Prediction of Heatwaves over India region using Artificial Intelligence and
Machine Learning models
Nagaraju Chilukoti
Climate Change Research Lab
Dept. of Earth and Atmospheric Sciences, NIT Rourkela, Odisha, India
chilukotin@[Link] / chilukotinagaraju@[Link]
India suffers from major heatwaves during March to June. Heatwave causes adverse effect
on health and discomfort to many people; severe heatwaves may lead to the human and animal
mortality. Heatwaves also significantly impact the agriculture production, tourism, industrial
activity and many other sectors. The increase in frequency and intensity of these heatwaves in
recent years has been attributed to global warming. It is necessary to predict these extreme events
in advance to take appropriate mitigation activities. The accurate and sufficient lead time
predictions of heatwaves will be beneficial for the society. In this study we have developed three
different types of models using Artificial Intelligence (AI)/Machine Learning (ML) (CNN, LSTM,
Random Forest) tools to predict heatwaves. With the help of these models, the prediction of
heatwaves over two regions are developed (North and Central India, East Coast region in India).
Based on correlation between predictors five influencing parameters/variables are selected as input
to the models, by using these variables three models are developed and run over period 1952-2020,
the data from 1952-2006 used to train models and remaining period used to estimate the model
skill in predicting the heatwave events. The results revealed that among three AI & ML models
Random Forest has shown promising results compared to other models (CNN, LSTM) with
robustness of the developed models. We have also compared our AI model result with NWP
(WRF) model predictions, results shows that AI model shown reliable forecast over WRF model.
Keywords: Heatwave, Artificial Intelligence, Machine Learning, CNN, LSTM, Random Forest,
WRF.
The Prediction of Heatwave events in
India using Artificial Intelligence and
Machine Learning models.
By
Dr. Nagaraju Chilukoti
Dept. of Earth & Atmospheric Sciences
NIT Rourkela
chilukotinagaraju@[Link]
National seminar on CURTAIN RAISE for Platinum Jubilee Celebrations
18-20 September 2024
Andhra University, Visakhapatnam,
Introduction
u India suffers from major heatwaves during March -June.
u Heatwave causes deaths of So many people and has effect on human
mortality.
u In May 2015, a severe heat wave affected parts of southeastern parts
of India (Telangana and Andhra Pradesh) which claimed lives of
more than 2500 people.
u There has been a number of mega heatwaves in recent years
worldwide.
u Rising global temperatures are causing increases in the frequency
and severity of extreme climatic events, such as floods, droughts,
and heat waves.
Source: Rohini, P., Rajeevan, M., & Srivastava, A. K. (2016). On the variability and increasing trends of heat
waves over India. Scientific reports, 6(1), 1-9.
Time series of (a) frequency (b) total duration (days) and (c) maximum duration (days) of heat waves averaged over
the region 22°–31°N, 70°–77°E. The linear trend line (red line) and the 95% confidence limits (black lines) are also
shown.
Source: Rohini, P., Rajeevan, M., & Srivastava, A. K. (2016). On the variability and increasing trends of
heat waves over India. Scientific reports, 6(1), 1-9.
Standardized trends show the
correspondence among the
three variables. In years when
heat wave days (yellow) and
summer mean temperature
(red) are above average, heat-
related deaths also spike
upward.
Source: Mazdiyasni, O., AghaKouchak, A., Davis, S. J., Madadgar, S., Mehran, A., Ragno, E., ... & Niknejad, M.
(2017). Increasing probability of mortality during Indian heat waves. Science advances, 3(6), e1700066.
Study Region
North & Central India
East Coast region
u In this study we have developed three different types of models using Artificial
Intelligence (AI)/Machine Learning (ML) (CNN, LSTM, Random Forest) tools
to predict heatwaves.
u Based on correlation between event and predictors, three influencing
parameters/variables are selected as input to the models, by using these
variables three models are developed and run over period 1952-2020, the data
from 1952-2006 used to train models and remaining period used to estimate the
model skill.
Variable Source Period
Temperature IMD Gridded Data
1952-2020
RH, U,V, MSLP NCEP Reanalysis
Heat wave events
North and central India East coast India
18 - 24 May 2013 15 - 21 May 2007
21 - 30 May 2015 14 - 20 May 2008
22 - 28 May 2016 26 May- 03 Jun 2010
22 - 30 May 2018 30 May- 06 Jun 2012
23 - 28 May 2020
CNN model Prediction
[Link] Events CORREL RMSE
1 15-21 May 2007 0.84 0.92
2 14-20 May 2008 0.89 0.40
26 May – 3 Jun
3 0.42 1.40
2010
30 May - 6
4 0.34 0.60
Jun2012
5 18-24 May 2013 -0.15 0.09
6 21-30 May 2015 0.63 0.19
7 22-28 May 2016 0.35 3.52
8 22-30 May 2018 0.76 6.74
9 23-28 May 2020 0.69 2.91
LSTM model Prediction
[Link] Events CORREL RMSE
1 15-21 May 2007 -0.39 1.69
2 14-20 May 2008 0.94 0.52
26 May – 3 June
3 0.90 1.08
2010
30 May - 6
4 0.40 0.02
June2012
5 18-24 May 2013 0.23 0.97
6 21-30 May 2015 0.59 0.92
7 22-28 May 2016 0.45 2.77
8 22-30 May 2018 0.69 4.52
9 23-28 May 2020 0.40 4.13
RANDOM FOREST
model Prediction
S no Events CORREL RMSE
1 15-21 May 2007 0.22 0.83
2 14-20 May 2008 0.97 1.43
26 May – 3 June
3 0.92 3.15
2010
30 May - 6
4 0.35 0.28
June2012
5 18-24 May 2013 0.54 1.25
6 21-30 May 2015 0.55 2.46
7 22-28 May 2016 0.77 2.58
8 22-30 May 2018 0.71 3.55
9 23-28 May 2020 0.52 3.47
Comparison of AI/ML with WRF model
MODEL CORREL
ATION BIAS RMSE
CNN 0.98 1.07 0.53
LSTM 0.79 0.35 0.41
Random 0.99 0.23 0.17
Forest
WRF 0.91 1.43 0.72
Summary & Conclusions
u CNN model is able to predict heat wave events with average
correlation of 76%
u LSTM model able to predict heat wave events with correlation of
87% and biases are less than CNN.
u Random Forest model able to predict heat wave events with
correlation of 96%.
u WRF model able to predict events but not as good as AI&ML
models.
u We have noticed that, Random Forest is having good correlation of
99% among all models with reasonable bias.
