the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Apr 2023
Neural Network Model for Automated Prediction of Avalanche Danger Level
Vipasana Sharma
Sushil Kumar
Rama Sushil
Abstract. Snow avalanches cause danger to human lives and property worldwide in high-altitude mountainous regions. Mathematical models based on past data records can predict the danger level. In this paper, we are proposing a neural network model for predicting avalanches. The model is trained with a quality-controlled sub-dataset of Swiss Alps. Training accuracy of 79.75 % and validation accuracy of 76.54 % have been achieved. Comparative analysis of neural network and random forest models concerning metrics like precision, recall, and F1 has also been carried out.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(835 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(835 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Vipasana Sharma et al.
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-701', Anonymous Referee #1, 09 May 2023
Title: Neural Network Model for Automated Prediction of Avalanche Danger Level
Scientific Significance:
- Research work is contributing for better understanding of Snow avalanches prediction through neural network.
- Sufficient size of data set is used.
Scientific Quality:
- Valid methods used for avalanche prediction.
- Proposed Neural Network model NNM-1 is useful for other researchers to apply for other data set.
- Appropriate references have been used.
Presentation Quality:
- In paper a clear presentation is given for data used, results and conclusions.
- Well written and structured paper.
- Quality of figures/tables is good.
- Appropriate words are used technical description.
Citation: https://doi.org/10.5194/egusphere-2023-701-RC1 -
EC1: 'Reply on RC1', Orsolya Kegyes-Brassai, 09 May 2023
Thank you very much for your fast evaluation. :D
Citation: https://doi.org/10.5194/egusphere-2023-701-EC1 -
AC1: 'Reply on RC1', Vipasana Sharma, 25 May 2023
Dear RC1,
Thank you very much for your kind review and positive statements.
Kind regards,
Vipasana, on behalf of the authors
Citation: https://doi.org/10.5194/egusphere-2023-701-AC1
-
RC2: 'Comment on egusphere-2023-701', Anonymous Referee #2, 16 May 2023
Summary of the review of the manuscript
The following main points highlight the strength of the paper:
- Authors have proposed a neural networkmodel for predicting avalanches, a new approach on the available good quality of data.
- A mathematical model based on past data record is trained with a quality-controlled sub-dataset of Swiss Alps to predict the avalanches danger level.
- The model Training accuracy of 79.75% and validation accuracy of 76.54% have been achieved, which is quite significant.
Introduce the problem clearly and well written in structured manner:
- Accurate prediction of snow avalanches can help ensure people's safety in snow-covered regions.
- The major hurdle in developing machine learning models is the lack of sufficient and reliable data. This issue has been resolvedto a great extent by the WSL Institute of Snow and Avalanche Research, Switzerland, by collecting 20 years of data inavalanche forecasting.
- The data set has been further refined with data selection techniques.
- The dataset combines differentfeature sets with meteorological variables.
- This unique dataset has enabled experimentation with machine learning models like neural networks and compared its performance with the random forest machine learning technique.
Analyses of the data have been done for prediction purpose:
The dataset used for the training ofneural networks is described well. Authors explained the neural network model, tuning of hyper parameters and evaluation metrics. Random Forest machine learning method details applied to the same dataset are described.
Relevant papers are cited. Following two more references may be added in the paper which are using different techniques for snow avalanche.
- Amreek Singh, AshwagoshaGanju, 2008. Artificial Neural Networks for Snow Avalanche Forecasting inIndian Himalaya. “The 12th International Conference ofInternational Association for Computer Methods and Advances in Geomechanics (IACMAG)1-6 October, 2008Goa, India.
- Singh, A., Srinivasan, K. and Ganju, A. 2005. Avalanche Forecast Using Numerical Weather Prediction in Indian Himalaya,Cold Regions Science and Technology, Vol. 43, 83-92.
On the basis of mentioned facts and importance of the topic I recommend this work for publication.
-
EC2: 'Reply on RC2', Orsolya Kegyes-Brassai, 19 May 2023
Thank you for the review. I'll ask the author to take your recommendation into account and consider the suggested references.
Citation: https://doi.org/10.5194/egusphere-2023-701-EC2 -
AC2: 'Reply on RC2', Vipasana Sharma, 25 May 2023
Dear RC2,
Thank you so much for the review and suggestions regarding new references. We have included the suggested references in the updated manuscript.
Kind regards,
Vipasana, on behalf of the authors
Citation: https://doi.org/10.5194/egusphere-2023-701-AC2
-
RC3: 'Comment on egusphere-2023-701', Anonymous Referee #3, 16 May 2023
The authors have presented a work entitled:Neural Network Model for Automated Prediction of Avalanche Danger Level. It is an interesting and is very much relevent to scenarios whre avalanches are a common phenomena.
The paper is well written and well explained and shows an interesting application of AI. The methodology is clearly explaned and the numrical results justify the potential of the present work.
Some minor suggestions to the authors are to add a few more recent references relevant to the present work. It will also be good f the authors can write a word or two about the complexity of data used in the present work. They can mention if the data acquired needed some pre-processing and if the answer is es than what type of pre-rocessing was used.
Citation: https://doi.org/10.5194/egusphere-2023-701-RC3 -
EC3: 'Reply on RC3', Orsolya Kegyes-Brassai, 19 May 2023
Thank you for the review. I'll ask the author to take your recommendation into account and to add explanation about the data, and to consider updating the references.
Citation: https://doi.org/10.5194/egusphere-2023-701-EC3 -
AC3: 'Reply on RC3', Vipasana Sharma, 25 May 2023
Dear RC3,
Thank you so much for the review and suggestions. Two recent references on use of deep learning techniques for analysing snow avalanches have been included.
Some pre-processing techniques used to clean the data are:
1.Handling the missing values:
In order to make sure that the results are not hampered due to missing values in the data set we replaced all the blanks with NAN using the python code df = df. fillna(0) in the dataset.
2.Removing the Duplicates:
Duplicates in the data can skew analysis results and introduce biases. Identifying and removing duplicate records was done manually in order to preserve the data integrity.
3.Data normalization:
Normalization is the process of scaling numeric data to a standard range or distribution. The dataset's variables whose values lie in different ranges do not have an equal contribution to the model's fit parameters and training function and may even lead to bias in the predictions made with that model. Hence, we have used the StandardScaler () function to standardize the data values into a standard format which is offered by the Python sklearn library
4.Data type conversion:
The categorical variable used here "Danger levels" was changed to numerical representation of one-hot encoding using the OneHotEncoder () function of the Python sklearn Library.
5.Removing irrelevant features:
Some irrelevant columns with information of date, sector ID, name of the sector region, elevation_width, elevaton_station and warning were removed from the data set.
Above mentioned pre-processing techniques used are now referred to in the manuscript in Sec-3 (Dataset) in the updated manuscript.
Citation: https://doi.org/10.5194/egusphere-2023-701-AC3 -
EC4: 'Reply on AC3', Orsolya Kegyes-Brassai, 31 May 2023
Dear Authors,
Thanl you for ypour response and also that you have concidered reviewers suggestions.
Can you please upload the final, corrected version?
Thanks,
Editor
Citation: https://doi.org/10.5194/egusphere-2023-701-EC4
-
EC4: 'Reply on AC3', Orsolya Kegyes-Brassai, 31 May 2023
-
EC3: 'Reply on RC3', Orsolya Kegyes-Brassai, 19 May 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-701', Anonymous Referee #1, 09 May 2023
Title: Neural Network Model for Automated Prediction of Avalanche Danger Level
Scientific Significance:
- Research work is contributing for better understanding of Snow avalanches prediction through neural network.
- Sufficient size of data set is used.
Scientific Quality:
- Valid methods used for avalanche prediction.
- Proposed Neural Network model NNM-1 is useful for other researchers to apply for other data set.
- Appropriate references have been used.
Presentation Quality:
- In paper a clear presentation is given for data used, results and conclusions.
- Well written and structured paper.
- Quality of figures/tables is good.
- Appropriate words are used technical description.
Citation: https://doi.org/10.5194/egusphere-2023-701-RC1 -
EC1: 'Reply on RC1', Orsolya Kegyes-Brassai, 09 May 2023
Thank you very much for your fast evaluation. :D
Citation: https://doi.org/10.5194/egusphere-2023-701-EC1 -
AC1: 'Reply on RC1', Vipasana Sharma, 25 May 2023
Dear RC1,
Thank you very much for your kind review and positive statements.
Kind regards,
Vipasana, on behalf of the authors
Citation: https://doi.org/10.5194/egusphere-2023-701-AC1
-
RC2: 'Comment on egusphere-2023-701', Anonymous Referee #2, 16 May 2023
Summary of the review of the manuscript
The following main points highlight the strength of the paper:
- Authors have proposed a neural networkmodel for predicting avalanches, a new approach on the available good quality of data.
- A mathematical model based on past data record is trained with a quality-controlled sub-dataset of Swiss Alps to predict the avalanches danger level.
- The model Training accuracy of 79.75% and validation accuracy of 76.54% have been achieved, which is quite significant.
Introduce the problem clearly and well written in structured manner:
- Accurate prediction of snow avalanches can help ensure people's safety in snow-covered regions.
- The major hurdle in developing machine learning models is the lack of sufficient and reliable data. This issue has been resolvedto a great extent by the WSL Institute of Snow and Avalanche Research, Switzerland, by collecting 20 years of data inavalanche forecasting.
- The data set has been further refined with data selection techniques.
- The dataset combines differentfeature sets with meteorological variables.
- This unique dataset has enabled experimentation with machine learning models like neural networks and compared its performance with the random forest machine learning technique.
Analyses of the data have been done for prediction purpose:
The dataset used for the training ofneural networks is described well. Authors explained the neural network model, tuning of hyper parameters and evaluation metrics. Random Forest machine learning method details applied to the same dataset are described.
Relevant papers are cited. Following two more references may be added in the paper which are using different techniques for snow avalanche.
- Amreek Singh, AshwagoshaGanju, 2008. Artificial Neural Networks for Snow Avalanche Forecasting inIndian Himalaya. “The 12th International Conference ofInternational Association for Computer Methods and Advances in Geomechanics (IACMAG)1-6 October, 2008Goa, India.
- Singh, A., Srinivasan, K. and Ganju, A. 2005. Avalanche Forecast Using Numerical Weather Prediction in Indian Himalaya,Cold Regions Science and Technology, Vol. 43, 83-92.
On the basis of mentioned facts and importance of the topic I recommend this work for publication.
-
EC2: 'Reply on RC2', Orsolya Kegyes-Brassai, 19 May 2023
Thank you for the review. I'll ask the author to take your recommendation into account and consider the suggested references.
Citation: https://doi.org/10.5194/egusphere-2023-701-EC2 -
AC2: 'Reply on RC2', Vipasana Sharma, 25 May 2023
Dear RC2,
Thank you so much for the review and suggestions regarding new references. We have included the suggested references in the updated manuscript.
Kind regards,
Vipasana, on behalf of the authors
Citation: https://doi.org/10.5194/egusphere-2023-701-AC2
-
RC3: 'Comment on egusphere-2023-701', Anonymous Referee #3, 16 May 2023
The authors have presented a work entitled:Neural Network Model for Automated Prediction of Avalanche Danger Level. It is an interesting and is very much relevent to scenarios whre avalanches are a common phenomena.
The paper is well written and well explained and shows an interesting application of AI. The methodology is clearly explaned and the numrical results justify the potential of the present work.
Some minor suggestions to the authors are to add a few more recent references relevant to the present work. It will also be good f the authors can write a word or two about the complexity of data used in the present work. They can mention if the data acquired needed some pre-processing and if the answer is es than what type of pre-rocessing was used.
Citation: https://doi.org/10.5194/egusphere-2023-701-RC3 -
EC3: 'Reply on RC3', Orsolya Kegyes-Brassai, 19 May 2023
Thank you for the review. I'll ask the author to take your recommendation into account and to add explanation about the data, and to consider updating the references.
Citation: https://doi.org/10.5194/egusphere-2023-701-EC3 -
AC3: 'Reply on RC3', Vipasana Sharma, 25 May 2023
Dear RC3,
Thank you so much for the review and suggestions. Two recent references on use of deep learning techniques for analysing snow avalanches have been included.
Some pre-processing techniques used to clean the data are:
1.Handling the missing values:
In order to make sure that the results are not hampered due to missing values in the data set we replaced all the blanks with NAN using the python code df = df. fillna(0) in the dataset.
2.Removing the Duplicates:
Duplicates in the data can skew analysis results and introduce biases. Identifying and removing duplicate records was done manually in order to preserve the data integrity.
3.Data normalization:
Normalization is the process of scaling numeric data to a standard range or distribution. The dataset's variables whose values lie in different ranges do not have an equal contribution to the model's fit parameters and training function and may even lead to bias in the predictions made with that model. Hence, we have used the StandardScaler () function to standardize the data values into a standard format which is offered by the Python sklearn library
4.Data type conversion:
The categorical variable used here "Danger levels" was changed to numerical representation of one-hot encoding using the OneHotEncoder () function of the Python sklearn Library.
5.Removing irrelevant features:
Some irrelevant columns with information of date, sector ID, name of the sector region, elevation_width, elevaton_station and warning were removed from the data set.
Above mentioned pre-processing techniques used are now referred to in the manuscript in Sec-3 (Dataset) in the updated manuscript.
Citation: https://doi.org/10.5194/egusphere-2023-701-AC3 -
EC4: 'Reply on AC3', Orsolya Kegyes-Brassai, 31 May 2023
Dear Authors,
Thanl you for ypour response and also that you have concidered reviewers suggestions.
Can you please upload the final, corrected version?
Thanks,
Editor
Citation: https://doi.org/10.5194/egusphere-2023-701-EC4
-
EC4: 'Reply on AC3', Orsolya Kegyes-Brassai, 31 May 2023
-
EC3: 'Reply on RC3', Orsolya Kegyes-Brassai, 19 May 2023
Peer review completion
Journal article(s) based on this preprint
Vipasana Sharma et al.
Vipasana Sharma et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 238 | 151 | 29 | 418 | 5 | 4 |
- HTML: 238
- PDF: 151
- XML: 29
- Total: 418
- BibTeX: 5
- EndNote: 4
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(835 KB) - Metadata XML