Time-Series Analysis for Forecasting Climate Parameters of Kashmir Valley Using ARIMA and Seasonal ARIMA Model.

Tracking #: 742-1722

Authors:

	Name	ORCID
	Sheikh Umar	https://orcid.org/0000-0002-0382-4035
	Fahad Farooq Jamie	https://orcid.org/0000-0002-7566-0076
	Khalid Hussain	https://orcid.org/0000-0003-4052-0468
	Salim Mir	https://orcid.org/0000-0002-8103-2017

Responsible editor:

Thomas Chadefaux

Submission Type:

Research Paper

Abstract:

The valley of Kashmir being an extremely ecologically sensitive area deserves significant attention with regard to climate change. As the region is part of the Indian Himalayan Region (IHR) and home to multiple rivers and glaciers, it holds significant geopolitical, economic, as well as geographical importance. Climate change in the region can have ripple effects on the Indian subcontinent in general. As such effective prediction of the different climate variables of the region is of paramount importance and can help us to reverse or slow down the negative climate change process. In this paper, we investigate CRU TS4.04 time series data of 120 years (1901-2019) for three major climate variables - precipitation, temperature, and cloud cover observed in the region. The analysis revealed some key and concerning trends. Thereafter, forecasting of the climate parameters for the next 80 years (2020-2100) was performed using the ARIMA and S-ARIMA models after a thorough analysis of the data and proper hard tuning of the model hyper-parameters, separately for each of the three climate parameters. The resulting projections show drastic temperature changes, with a projected increase of about 2°C by the end of the century, slight changes in precipitation and cloud cover, and other alarming climatic conditions. This study also brings forth the inability of the ARIMA model to substantially forecast erratic changes in any climate variable. The insights gathered in this study may serve as a presage for the concerned government and stakeholders, and will pave way for the development of robust and efficient plans to tackle climate change in the area. This is the first kind of work that is using effective machine learning-based time-series forecasting models trained over such a large data set for all three major climate variables.

Manuscript:

ds-paper-742.pdf

Data repository URLs:

https://github.com/fahad-farooq/climchange_jk

Date of Submission:

Wednesday, January 11, 2023

Date of Decision:

Tuesday, April 18, 2023

Nanopublication URLs:

Decision:

Reject

Solicited Reviews:

Review #1 submitted on 23/Jan/2023

By Thomas SCHINCARIOL ORCID logo

https://orcid.org/0000-0001-8807-3367

Review Details

Reviewer has chosen not to be Anonymous

Overall Impression: Weak
Suggested Decision: Reject
Technical Quality of the paper: Weak
Presentation: Weak
Reviewer`s confidence: Medium
Significance: High significance
Background: Reasonable
Novelty: Lack of novelty
Data availability: All used and produced data (if any) are FAIR and openly available in established data repositories
Length of the manuscript: The length of this manuscript is about right

Summary of paper in a few sentences:

This paper has a twofold objective : the first one is to analyze the past dynamic of three climatic variables (temperature, cloud cover and precipitation). The authors show an increasing value in the temperature trend in the last 120 years. In a second part, the three climate variables are forecasted in the 80 next years (2020-2100). The authors show a predicted increased value in temperature of 2 °C, and slight changes in precipitation and cloud cover. To perform these predictions, the authors used a Seasonal ARIMA model.

Reasons to accept:

Paper with some interesting results. The method and model are well-known and already applied in climate change analysis/prediction. The novelty is the application to this specific region. It also adds a new argument to the increase of temperature in the next couple of decades and comfort the actual predictions made in other part of the world.
The results with temperatures (mean, minimum and maximum) are good, and the model shows a good fit and great prediction for the validation set.

Reasons to reject:

General reasons :

First, the novelty here is debatable. The only new thing is the application to this specific region. The model is widely used in the existing literature.
Then, as said before, results with temperatures are good. However, on Fig 16-18, we can spot a high seasonality. It seems that the model is predicting the same pattern and does not catch the little annual variation and trend. The analysis and prediction of the trends (as shown in Fig 2-6), monthly or split in season, seems to be a better option. Like this, the focus would be on the trend (the real interest of the study) and less on fitting the monthly seasonality, that we already know and assume.
For precipitation and cloud cover, the results are mixed. As the rainfall are extremely fluctuant, it is way harder to fit it with an (S)ARIMA model. However, as said in the literature review, the main problem with rainfall is heavy rains and drought. In the region, it already provoked catastrophes and might be one of the major issues caused by climate change in the next decades. The problem could then be adapted by focusing on the extreme events related to rainfall. For example, a study could be run counting the number of months/weeks exceeding a dangerous threshold. ARIMA model could fit and then forecast this metrics. Like this, we would be able to have a better idea of the extreme rainfall events coming with climate change.

More specific remarks :

Fig 1, page 2 : Good overall figure but maybe on the left part, print the overall shape of India for the readers to have a more overall idea of the localization of study zone.
Fig 2 to 5, page 7-8 : Combining the figure would let the reader comparing the four seasons’ trends with putting the point in transparency (alpha=0.3 for example). Same for the regression tables 1-3 and figure 6, combining them would make the presentation clearer, with subtitles.
Fig 7 and 8 : Scatter plot would be more appropriate than line. Regression line could be eventually added.
Model optimization : ACF and PACF are good to determine the ARIMA and SARIMA parameter. However, it may just be a starting point for the parameter tuning. It is not clear if you test other set of parameters with value around the ones extracted from ACF and PACF curves analyze. Maybe it would be a good idea to try with +/- 1 value on each parameter to see if it better the evaluation metric (AIC or BIC). For example, if the best MA value is 3 based on the ACF curve, try 2 and 4 to see if it gets a better AIC and try then with the new best value, and so on.

Nanopublication comments:

Further comments:

To conclude, this paper provides a new argument in the temperature change in the next couple of years of Kashmir Valley. However, some things should be changed (model tuning, metric used for optimization, the study around rainfall) to make it even stronger and a solid paper on the future issues we might face with climate change.

Review #2 submitted on 14/Apr/2023

By Tobias Kuhn ORCID logo

https://orcid.org/0000-0002-1267-0234

Review Details

Reviewer has chosen not to be Anonymous

Overall Impression: Weak
Suggested Decision: Reject
Technical Quality of the paper: Weak
Presentation: Good
Reviewer`s confidence: Medium
Significance: High significance
Background: Incomplete or inappropriate
Novelty: Lack of novelty
Data availability: All used and produced data (if any) are FAIR and openly available in established data repositories
Length of the manuscript: The length of this manuscript is about right

Summary of paper in a few sentences:

The paper presents a study to analyze the past climate of the region of the Kashmir Valley, and presents a prediction of its future.

Reasons to accept:

- very relevant topic
- well written

Reasons to reject:

- analysis of existing data with existing techniques is not novel enough
- prediction of future climate is not convincing

Nanopublication comments:

Further comments:

The paper is generally well written, in particular in the introduction of the specific problem and region, and it tackles an interesting and highly relevant problem. However, the paper has some crucial weaknesses.

The related work section is mostly helpful but sometimes also uses exaggerated language, e.g. "exceptionally accurate results". The related work is weak on the multi-decade prediction aspect, however. The authors note that none of the existing studies aimed to "forecasting the future climate trends", but do not discuss why this is the case (most obviously because it is incredibly hard to do so in a convincing way).

The shown analyses seems technically correct, even though I cannot judge the details. The analysis of the past data is certainly interesting, even though often done in too much detail to be really insightful, but the novelty of these results is in any case not sufficiently established. Descriptively analyzing public data doesn't meet the standards for novelty for this journal.

The future predictions form the more innovate and more interesting aspect, but this part is unfortunately not convincing, and could even be misleading to lay audiences. Making multi-decade predictions of the climate of a region by only extrapolating (or "learning") from the past data of the given region seems naive. We are most probably not in some sort of slowly shifting equilibrium, but in a highly complex global system with wide-ranging interactions and possibly major feedback loops. At the very least the *global* climate would have to be taken into account for predicting the climate of the given region. To make it actually convincing, it would have to involve also variables that reflect several possible trajectories of important variables like global CO2 emission, technological progress, and policies that cannot be inferred from past data. The paper also doesn't really highlight these limitations in the discussion. For these reasons, the interpretations we can draw from the results are very limited.

For these reasons, I suggest to reject the paper, as I think part of it can be more misleading than enlightening.

2 Comments

Review the paper and comment.

Submitted by Malik Jawarneh on Tue, 03/07/2023 - 05:30

Structure your abstract as follows- 1) Background 2) Aim/Objective 3) Methodology 4) Results 5) Conclusion. Write 2-4 lines for each and merge everything in one paragraph without any subheading
Abstract must contain the motivation and objective of the article. The Abstract must be very clear and the motive of the paper should be represented in a nutshell.
Introduction should be of 5-7 solid paragraphs and provide structure of work at the end of the Introduction section.
Add more contribution to your study field.
The purpose of study not clear.
Make highlight for objectives
Remove any table or figure which is taken from web. Otherwise you have to get approval from publisher and author in a provided form by springer.
Please avoid to write definitions of terms ARIMA; climate; Kashmir valley; analysis; time-series forecasting;etc., which are already available over web, try to cite work for such information.
In summary, only provide useful content in your work.
These are Title related to your area, you may use.
Add your references section
Please list all abbreviations used in your manuscript under the heading "Abbreviations" after the conclusions section
The authors have not discussed any limitations or challenges faced during the development of the dataset or the machine learning system. There is also no comparison to other existing datasets and systems.

JAWARNEH, M. (2023). Development of Machine Learning Based Security Model for IoT Network.
Nageswaran, S., Arunkumar, G., Bisht, A. K., Mewada, S., Kumar, J. N. V. R., Jawarneh, M., & Asenso, E. (2022). Lung cancer classification and prediction using machine learning and image processing. BioMed Research International, 2022.
Ansari, A. S., Jawarneh, M., Ritonga, M., Jamwal, P., Mohammadi, M. S., Veluri, R. K., ... & Shah, M. A. (2022). Improved Support Vector Machine and Image Processing Enabled Methodology for Detection and Classification of Grape Leaf Disease. Journal of Food Quality, 2022.
Pallathadka, H., Jawarneh, M., Sammy, F., Garchar, V., Sanchez, T., & Naved, M. (2022, April). A Review of Using Artificial Intelligence and Machine Learning in Food and Agriculture Industry. In 2022 2nd International Conference on Advance Computing and Innovative Technologies in Engineering (ICACITE) (pp. 2215-2218). IEEE.

Meta-Review by Editor

Submitted by Tobias Kuhn on Tue, 04/18/2023 - 05:32

Below is a summary of the main issues raised by the reviewers:

Lack of novelty: Both reviewers agreed that the novelty of the paper is limited, with the only new aspect being the application to a specific region. The method and model used are well-known and widely used in the existing literature.
Weaknesses in the analysis: While the analysis of past data is interesting, the novelty of these results is not sufficiently established. The future predictions, which form the more innovative and interesting aspect of the paper, are not convincing due to their reliance on extrapolation from past data without taking into account global climate or other important variables. The interpretation of the results is therefore limited.
Presentation and clarity: The reviewers suggested improvements to the presentation of figures and tables to enhance clarity and facilitate comparison of results.
Model optimization: One reviewer suggested that model optimization could be improved by testing different parameter sets and better evaluation metrics.

While the paper presents some interesting results and adds to the current understanding of temperature increases in the coming decades, the issues highlighted above prevent us from accepting this submission. We appreciate your interest in our journal and hope that the feedback provided will be helpful for any future submissions.

Once again, we apologize for the delay in our decision, and we thank you for considering our journal for your work.

Thomas Chadefaux (https://orcid.org/0000-0002-8456-8124)

Data Science