more sophisticated paradigms" is indeed too vague. Therefore, it has been revised in the paper to "promoting the transformation of in-situ leaching uranium mining technology toward a data-driven and optimization-oriented intelligent transformation," which should be more appropriate.”

Comment 5:“Please clarify the description of “13 technicians’ understanding of the mining area”, this statement needs refinement.”

Revised version 5 ：“In traditional in-situ leaching (ISL) uranium mining, the injection volume depends on technicians' on-site experience.”

Response 5:“The original sentence was somewhat redundant and has now been revised.”

Comment 6:“The description of CNNs is somewhat generic. Consider shortening and focusing more directly on how CNN features are applied in this specific context (time-series features for uranium leaching prediction).”

Revised version 6 ：“CNNs excel in extracting meaningful features from ordered temporal sequences by sliding 1D filters (convolution kernels) along the time axis. This sliding operation enables element-wise multiplication and summation across consecutive time steps, generating feature maps that encode critical local patterns such as short-term fluctuations, periodic pulses, or transient trends in the time series.

By stacking multiple 1D convolutional layers, the network progressively aggregates these local temporal features into higher-level representations—for example, linking hourly leaching rate variations to daily or weekly trend patterns—thereby capturing both fine-grained dynamics and coarse-grained temporal structures [18–20]”

Response 6:“The description of CNNs has been revised to focus on describing its advantages in time-series prediction.”

Comment 7: “References: Ensure consistency in formatting (journal names should be italicized and follow Processes reference style).”

Response 7: “The citation format has been revised”

Sincerely,

Dr. Zhi-feng Liu

Sept. 16, 2025

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

After the revision process, the authors improved the academic quality of their articles and they have made the necessary corrections. Congratulations to the authors. I recommend that the article be accepted for publication.

Reviewer 2 Report

Comments and Suggestions for Authors

Accept in present form.