Sunday, November 21^st, 2021

Ref.: NHESS 2021_294 titled: "A Strategic Framework for Natural Disaster-Induced Cost Risk Analysis and Mitigation: A Two-Stage Approach Using Deep Learning and Cost-Benefit Analysis"

Dear authors,

I read with interest your manuscript titled: " A Strategic Framework for Natural Disaster-Induced Cost Risk Analysis and Mitigation: A Two-Stage Approach Using Deep Learning and Cost-Benefit Analysis". The manuscript develops a strategic evaluation framework for natural disaster risk mitigation strategies. A two-stage model: using a deep learning and cost-benefit analysis is introduced. Data of insured loss amounts due to natural disasters associated with the risk indicators were collected and used to train and develop deep learning model. A data set of 458 cases was used to train the model and test its validity. Root Mean Square Error was used to assess the Deep Neural Network (DNN) precision of the prediction model and was compared to Multi Regression Analysis (MRA). The DNN was found to be precise compared to MRA with number of hidden layer – 3, and 25 hidden nodes at each layer. 474 disaster risk reservoirs and 291 maintenance projects were used as validation case for the BCR analysis stage. Loss rate was found to significantly decrease (82%). Despite the fact that rainfall during the test cases decreased by an average of 35% the model yielded significant improvement in disaster risk mitigation.

The paper deserves minor revisions as follows:

1. The title of the manuscript is too long, please consider shorter title such as: "Natural Disaster- Mitigation Using Deep Learning and Cost-Benefit Analysis".
2. The abstract does not reflect the novelty of the methodology. Please consider revision: stress the validity of findings and efficiency of the framework.
3. The literature review may be extended to applications of Deep Learning to risk assessment – Please review and consider the following literature: ( (Khosravi et al. 2020; Zhang et al. 2022; Yi et al. 2020; Al Najar et al. 2021; Moishin et al. 2021; Shane Crawford et al. 2020; Sugiyarto and Rasjava 2020; Kim et al. 2021));
4. Please add a research framework diagram, emphasize the core phases of the methodology;
5. The training phase of the DNN must be significantly improved: please provide figures with distributions of the core variables, Please provide distribution figures of the training data (Loss ratio, building type, maximum wind speed, rainfall and PGA), the MRA model and the DNN output. This is essential to for legibility and scientific soundness.
6. Please add keyworks.
7. Please see further remarks and corrections in the attached file.

Good Luck!

Ref.

Gradual increases in the frequency and severity of natural disasters increase loss risk of human life and property. It is important to estimate the damages caused by the disaster before such events occurs. The authors attempted to to develop a strategic evaluation framework  and analyzed the natural disaster risk mitigation strategies by DNN methods with data of insured loss amounts between 2009 to 2019.  The ideas is Scientific and topic is suitable for NHESS.

After reading the manuscript carefully, I found some important messages were missing or not clearly stated. I listed them below for improving the presentation:

1. The quality of the data determines whether the analysis results are scientific. The authors carry out the research with data on claim payout for storm and flood damage insurance from from the Korea Insurance Development Institute. But How many pieces? What information in the data?  The author only gave the data name and resources. It is necessary to describe the data information in detail.
2. The data collected are claim payout for storm and flood damage. But the selected disaster factors include wind speed and Peak Ground Acceleration. What is the reason? What is the relationship between them?
3. It is an interesting attempt to use DNN to predict disaster loss. The authors just trained the model and compared it with MRA model. I think it is import to analyze it in detail and use it in resent disaster cases.
4. The authors describe two stages in their work. Stage I is to create a predictive model based on a deep learning algorithm with data on claim payout for storm and flood damage insurance.  Stage II is to analyzed the cost-benefit with data on natural disaster risk reduction projects. There seems to be no logical relationship between the two. What is the relationship between the two stages? The authors should describe it clearly.
5.  Sections about the methods and results should be revised. The authors mixes methods and results together in Section 3 and Section 4, which are nor clearly presented.