Ensemble Learning for Predicting TOC from Well-Logs of the Unconventional Goldwyer Shale
Round 1
Reviewer 1 Report
This study provides an efficient TOC estimation workflow from wire-line log responses through a robust ensemble learning approach. It is good data interpretation and good addition to the existing knowledge , since information in this field has been insufficient.
Language is written by native speaker so no errors. Good work.
Author Response
Dear Reviewer,
We gratefully thank you for a careful and constructive perusal of our manuscript. A thorough grammar and spell check were done. The changes in the manuscript are highlighted by track change.
Reviewer 2 Report
Line 43 add newer citation - https://doi.org/10.3390/en14112995
Author Response
Dear Reviewer,
We gratefully thank you for a careful and constructive perusal of our manuscript. The changes in the manuscript are highlighted by track change.
The manuscript has been checked by a native English-speaking colleague.
The requested citation has been added. (Line 73)
Reviewer 3 Report
Congrats to the authors! for an interesting paper. The results of the study are encouraging but I wonder if the workflow proposed can be applied in any geological environment or at least to any shale-gas reservoir.
Regarding Technical aspects:
In the paper the authors are trying to modelling via machine learning the prediction of the TOC values from well-logs in the Goldwyer shale formation (Lower Paleozoic) of the Canning Basin, WA. The Canning Basin is a relatively underexplored sedimentary basin although the Goldwyer shale formation (Ordovician) shows good petroleum prospectivity related to these organic-rich mudstones.
The study was focused on the development of a workflow for generating a robust TOC prediction model using the ensemble learning approach and as a result, the potential of the Goldwyer shale can be assessed. The workflow consists of two stages: data preparation; and (ii) model generation by ensemble learning and TOC prediction. The ensemble learning is further sub-divided into four stages. The standard multi-linear regression (MLR) statistical method is also considered for modelling a linear relationship between the log response and the core TOC.
The accurate TOC assessment is essential for assessing the prospectivity of a shale gas play and further for the development and production of the shale gas reservoir. In this line the study provides an efficient TOC estimation workflow from wire-line log responses through a robust ensemble learning approach that can integrate multiple models to build a predictive TOC model and improve the prediction performance of TOC.
The final prediction TOC model is the average predictor model which is aggregate of four ensemble learning models: Multi-layer Perceptron (MLP) an ANN algorithm; bootstrapping method – RF; projection onto a higher dimension space with different kernel – SVM; and Gradient boosting regressor (GBR). With a proper calibration I think the model proposed in the paper can be applied in any geological environment.
The results presented lead to the conclusion that the paper proposed for review has a high scientific merit and a significant contribution to the academic debate.
Regarding Ease of Understanding:
The abstract of the paper is informative and concise.
The structure of the paper is continuous and logical.
The methodology and techniques used are proper described.
The pictures and tables are necessary and appropriate.
The results are promising.
The interpretations and conclusions are supported by the data presented.
The most of the references are recent.
The quality and the merit of the paper are indisputable. However the paper is a long one (35 pages) and the reader is lost in details. I recommend if is possible restructuring of the paper in the way to become more concise, short and crisp.
Considering the above I recommend the MS proposed for review (Energies-1483489) Publishable with Minor Revisions.
Author Response
Dear Reviewer,
We gratefully thank you for a careful and constructive perusal of our manuscript. We address your comments and suggestions below. The changes in the manuscript are highlighted by track change.
- The ensemble learning approach has been applied in Goldwyer gas shale formation in the onshore Canning Basin, Western Australia. Considering its success, the same methodology can be applied in other gas shale reservoirs as well. However, we have not checked its validity beyond Goldwyer gas shale.
-
Following your suggestion, we have moved the technical details of ensemble learning algorithms to the Appendix to make the manuscript more concise and crispier. We are sure that the editorial team will adjust the formatting and make it a standard double-column format.
