Reviewer has chosen to be Anonymous

Summary of paper in a few sentences:

This paper focuses on the forecasting of numerous meteorological variables for Bangladesh using various machine learning algorithms. Data was acquired from 33 weather stations across Bangladesh for 2012—2018. The study claims to have developed skillful forecasts for a year of weather variables in Bangladesh (line 3, page 4), but fails to provide methodological details to support this claim. The paper as it stands is too long and provides excessive background information on concepts that are not relevant, yet fails to provide details and justifications where vitally needed to understand the study and its technical quality. Incorrect grammar and weather/climate terminology also make it difficult to understand the text.

Reasons to accept:

N/A

Reasons to reject:

1. Insufficient and inappropriate research background: The paper’s background listed a number of past studies and what they did/found, yet some of these studies are irrelevant to the study being conducted. For example, studies regarding prediction of rice cultivation time, road traffic, and resource utilization in cloud computing are not relevant to the study at hand (page 3), which is related to forecasting weather variables for Bangladesh. A stronger research background would have detailed the state of the science with regards to forecasting of the Bangladesh weather parameters being analyzed and data being used in this study (e.g., With how much lead time can these parameters be skillfully forecast in Bangladesh, currently? How accurate are the weather measurements provided by BMD?). These details are critical to establish the novelty and impact factor of this research.

2. Insufficient and inappropriate methodology: The paper describes error metrics and machine learning algorithms, but does not detail how they were used in this study. For example, what do X and Y specifically represent in MAE, MSE, and MAPE (pages 4-5)? Is X the weather variable training data and Y the test data? Are data being standardized/normalized prior to being input into the machine learning algorithms? What percentage of data is being used for training and testing (lines 1-7, page 9)? What specific location is being forecasted for (latitude, longitude)? The omission of these details makes it difficult to understand and reproduce the study. It is also difficult to assess whether these metrics and algorithms were properly applied given the lack of specific information regarding their application in this study.

3. Results section: In general, it was not clear to me why certain machine learning algorithms performed better than others throughout the text, nor could I interpret the results myself given that I was unclear about the study methodology. Additionally, details regarding wind speed, humidity, max and min temperatures, etc. should not be in the results. These should be in the methods and considerably consolidated. I also have concerns about the calculation for maximum and minimum temperature. The authors describe maximum temperature as the “yearly summation of high temperature” (line 45, page 13), which I do not understand. The authors also state that maximum and minimum temperatures were determined between 7pm and 7am the next day, but they should be computed within a 24-hour period.

4. Grammatical and word-choice selections throughout the paper: Critically, this paper contains word usage that is at times inappropriate. For example, phrases like “tillage stuff” (line 41, page 1), “wind stream speed” (line 27, page 10), “intertropical union zone” (line 31, page 12; should be intertropical convergence zone), “rainstorm trough” (line 31, page 12), “rain creation is dampness” (line 23, page 12), “most extreme dampness” (line 31, page 13), “much fruitful” (line 36, page 20), to name just a few, are either grammatically incorrect or incorrect terms in meteorology/climate science. Some sentences were very difficult to understand throughout the text. For example, lines 22-23, page 13 states “The measure of water vapor expected to accomplish immersion increments as the temperature increments”, which I do not understand. Correcting these issues requires a major rewrite of the paper.

Nanopublication comments:

Further comments:

Reviewer has chosen to be Anonymous

Summary of paper in a few sentences:

In this manuscript, authors have presented several machine learning techniques for weather forecasting in Bangladesh.

Reasons to accept:

Reject

Reasons to reject:

In this manuscript, authors have presented several machine learning techniques for weather forecasting in Bangladesh.
The manuscript is not suitable for publication. Only six years (2012-2017) years of weather data have been used to train the models. Six years of data is not sufficient to capture decadal and multi-decadal scale of climate variability. It’s not clear why long record of weather data has not been considered. It’s not clear but it appears that authors have assumed the past values of an environmental variable to the predictor of that environmental variable; i.e., past values of rainfall have been used to forecast future precipitation value. If this is true, then it is a very weak assumption. Wind speeds, humidity, temperature, and many other environmental variables could be predictors of rainfall.

Furthermore, there is no novelty presented in the study. Authors have simply taken some weather data and applied it to standard machine learning methods to do weather forecast as authors claim. This is not weather forecasting.

Writing is verbose. Lots of unnecessary details have been written in Introduction and Methodology. Details about the performance metrics and the machine learning method are available in several Machine Learning textbook. Citations to some of the known books or literatures would be sufficient. The usage of standard grammar rule is not followed, especially the use of ‘which’ and ‘that’; which requires a common unlike that.

The manuscript should not be considered for publication.

Nanopublication comments:

Further comments:

By Ryan Lagerquist https://orcid.org/0000-0002-8409-415X

Review Details

---

Reviewer has chosen not to be Anonymous

Summary of paper in a few sentences:

The authors use a variety of machine learning (ML) methods to predict basic weather variables -- temperature, humidity, wind speed, and rainfall -- in Bangladesh. They compare the ML methods to determine which performs best for each weather variable.

Reasons to accept:

They compare a large number of ML methods, so their results could serve as a valuable guide for those who want to implement similar methods for weather prediction.

Reasons to reject:

The paper is poorly written. Specific complaints are listed below.

[1] The quality of the English is poor. This makes the paper difficult to follow and makes some parts completely incomprehensible. Some examples from just the first two pages are as follows.

"Nearly 70% of its total population live in rural areas and 60% of them earn their livelihood from tillage stuff." -- What is 'tillage stuff'?

"Weather forecasting is supposed to be a prime factor for Bangladesh’s economy as agriculture plays a vital role in country’s overall Gross Domestic Product (GDP) which accounts for approximately 20% of the total amount. . .Even a significant amount of total annual exports are from farming products which tend to be in the region of 13-18% of the country's total GDP." -- The first sentence seems to claim that 20% of the GDP comes from agriculture, and the second seems to claim that 13-18% of the GDP comes from agriculture -- unless there's a difference between agriculture and 'farming products,' but this is unclear.

"Since the dawn of technological advancement, weather forecasting is that much noteworthy which have been endeavored to predict accurately as much as possible for many years by the experts." -- The second part of this sentence (after the comma) doesn't make sense.

"Intelligent weather prediction techniques can help us to a certain degree that can help us to make effective decisions which can save valuable lives, times and property at a time. With the passage of time, science and technology have advanced to the next level and weather pattern discovery has attracted more attention. It involves the anticipation of how the current circumstance with the air will change in which current climate conditions are taken via ground discernments e.g. boats, radar, satellite, airplanes, etc." -- Although this part is comprehensible, like much of the paper, it is long-winded and full of grammatical errors, which makes it difficult to parse.

[2] The paper contains 32 figures, which is far too many. I understand that the authors want to show everything, but the paper would be much more effective if they showed only the most interesting/important results. They could relegate other results to an online supplement.

[3] Most of the figures are ineffective. Figures 1-17 are too grainy; Figures 18, 21, and 30 have three titles that are too small to read; and lines in Figures 19, 23, 25-26, 28-29, and 31-32 cannot be differentiated.

[4] The paper contains 10 tables, each with 27 numbers. These tables would be better served in a graphical format such as line graphs or bar graphs, as it is difficult to make general conclusions from staring at 27 numbers (especially when one has to do this 10 times).

[5] In general, the paper contains too many figures/tables and too little interpretation of the data shown in the figures/tables.

Nanopublication comments:

Further comments:

Poor presentation and grammar make the paper difficult to read, so I read only the first three pages thoroughly, skimming thereafter. This is why the majority of my comments focus on the presentation and not the science. However, I believe these comments are sufficient to reject. Even if the science were flawless, I can't imagine the paper having a positive impact.