International Journal of Computer Applications (0975 – 8887)

Volume 184– No.16, June 2022

Heat Wave Prediction using Machine Learning

Techniques: A Review

F.S. Sourjah W.P.J. Pemarathne

Department of Computer Science, Department of Computer Science,
Faculty of Applied Sciences, Faculty of Computing,
University of Sri Jayewardenepura General Sir John Kotelawala Defence University

ABSTRACT and diseases[3]. Heat waves can affect the quality of the water
encouraging the growth of hazardous algae that can produce
Heat waves have become one of the major calamities in the toxins in drinking water, bodies, increasing the population's
present world because of global warming and the rise in danger of poisoning[4] 2016.
temperature. Exceptional temperatures that are recorded over
a period can be known as extreme heat events or heat waves. It is very important to measure the heatwaves to estimate the
Even though heat waves are less exciting than other disasters danger it brings upon the human life. A threshold must be
like floods, tornados and earthquakes, they are an equivalent established to determine whether temperature levels above the
hazard to the existence of life on earth. It is critical to forecast average should be deemed a heat wave. The threshold
heatwaves throughout the year and take the necessary includes not only the ambient temperature but also the thermal
precautions to avoid losses. Due to the advancement in sensation encountered by people, relative humidity and human
machine learning (ML) techniques, ML can become handy in activities as factors[4]. The most well-known heat index has
predicting heat waves. Our main objective is to study different been recommended by the United States National
existing ML algorithms that can be used to predict extreme Oceanographic Administration (NOAA). But the main
heat events. The main approach to this study is to review the drawback to the heat index is the lack of access to relative
existing studies of machine learning techniques like the use of humidity information. These researchers discovered that basic
regression algorithm, K-Nearest Neighbour algorithm, Deep indices based only on temperature may be the most useful for
Learning algorithms and other significant ML algorithms that use in alarm systems, but that for regional analysis, all of the
are being used to predict heatwaves and heatwave-related relevant temperature thresholds should be incorporated[5].
predictions. This research study will help to compare and Heat waves can have a greater impact on highly populated
contrast different ML algorithms to help this research derive places than on other locations. Preventive measures can be
the best ML model that can be developed for future heat wave taken against heatwaves by creating sustainable cities and
predictions. communities. That is, establishing settlements that are
resilient, safe and sustainable as an adaptation to extreme
General Terms heat[6].
Machine Learning, Algorithms Heat waves due to global warming and climate change have
become a global issue. The negative impact of heatwaves on
Keywords human life and society is tremendous. It is very important to
Heat Waves, Global Warming, Hazard, Machine Learning predict extreme temperatures in the future to mitigate the
(ML), Prediction consequences. Machine learning (ML) algorithms like
1. INTRODUCTION artificial neural networks can be used to build climate models
Heatwaves have become a rising threat to life on earth. The to identify heat waves[7]. The average temperature over a
recent report on climate science by the U.N indicates that the period can be used as data to the machine learning models to
extreme heat wave that used to strike once every 50 years is predict the thermal waves.
now experienced very frequently due to human-induced With the use of the implications of the foremost literature, this
climate change[1]. Also, the report predicts that the study presumed that the application of machine learning
temperature will continue to increase with global warming. techniques can be used to predict the heatwaves. Therefore,
The earth's temperature is governed by how much heat energy this study has deduced two objectives.
from the sun is directed at it and how much of that energy is
contained by the greenhouse gases in the earth’s atmosphere. 1) To investigate the use of ML techniques to predict
Global warming has very adverse effects on human life. the heatwaves to mitigate the hazard caused to the
Drought, hunger and local disputes over resources killed human life and environment.
hundreds of thousands of people in some parts of Africa. [2]. 2) To discover means to inspire the climate scientist
Extreme hot weather, along with high humidity, is known as a and researchers to initiate their work on heatwave
heat wave, which is a consequence of global warming. These prediction and climate changes.
heatwaves are hard to explain in terms of natural climatic
The review conducted in this study has assessed previous
changes without human-made climate variations. The severe
literature available on heat wave prediction using ML
2003 heat wave, according to the World Health Organization
techniques using a group of keywords namely heatwaves,
(WHO) and different national studies, resulted in around
weather changes, climate changes, global warming, heatwave
70,000 more fatalities, especially in France and Italy. Heat
prediction using ML and machine learning techniques. This
waves cause widespread agricultural loss, several wildfires

33
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

study which exploits these keywords to review the proposed variables used to train the model were Latitude, longitude,
and implemented methods, models and applications in elevation, slope, and solar radiation. Machine learning (ML)
predicting the heatwaves using ML techniques. models were used to create prediction models on the data.
Random Forest model which is the best model for multi-class
2. LITERATURE REVIEW classification and regression problems[12][13] was trained.
2.1 Heat wave prediction using Regression Second model, an Artificial neural network (ANN) which has
a networked structure that mimics the functions and
Algorithm connections of actual neurons in the human brain was used
Global warming and the temperature rise have caused severe [14][15]. Also, a Support Vector Regression algorithm (SVR)
heat events to occur in several parts of the world. Heat waves was used to result prediction with minimum errors. Finally, a
can cause severe damage to human life. Predicting the Multi Model Ensemble was used after comparing the results
occurrence of heat waves is crucial in preventing human of other ML models. MME combines multiple ML models to
deaths from privations and sickness. A statistical model can achieve better accuracy and resilience [16][17].
be proposed to predict the heat waves in a chosen area. A
Quantile Regression model was proposed to predict the heat Global warming has caused a high heat flow in the region of
waves in Pakistan. The dataset used to train this model Antarctica. High Geo Thermal (GT) heat flow can thwart the
contains the daily temperature data from Princeton's Global glaciers in the arctic region [18]. Melting of Ice in the
Meteorological Forcing (PGF). The variables used to train the northern region can result in numerous adverse effects like sea
model was selected from data collected from National Centers level rising, impact on climate to cause extreme weather
for Environmental Prediction (NCEP). The model was able events, extinction of vulnerable animal species, etc. A Radient
forecast the days of occurrence of heat events and the Boosted Regression Tree algorithm is used to predict GT heat
departure dates in three months May, June and July with an flow in Antarctica. Dataset was created by collecting several
accuracy of about ±5 days. The evaluation of the variables reliable geophysical and geological global datasets. The
used in the training of the model revealed that the relative normalized root-mean-square error (RMSE/mean) of the test
humidity and the wind were two determining factors of the set, as well as the model score and R2 score, are calculated in
heat in Pakistan. Furthermore, the performance of the the same way as [19]. In the test set, the established prediction
Quantile Regression model (QRF) was compared to that of model has an R2 score of 0.44 and a relative RMSE of 29%.
the Random Forest model (RF). QRF produced better For the model evaluation the predicted values are compared
predictions compared to RF, proving its efficacy [8]. with the actual values [20].

The [9] publication from the EGU Assembly held in April Land Surface Temperature (LST) observed by satellites is an
2015 in Vienna shows that there is a huge impact of important factor in predicting heat waves. But due to the
urbanization on extreme temperatures, leading to severe heat disturbances from the clouds, the clarity of the LST recorded
events occurring in metropolitan areas. To mitigate can be poor, which can affect the poor prediction of the
temperature in the urban areas, this impact has drawn much heatwaves [26]. A regression model was developed to predict
attention. The research study [10] integrates spatial analysis the LST, which will be used in accurate heatwave prediction.
(satellite measurements), ground based measurements and The LST predicted using the ML model will fill in the gap
machine learning to create a model that can predict the urban between the satellite predicted LST and the actual LST[21].
areas with extreme heat waves. Using sensors mounted to The heatwave days per annum in Iran were predicted by
automobiles with a global positioning system (GPS), extracting the daily temperature from 27 points and trained
temperature data was collected from a few urban cities on using Ada-Boost Regression and decision tree (ABR-DT).
days with temperatures in the 90th percentile of historic The feature variables used to train the data were extracted for
averages. A Random Forest Regression model was used to do four different pressure levels. The model was evaluated by a
predict temperatures with high predictive power and low Root grid-point based evaluation by obtaining a corelation
mean square error (RMSE). An evaluation technique of cross coefficient of about 0.86 and a mean squared error of 6.929.
validation was done to avoid the over fitting of data. 70 The research work identified the best set of feature variables
percent of the data was used for training the model and 30 to predict the heatwave days in a year [22].
percent was used for testing [10].
There are few existing systems that can predict the number of
A quantile regression model is trained to predict the urban city heat strokes. Most of these systems are developed only using
temperature. The dataset used combines sensor data from meteorological variables, and none of the socio-economic
different sources, data from spatial and meteorological variables are in cooperation, which leads to a lack of
predictors. All this data was collected between June 25 and verification of these models and poor accuracy[23]. A
June 30, 2019, and a total of 85,942 data points were used. machine learning model was developed using both
The data collected (16 per cent) on 30th June was used to test meteorological and economic variables to predict heat stroke
the model. The model was evaluated in terms of RMSE. The to reduce the impact of the extreme heat events that occur in
objective of this study is to create high resolution temperature the cities of China. The daily heat stroke occurrences of
maps of urban areas to see extreme heat events and avoid the certain cities in China in 2012- 2014 were collected. The
serious ill effects that can occur[11]. socio-economic variables were collected from the statistical
year books of the cities in China. The novelty of this study is a
An overpopulated metropolitan city of Seoul in South Korea search index factor is used in the model. The term “Heat
was chosen as the area of the study to create a model that can Stroke” was used as the key word. The network search data
predict extreme heat events. From 2013 to 2017, the next-day was compiled using China's most popular search engine. The
maximum and lowest air temperatures were acquired from 25 search index was provided daily as a public data source, and it
Automatic Weather Stations (AWSs) run by KMA in Seoul was proven to have a greater connection with other
throughout July and August. The current day's maximum and meteorological characteristics such as temperature and
lowest air temperatures were calculated using hourly air relative humidity, demonstrating that the search index is a
temperature data from the 25 (AWS)s. Some of the feature reasonable predictor [23]. A Random Forest model was used

34
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

to train the data. The Boruta algorithm was used to filter the time and resources. Deep neural network models can
variables and maybe shorten the time it took to calculate the overcome these challenges and perform well in heatwave
follow-up random forest model establishment [24]. 90 per predictions [37]. A Convolution Neural Network (CNN) was
cent of data was trained and 10 per cent was used for testing used to train a very large dataset consisting of 1000 years of
and validation. The model performance was analysed using climate model data with the use of Transfer Learning. The
the linear fitting approach, and the model fitness was training and the test datasets were split into 900 and 100
measured using linear R2[25]. years, respectively, by random sampling. This was a
supervised classification problem trained using a CNN with a
Occurrence and the impact of heatwaves is higher in urban 4 layered architecture. Under sampling techniques were used
areas than the rest. The extreme temperature in the urban areas to overcome the challenge of having imbalanced classes in the
can greatly affect the vulnerable population like the elderly training set to predict the rare events.[38]. In predicting high
and the infants [26][27][28]. The fatality rates due to intensity heatwaves Transfer Learning was used since under
heatwaves were greater than 4000 in the urban city of Paris in sampling can degrade the performance [39].
the year 2003 [29]. Predicting higher air temperatures can be
helpful in designing early warning systems to alert the Marine Heat waves (MHW) can be a huge threat to the
population of extreme heat events. Crowd sourced data of aquatic fauna, flora and millions of humans whose main
higher air temperatures in the city of Berlin was used to train livelihood depends on the sea. Marine heatwaves can be a
ML models like Random Forest (RF), Stochastic Gradient great impact to the biodiversity in and out of the ocean. The
Boosting, and Model Averaged Neural Network comparativel. extreme heat event occurred in 2011 destroyed hundred
RF model performed well with a R2of 0.512. One of the kilometres of underwater ecosystem wiping out very rare
novelties of this work is that it is done using open source species of aquatic plants like kelp forests, seagrass meadows,
datasets and therefore can be adapted to predict extreme and coral reefs in the western coast of Australia [40]. It is
temperatures in other parts of the world[30]. important to forecast these extreme under water events to
prevent the aquatic life. The research work suggests to train
2.2 Heat Wave Prediction using K-Nearest several modified CNN like Spherical CNNs, CNN-LSTMs,
Neighbour Algorithm CNN-AEs and GANs and GNNs on Spatiotemporal data.The
A prediction tool was designed integrating the K-Nearest evaluated CNNs were integrated to achieve a better MHW
Neighbors model and weighted moving average algorithm to prediction [41]. LSTM a recurring neural network was
predict the flood and heatwaves that can occur using the integrated with the existing convolution neural network to
weather data obtained from the city of Mangalore, India. A train on satellite temperature data to predict the Ocean heat
heat wave is recognized by looking for abnormal temperature waves in Korea. The prediction of the model was evaluated by
behaviour in weather data over a long period of time. The the RMSE between the predicted and observed values [42].
model was evaluated by comparing the predicted temperature
against the actual average temperature [31]. 2.4 Prediction of Heatwaves using other
ML Algorithms
Extreme climatic events and heatwaves can destroy The industrialization and and the high emission of harmful
cultivation due to the lack of irrigation and finally can result gases into the atmosphere have caused a rise in the
in droughts. Drought forecasting is extremely useful in temperature of the earth. A predictive model was created
providing early warning and preparing the most vulnerable using cellular automata (CA) and artificial neural networks
areas for the worst effects of drought. Three ML models the (ANN) to predict the surface temperatures. The land satellite
K-Nearest Neighbours, Support-Vector machine and Artificial images from five-year intervals from 1995-2020 were used to
Neural Network were used to create a model to forecast create a dataset to train the models to forecast temperature.
droughts. The data set consist of data obtained from National Study computed overall accuracy, user accuracy, producer
Centres for Environmental Prediction/National Centre for accuracy, and Kappa statistics. Regression analysis was used
Atmospheric Research (NCEP/NCAR) reanalysis database. A to check the accuracy. One of the limitations of this study was
unique feature selection technique called Recursive Feature the CA model has limited ability to identify the explicit
Elimination (RFE) was also employed for the first time in relationship between the influential variables [43].
drought modelling to determine optimum sets of predictors. In
validation, KNN, which was utilized for the first time in Predicting the heat wave days in a year is important to get
building drought models, performed poorly when compared to prepared for the losses that may have to be faced in the future.
SVM and ANN-based drought models. The main limitation of Many of the heat wave prediction models perform poor due to
this study is the use of one dataset can lead to some the constant and inevitable changes in the climate [44]. The
uncertainties of prediction [32]. change in the ocean-atmospheric variable can affect the
performance of these forecast models in higher levels.
2.3 Prediction of Heatwaves using Deep Therefore, a constant update in the models is vital to maintain
Learning Algorithms the models reliable. A heat wave day (HWS) prediction model
Extremely hot summers and heatwaves pose a serious threat was developed using the Atmospheric Research reanalysis
to human life, even though people are not as conscious of data to predict the occurrence of heatwaves in Pakistan. The
heatwaves as they are of other calamities. There were so many machine learning models used in the work are Support Vector
reported incidents of high death tolls due to extreme Machines (SVM), random forest and artificial neural
heatwaves over many parts of the world, like over Russia in networks. The SVM model was selected comparatively based
2010[33], [34]or during the summer of 2021 in North on its best performance, with an R2 value of 0.86, to predict
America[35] or in western Europe during the summer of the heat wave days. An approach of updating the link between
2003[36]. Therefore, predicting heatwaves is important to the predictor and predictor was used to update the model
mitigate the losses. A physics driven approach to predict constantly to keep up with climate change[45].
heatwaves need a significant amount of computational cost,

35
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

Extreme weather conditions like heatwaves can be forecasted zero-inflated regression model was trained on Korean
by proper prediction of temperature [46]. Time Series Time- statistical data to predict the number of heat deaths in Korea.
Delay Neural Network, a feedforward network was trained on Some of the main parameters on which the model was
the temperature dataset obtained from the Morocco developed are the weekly mean of the daily temperature, the
meteorological administration to predict accurate temperature. number of elderly people in the area, the number of heatwave
The model predicted results with a corelation coefficient R of days and the vulnerability of people to heatwaves. The model
about 0.99. The model predicted temperatures were evaluated was evaluated by comparing the predicted number of heat
against the actual temperature values [47]. deaths with the actual number of heat deaths [50]. Another
regression model was used to predict the number of deaths
2.5 Prediction of Heat Wave Related due to heat disorders (NDHD). This model assesses the direct
impact of heatwaves on the human fatality rates. The model
Factors was trained on the National Statistics database on deaths due
The climate issue has resulted in high-intensity heatwaves that to heatwaves in the period 1994–2012. The model produced a
have wreaked havoc on human existence and caused high R2 value of 0.96 and a very low mean square error of 1.4
innumerable deaths. It is very important to take actions to deaths. The model was evaluated by using scatter and time
prevent and minimize the damages caused by these heat series plots to compare the estimated and observed fatalities.
events. Depending merely on the human instincts in taking Also, the prediction results from this model show that there
measures to disaster can lead to unproductive decision- will be extreme heatwaves in the year 2050 that will cause
making. Artificial intelligence-based analytical models can be about 250 deaths[51].
used to make effective solutions during catastrophe response.
A Random Forest model, a strong ML model that can be
applied both for classification and regression [48] was used to 3. RESEARCH METHODOLOGY
predict the heat wave related damages. The dataset was A systematic literature review has been conducted to study
created using the statistical, meteorological, and floating and review the recent literature on heatwaves and heatwave-
population data from South Korea in the years 2015-2018. related predictions with the aim of discovering how much
The Random Forest model was evaluated by comparing other work has been done using ML techniques in heat wave
traditional regression models in terms of the coefficient of prediction. Searching online research databases such as
determination, mean absolute error, root mean squared error, Google Scholar, IEE Explorer and Science Direct yielded
root mean squared logarithmic error, and mean absolute error high-quality research articles. Specific keywords were used to
(R2). The suggested model has an R2 value of 0.804 in a search both conference and journal articles in the database.
comparison with observed data. The loss functions were used Only recent research studies (from 2016 to 2021) were chosen
to evaluate the results with the actual values. The loss for the review. After the selection of relevant research work, a
functions of the random forest model used in this study were filtration process was carried out to select the best quality and
the Mean Absolute Error (MAE) and the Mean Square Error most relevant articles.
(MSE) [49]
The online databases were searched with the keywords
While predicting heatwaves are important its vital to predict "Heatwave prediction using machine learning" and "Extreme
the heat deaths to understand the intensity of the heatwaves. temperature prediction using machine learning". When
Heat deaths can occur by exposure to the excessive heat. searching the databases, there were filters applied to only
Predicting the number of heat deaths can give us an idea about show the research articles that had been done after the year
how much the heatwaves going to affect the human life. A 2016. Only the articles related to subject areas Environmental
science or Computer Science were selected. 31 articles which
are most relevant and most recent were selected for Exclusion Factors
thescreening process. The acquired research articles were
filtered with the following inclusion and exclusion factors.  Physics-driven heatwave prediction articles.

Inclusion Factors  Literature reviews on the heatwave predictions.

 Topics related to heatwave and heatwave related After the exclusion of the 31-research article used for the
predictions. screening process, finally 20 articles were selected for the
review of this study. Figure 1 shows the systematic
 Only the predictions using machine learning filteringprocess of the research works.
techniques.
 Only the articles technically addressing the
heatwave prediction.

36
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

Fig 1: Systematic filtering process of the research works

4. RESULTS heatwaves is as important as predicting the heatwaves

A deep analytical literature review was done on 17 out of 20 themselves to take necessary prevention measures.
(85%) papers which included work on heatwave predictions
(HWP). When reviewing this category, the research works The databases like IEEE and Science Direct yielded very few
were categorized according to the ML algorithm used by the results when the keyword "Heatwave prediction using ML"
research work. 7 out of 17 (41%) papers were reviewed under was used to search. When the same keyword was used in
the topic “Heatwave predictions using Regression models”. 2 other databases like Microsoft academic and Taylor and
papers out of 17 (12%) were reviewed under the topic Francis online, it yielded no results at all. That is one reason
“Heatwave predictions using Random Forest models”. 2 the authors had to search a social research platform like
papers (12%) were reviewed under the topic “Heatwave Google Scholar. When the same keyword was searched in the
predictions using K-Nearest Neighbour models”. 3 out of 17 database Google Scholar, it produced considerably large
papers (18%) were under the topic “Heatwave predictions results. But the relevance of the work was not compatible. So,
using Deep Learning models”. The rest of the 3 papers were a great effort was made to select the most recent and most
studied under the topic “Heatwave predictions using other ML relevant work (31 articles) for the screening process. Since
models”. This approach of study allows us to investigate the there were low results for the keyword "heatwave prediction
application of different ML algorithms in predicting the using ML," another related keyword, "extreme temperature
heatwaves. prediction using ML," was used.
Apart from the heatwave predictions its crucial to
predict the heatwave related factors too. Identification of how The result of this study shows that 41% of the literature used
much damage and deaths can an extreme weather like regression models for prediction. Although there are so many
heatwaves can cause allow us to predict the intensity of the advancements in the ML field, like the latest deep learning
heatwaves. This also will help in taking necessary measures in and transfer learning techniques, a considerably low amount
protecting human lives and properties. Therefore 3 such of work has been done using those models. Even when
research works were reviewed under the topic “Heatwave searching the databases, the results of literature using deep
Related Predictions”. learning for heat wave prediction were low. This work
emphasizes the fact that more work can be done to predict
heatwaves in the future using advanced ML techniques like
5. DISCUSSION. deep learning.
In this study, the research articles were reviewed to
understand the key points such as the ML algorithm used, the 6. CONCLUSION, LIMITATIONS, AND
dataset used, the evaluation of the model, and the limitations
of the work. This kind of study can give a complete idea of FUTURE DEVELOPMENTS.
how much work has been done in heatwave prediction and This systematic literature review has been done by deeply
can pave the way for future developments in this area. reviewing recent studies on heatwave predictions. For the
review, the research works were filtered and methodically
Few of the research papers reviewed in this literature review analyzed. The results of this show how much work has been
used multiple algorithms or the integration of multiple models done in heatwave predictions using different ML models.
to forecast heatwaves. But when grouping according to the According to the results, regression models are widely used
used ML algorithms, the model that is primarily used for the for heatwave predictions, and the amount of work that has
study is used. This study not only reviews the heatwave been done using the other ML models is considerably low.
prediction work but also considers the work related to the Significantly, the literature using advanced ML techniques
heatwaves. Predicting the damage and deaths caused by like deep learning models is low.

37
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

One of the limitations of this work is that there is not adequate Lead Detection,” Remote Sens., vol. 8, no. 9, p. 698,
recent literature available in the databases. The amount of Aug. 2016, doi: 10.3390/rs8090698.
work done in the field of heatwave prediction is very low,
even though heatwaves are a major climate threat. A greater [13]S. Yoo, J. Im., and J. E. Wagner, “Variable selection for
effort was put into selecting research papers needed for the hedonic model using machine learning approaches: A
screening process. After the exclusion of papers, limited case study in Onondaga County, NY,” Landsc. Urban
number studies are eligible for the review. Plan., vol. 107, no. 3, pp. 293–306, Sep. 2012, doi:
10.1016/j.landurbplan.2012.06.009.
Since there are so many recent advancements in the use of [14]R. Özçelik, M. J. Diamantopoulou, J. R. Brooks, and H.
ML techniques, a better heatwave prediction system can be V. Wiant, “Estimating tree bole volume using artificial
built using deep learning models and transfer learning with neural network models for four species in Turkey,” J.
better accuracy and performance as the future development of Environ. Manage., vol. 91, no. 3, pp. 742–753, Jan.
this literature review. 2010, doi: 10.1016/j.jenvman.2009.10.002.
[15] S. Tiryaki and A. Aydın, “An artificial neural network
7. REFERENCES model for predicting compression strength of heat-treated
[1]X. Gu et al., “Attribution of Global Soil Moisture Drying
woods and comparison with a multiple linear regression
to Human Activities: A Quantitative Viewpoint,”
model,” Constr. Build. Mater., vol. 62, pp. 102–108, Jul.
Geophys. Res. Lett., vol. 46, no. 5, pp. 2573–2582, Mar.
2014, doi: 10.1016/j.conbuildmat.2014.03.041.
2019, doi: 10.1029/2018GL080768.
[16]P. J. L. Adeodato, A. L. Arnaud, G. C. Vasconcelos, R. C.
[2]M. Munasinghe, “Rising temperatures, rising risks,”
L. V. Cunha, and D. S. M. P. Monteiro, “MLP ensembles
Finance Dev., vol. 45, Mar. 2008.
improve long term prediction accuracy over single
[3]W. Marx, R. Haunschild, and L. Bornmann, “Heat waves: networks,” Int. J. Forecast., vol. 27, no. 3, pp. 661–671,
a hot topic in climate change research,” Theor. Appl. Jul. 2011, doi: 10.1016/j.ijforecast.2009.05.029.
Climatol., vol. 146, no. 1–2, pp. 781–800, Oct. 2021, doi:
[17]D. Cho, C. Yoo, J. Im, and D. Cha, “Comparative
10.1007/s00704-021-03758-y.
Assessment of Various Machine Learning‐ Based Bias
[4]P. Martinez and E. R. Bandala, “Heat Waves: A Growing Correction Methods for Numerical Weather Prediction
Climate Change-related Risk,” p. 4. Model Forecasts of Extreme Air Temperatures in Urban
Areas,” Earth Space Sci., vol. 7, no. 4, Apr. 2020, doi:
[5]M.-A. P. F and B. E. R., “Heat Waves: Health Effects, 10.1029/2019EA000740.
Observed Trends and Climate Change,” in Extreme
Weather, P. J. Sallis, Ed. InTech, 2018. doi: [18]R. Dziadek, F. Ferraccioli, and K. Gohl, “High
10.5772/intechopen.75559. geothermal heat flow beneath Thwaites Glacier in West
Antarctica inferred from aeromagnetic data,” Commun.
[6]R. Ortlepp and D. Schiela, “Perspectives of climate change Earth Environ., vol. 2, no. 1, p. 162, Dec. 2021, doi:
adaptation of building areas against heat waves,” IOP 10.1038/s43247-021-00242-3.
Conf. Ser. Mater. Sci. Eng., vol. 615, no. 1, p. 012004,
Oct. 2019, doi: 10.1088/1757-899X/615/1/012004. [19]S. Rezvanbehbahani, L. A. Stearns, A. Kadivar, J. D.
Walker, and C. J. van der Veen, “Predicting the
[7]J. Safieh, R. Dara, and J. Forough, “Effect of climate Geothermal Heat Flux in Greenland: A Machine
change on heat waves in the South Sea of Iran,” Ukr. J. Learning Approach,” Geophys. Res. Lett., vol. 44, no. 24,
Ecol., vol. 10, pp. 87–93, Oct. 2020, doi: p. 12,271-12,279, 2017, doi: 10.1002/2017GL075661.
10.15421/2020_211.
[20]M. Lösing and J. Ebbing, “Predicting Geothermal Heat
[8]N. Khan, S. Shahid, L. Juneng, K. Ahmed, T. Ismail, and Flow in Antarctica With a Machine Learning Approach,”
N. Nawaz, “Prediction of heat waves in Pakistan using J. Geophys. Res. Solid Earth, vol. 126, no. 6, Jun. 2021,
quantile regression forests,” Atmospheric Res., vol. 221, doi: 10.1029/2020JB021499.
pp. 1–11, Jun. 2019, doi:
10.1016/j.atmosres.2019.01.024. [21]I. Buo, V. Sagris, and J. Jaagus, “Gap-Filling Satellite
Land Surface Temperature Over Heatwave Periods With
[9]M. Wang, X. Yan, J. Liu, and X. Zhang, “The contribution Machine Learning,” IEEE Geosci. Remote Sens. Lett.,
of urbanization to recent extreme heat events and a pp. 1–5, 2021, doi: 10.1109/LGRS.2021.3068069.
potential mitigation strategy in the Beijing–Tianjin–
Hebei metropolitan area,” Theor. Appl. Climatol., vol. [22]S. B. H. S. Asadollah, N. Khan, A. Sharafati, S. Shahid,
114, no. 3–4, pp. 407–416, Nov. 2013, doi: E.-S. Chung, and X.-J. Wang, “Prediction of heat waves
10.1007/s00704-013-0852-x. using meteorological variables in diverse regions of Iran
with advanced machine learning models,” Stoch.
[10]V. Shandas, J. Voelkel, J. Williams, and J. Hoffman, Environ. Res. Risk Assess., Oct. 2021, doi:
“Integrating Satellite and Ground Measurements for 10.1007/s00477-021-02103-z.
Predicting Locations of Extreme Urban Heat,” Climate,
vol. 7, no. 1, p. 5, Jan. 2019, doi: 10.3390/cli7010005. [23]T. Li, F. Ding, Q. Sun, Y. Zhang, and P. L. Kinney, “Heat
stroke internet searches can be a new heatwave health
[11]M. Zumwald, B. Knüsel, D. N. Bresch, and R. Knutti, warning surveillance indicator,” Sci. Rep., vol. 6, no. 1,
“Mapping urban temperature using crowd-sensing data p. 37294, Dec. 2016, doi: 10.1038/srep37294.
and machine learning,” Urban Clim., vol. 35, p. 100739,
Jan. 2021, doi: 10.1016/j.uclim.2020.100739. [24]M. B. Kursa and W. R. Rudnicki, “Feature Selection with
the Boruta Package,” J. Stat. Softw., vol. 36, no. 11,
[12]S. Lee et al., “Arctic Sea Ice Thickness Estimation from 2010, doi: 10.18637/jss.v036.i11.
CryoSat-2 Satellite Data Using Machine Learning-Based

38
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

[25]Y. Wang et al., “A random forest model to predict [37]J. A. Weyn, D. R. Durran, and R. Caruana, “Can
heatstroke occurrence for heatwave in China,” Sci. Total Machines Learn to Predict Weather? Using Deep
Environ., vol. 650, pp. 3048–3053, Feb. 2019, doi: Learning to Predict Gridded 500-hPa Geopotential
10.1016/j.scitotenv.2018.09.369. Height From Historical Weather Data,” J. Adv. Model.
Earth Syst., vol. 11, no. 8, pp. 2680–2693, 2019, doi:
[26]R. S. Kovats and S. Hajat, “Heat Stress and Public Health: 10.1029/2019MS001705.
A Critical Review,” Annu. Rev. Public Health, vol. 29,
no. 1, pp. 41–55, 2008, doi: [38]J. Johnson and T. Khoshgoftaar, “The Effects of Data
10.1146/annurev.publhealth.29.020907.090843. Sampling with Deep Learning and Highly Imbalanced
Big Data,” Inf. Syst. Front., vol. 22, Oct. 2020, doi:
[27]K. M. A. Gabriel and W. R. Endlicher, “Urban and rural 10.1007/s10796-020-10022-7.
mortality rates during heat waves in Berlin and
Brandenburg, Germany,” Environ. Pollut., vol. 159, no. [39]V. Jacques-Dumas, F. Ragone, P. Borgnat, P. Abry, and
8, pp. 2044–2050, Aug. 2011, doi: F. Bouchet, “Deep Learning-based Extreme Heatwave
10.1016/j.envpol.2011.01.016. Forecast,” ArXiv210309743 Phys., Oct. 2021, Accessed:
Nov. 29, 2021. [Online]. Available:
[28]D. Scherer, U. Fehrenbach, T. Lakes, S. Lauf, F. Meier, http://arxiv.org/abs/2103.09743
and C. Schuster, “Quantification of heat-stress related
mortality hazard, vulnerability and risk in Berlin, [40]N. J. Holbrook et al., “Keeping pace with marine
Germany,” ERDE – J. Geogr. Soc. Berl., vol. 144, no. 3– heatwaves,” Nat. Rev. Earth Environ., vol. 1, no. 9, pp.
4, Art. no. 3–4, 2013, doi: 10.12854/erde-144-17. 482–493, Sep. 2020, doi: 10.1038/s43017-020-0068-4.
[29]B. Dousset et al., “Satellite monitoring of summer heat [41]D. Ning, V. Vetrova, K. Bryan, and S. Delaux, “Deep
waves in the Paris metropolitan area,” Int. J. Climatol., Learning for Spatiotemporal Anomaly Forecasting: A
vol. 31, no. 2, pp. 313–323, 2011, doi: 10.1002/joc.2222. Case Study of Marine Heatwaves,” presented at the
ICML 2021 Workshop on Tackling Climate Change with
[30]S. Vulova, F. Meier, D. Fenner, H. Nouri, and B. Machine Learning, Jul. 2021. Accessed: Nov. 29, 2021.
Kleinschmit, “Summer Nights in Berlin, Germany: [Online]. Available:
Modeling Air Temperature Spatially with Remote https://www.climatechange.ai/papers/icml2021/87
Sensing, Crowdsourced Weather Data, and Machine
Learning,” IEEE J. Sel. Top. Appl. Earth Obs. Remote [42]S. Jung, Y. J. Kim, S. Park, and J. Im, “Prediction of Sea
Sens., vol. 13, pp. 5074–5087, 2020, doi: Surface Temperature and Detection of Ocean Heat Wave
10.1109/JSTARS.2020.3019696. in the South Sea of Korea Using Time-series Deep-
learning Approaches,” Korean J. Remote Sens., vol. 36,
[31] C. S. Bangera, P. S. Kotian, C. Dias, T. Divya, and G. no. 5_3, pp. 1077–1093, Oct. 2020, doi:
Aithal, “Flood and Heat Wave Prediction using 10.7780/KJRS.2020.36.5.3.7.
Weighted Moving Average, Anomaly Detection and K-
Nearest Neighbours for the City Of Mangalore,” in 2018 [43]A.- Al Kafy et al., “The operational role of remote
IEEE Distributed Computing, VLSI, Electrical Circuits sensing in assessing and predicting land use/land cover
and Robotics (DISCOVER), Mangalore (Mangaluru), and seasonal land surface temperature using machine
India, Aug. 2018, pp. 93–97. doi: learning algorithms in Rajshahi, Bangladesh,” Appl.
10.1109/DISCOVER.2018.8674113. Geomat., Aug. 2021, doi: 10.1007/s12518-021-00390-3.
[32]N. Khan, D. A. Sachindra, S. Shahid, K. Ahmed, M. S. [44]B. Wang et al., “Rethinking Indian monsoon rainfall
Shiru, and N. Nawaz, “Prediction of droughts over prediction in the context of recent global warming,” Nat.
Pakistan using machine learning algorithms,” Adv. Water Commun., vol. 6, no. 1, p. 7154, Nov. 2015, doi:
Resour., vol. 139, p. 103562, May 2020, doi: 10.1038/ncomms8154.
10.1016/j.advwatres.2020.103562.
[45]N. Khan, S. Shahid, T. B. Ismail, and F. Behlil,
[33]D. Barriopedro, E. M. Fischer, J. Luterbacher, R. M. “Prediction of heat waves over Pakistan using support
Trigo, and R. García-Herrera, “The Hot Summer of vector machine algorithm in the context of climate
2010: Redrawing the Temperature Record Map of change,” Stoch. Environ. Res. Risk Assess., vol. 35, no. 7,
Europe,” Science, vol. 332, no. 6026, pp. 220–224, Apr. pp. 1335–1353, Jul. 2021, doi: 10.1007/s00477-020-
2011, doi: 10.1126/science.1201224. 01963-1.
[34]F. E. L. Otto, N. Massey, G. J. van Oldenborgh, R. G. [46]Tushar Kanti Routh et al., “Artificial neural network-
Jones, and M. R. Allen, “Reconciling two approaches to based temperature prediction and its impact on solar
attribution of the 2010 Russian heat wave,” Geophys. cell,” in 2012 International Conference on Informatics,
Res. Lett., vol. 39, no. 4, 2012, doi: Electronics & Vision (ICIEV), Dhaka, Bangladesh, May
10.1029/2011GL050422. 2012, pp. 897–902. doi: 10.1109/ICIEV.2012.6317369.
[35]S. Y. Philip et al., “Rapid attribution analysis of the [47]A. Kabbori, J. Antari, R. Iqdour, and Z. E. Abidine El
extraordinary heatwave on the Pacific Coast of the US Morjani, “Temperature Prediction using Time Series
and Canada June 2021,” Earth Syst. Dyn. Discuss., pp. Time-Delay Neural Networks,” in 2019 7th International
1–34, Nov. 2021, doi: 10.5194/esd-2021-90. Renewable and Sustainable Energy Conference (IRSEC),
Agadir, Morocco, Nov. 2019, pp. 1–4. doi:
[36]R. García-Herrera, J. Díaz, R. Trigo, J. Luterbacher, and 10.1109/IRSEC48032.2019.9078299.“randomforest2001
E. Fischer, “A Review of the European Summer Heat .pdf.”
Wave of 2003,” Crit. Rev. Environ. Sci. Technol., vol.
40, pp. 267–306, Mar. 2010, doi: [49]M. Park, D. Jung, S. Lee, and S. Park, “Heatwave
10.1080/10643380802238137. Damage Prediction Using Random Forest Model in

39
 International Journal of Computer Applications (0975 – 8887)
Volume 184– No.16, June 2022

Korea,” Appl. Sci., vol. 10, no. 22, p. 8237, Nov. 2020, [51]D.-W. Kim, R. C. Deo, J.-H. Chung, and J.-S. Lee,
doi: 10.3390/app10228237. “Projection of heat wave mortality related to climate
change in Korea,” Nat. Hazards, vol. 80, no. 1, pp. 623–
[50]D.-W. Kim, “Weekly heat wave death prediction model 637, Jan. 2016, doi: 10.1007/s11069-015-1987-0.
using zero-inflated regression approach,” p. 16.

IJCATM : www.ijcaonline.org 40