CHAPTER-1

INTRODUCTION
 Chapter 1

Introduction

In earlier days, observing the sky was the first step in forecasting, which was
supplemented by the invention of weather forecasting instruments like Hygrometer,
Thermometer, and Anemometer etc. [1]. Observing techniques and forecasting
devices have developed to the state of art in recent years and launching exclusive
Meteorological satellites [2] and Radars has now made possible to monitor the
weather closely and accurately. In today’s fast telecommunication network, countries
exchange weather observations and updates rapidly through the help of
Meteorological satellites to produce near accurate predictions [3]. Apart from various
Government organizations and the weather observatory stations, many private
agencies have evolved to make predictions on the weather and this information are
shared through the latest smart devices is a good sign indicating the growth and the
development of Weather Forecasting and its technologies.

This Chapter discusses about Meteorology [4], its applications, Weather Forecasting
[5], Traditional Weather Forecasting methods, Forecasting Devices and Existing
Models. The chapter also focuses on the importance and the need for Weather
Forecasting, along with objectives of our proposed work and also highlights the social
significance of this research work.

1.1 Meteorology

Meteorology is the science of weather, in broad terms, it is the study of the

atmosphere [6] and is an integrated system constituting atmosphere, land and ocean
[7].The meteorology word comes from the Greek word called “meteoron” referring to
any phenomenon related to the sky [8]. Observing, understanding and interpretation
of weather are the three basic aspects of meteorology [9][10][11].Weather
observations are carried out through simple instruments like the thermometer [12],
anemometer [13] and are plotted on weather charts to be analyzed by meteorologists
at Weather stations. Recent advances in modern computers, supercomputers, radars
and satellites [12] have resulted in the noteworthy progress in the meteorology field

1
[14]. Weather and climate related changes have direct or indirect affect on most of the
economic, industrial, agricultural, social, commercial, transport related processes. The
atmosphere is affected and sustained upon lives of people, animals, pests, insects and
microorganisms, plants, trees, forests and marine life in their life cycle of growth and
development [15][16][17]. Meteorology, therefore, shows a major part in all facets of
modern human life. The Applications of Meteorology [18] are enormous and a few
are listed below:
 Weather forecasting
 Aviation meteorology
 Agricultural meteorology
 Hydro meteorology
 Military meteorology
 Nuclear meteorology
 Maritime meteorology

The Job of Weather forecasters differs like,

 An operational forecaster analyzes and issues day to day forecasts.
 A research meteorologist on disastrous events like earthquakes, tsunamis,
landslide, flash flood and erratic climate changes.
 In military, for safety and security of the nation.
 Airlines, updates the pilots about the weather status on take-off, land in and in
between.
 In Agriculture, for the cultivation of crops, meteorology plays a vital role in
crop production, the right period for sowing, in determining soil moisture, and
migration of pests etc.
Thus, Meteorology and Weather forecasting are closely related and serves the
mankind and all living beings for its existence and survival on this planet Earth. This
made us to draw to the conclusion of using Data Mining Techniques [19] for weather
forecasting.

1.2 Weather Forecasting

Weather Forecasting is high in complexity, multidimensional and a dynamic method

[20] as it includes a number of entities related to atmosphere along with various

2
theories of mathematics, thermodynamics, atmospheric physics, laws of motion,
conservation of energy, etc. So, meteorologists are continuously making efforts for
upgrading the forecasts related accuracy in spite of the sudden changes leading to
difficulty in predicting weather patterns and climate changes occurring worldwide due
to global warming [21]. Hence, Atmospheric Studies [22] and Weather forecasting is
being a continuous, updating and challenging phenomenon in today’s world [23].

Weather Forecasting is referred to as forecasting the forthcoming weather conditions

depending upon the historical data collection [24]. It is one of the vital applications of
Meteorology. Weather and Climate are the measures of time [25]. Weather is the
atmosphere’s condition during a shorter period of time, whereas, Climate is the
behavior of the atmosphere during the longer duration [26]. Out of several
meteorological elements major components, namely temperature, pressure, wind,
humidity, and precipitation has a significant impact in shaping the weather and the
climate of a place [27]. Analysis of these meteorological elements forms the basis for
forecasting the weather and to determine the climate of a location [28].

As a result of depending upon the number of factors, the procedure of weather related
predictions becomes complex and challenging as well [29]. Fluctuations in weather
conditions are noticed every few hours and extreme changes occur from time to time
[30]. Being aware about the weather conditions earlier itself leads to reduction in the
losses, helping us in numerous ways. Weather forecasting has wider applications
varying from being useful for a student to keep an umbrella when being aware that it
would rain in the evening to being useful for governmental establishments in
emptying a locality when being aware about the possibility of heavy rain in that area.
Forecasting is the undertaking of expectation of the environment at a future time and a
given zone [31]. In the early days, this has been completed through physical
conditions in which the air is considered as runny. The present condition of the earth
is studied, and the future phase is anticipated through unraveling those conditions
arithmetically, yet we cannot decide an exact climate for beyond what days and this
can be improved with the assistance of science and innovation [32].

Weather Forecasting is a process of distinguishing and foreseeing to a precise

exactness of climatic conditions by the usage of a number of innovations. Generally,
different procedures were utilized to foresee the climate, in view of perception of

3
ecological and meteorological components, for example, sunlight, clouds and animal
conduct. These predictions were not frequently logical or precise.

Climate estimates are made by gathering the most extreme measure of information
conceivable about the current scenario of the environment (specifically humidity,
temperature, and wind speed) and the usage of comprehension of barometrical
processes in deciding that how the atmospheric changes takes place later. In this
century, it is one of the most experimented and mechanically tested issue remaining
so far globally. This is predictable principally to two variables: first, it's utilized for
some human exercises and moreover, due to the advantage made by the various
mechanical advances which are justifiably identified by this solid research field, likely
to the progress of calculation and the enhancement in estimation frameworks. Making
a precise forecast is becoming one of the significant difficulties confronting
meteorologist everywhere globally. Climate prediction is one of the most intriguing
and entrancing area since ancient times [33]. Researchers have made attempts in
finding out the meteorological qualities by the use of many techniques, a part of these
strategies being more accurate than the other ones [34].

Weather prediction is attracting the enthusiasm of the exploration network, for the
most part on account of its impact on human life. An earlier admonition for potential
atmosphere calamities may possibly several numbers of lives throughout the years to
come [35]. The capability of the climate prediction has been demonstrating its value
in the horticultural division like arranging ranch tasks, for example, stockpiling and
transportation of nourishment grains. Earlier data of cyclones and heavy rainfalls are
vital for the security and insurance of human life [36]. It further aides in diminishing
the harm acquired as a national loss. Extremely worst climate condition likewise
hampers the transportation and administration division truly, causing an expansion of
dangers like mishaps, and postponement in the offered administration just as the
nature of service.

The procedure of weather forecasting is fundamentally reliant on model-based

techniques like recreating dynamical frameworks by using the intensity of a fractional
differential equation [37].The Numerical expectation of a climate gauge utilizes
computational models [38] planned based on notable physical standards related with
seas [39] and air, to improve the estimation of future environmental conditions, in

4
view of the present climate and other influencing parameters. The environment can be
demonstrated like a liquid. The present atmospheric state can be examined, and the
forthcoming state will be looked like by getting the arrangement of liquid elements
and thermodynamics conditions. Quantitative estimate such as humidity, rainfall and
temperature are proved to be significant in the horticulture zone, just like merchants
inside product markets [40][41]. These days, we are utilizing numerous
methodologies for climate estimation [42]. Statistical modelling, Mathematical
modelling and Artificial Intelligence strategies are some of them [43]. Mathematical
modelling of the climate to foresee future climate dependent on current climate
conditions is numerical climate prediction. It needs full information of atmospheric
elements and includes estimation with countless factors and datasets [44].

The traditionally used weather forecast procedures that used satellite images and
weather stations are costly due to the inclusion of processes being high in cost and
complexity both [45]. Weather forecast by the use of machine learning is low in cost,
takes lesser time, higher in convenience, real in time and precise in nature [37]. A few
of the present researches related to weather forecasting including machine learning
technique involved the usage of much of the former weather data [46]. The accuracy
of the forecasts depends upon the models being trained with. Thus, it becomes much
essential for any machine learning model to be trained with a highly precise data. The
data attained from a number of sources is not trustworthy all the time. Thus, it
becomes necessary to preprocess the data. Preprocessing of the data includes the
removal of needless columns that are not relevant to the model’s forecast, removal of
the zero values, merging the same columns and various other stages related to pre-
processing [47].

Machine Learning is generally strong to annoyances and doesn’t require some other
physical factors for expectation. Hence, AI is greatly improved open door in
advancement of climate estimation [48]. Prior to the progression of Technology,
climate estimation was a hard nut to separate. Climate forecasters depended upon
satellites, information model's environmental conditions with less precision. Climate
forecast and investigation has inconceivably expanded regarding exactness and
consistency with the utilization of Internet of Things since most recent years. Using
the intensity of strategies identified with AI, and to give a superior and proficient
arrangement of this information requesting test, which contains derivations over the

5
existence, has acquired a huge measure of consideration of the specialists in the last
decade [49]. While contrasting, numerical forecast of climate, focal point of
information driven methodologies is for using authentic climate information for the
preparation of AI models. The AI models, as a rule, have two kinds of design, shallow
and deep. The shallow AI strategies centers on dealing with crude information and the
expert of this area is basically required in the change of un-important information into
significant portrayals. Though, deep learning based models have been altering
different application spaces, promoting range from sound-related to vision signal
preparing. The key bit of leeway of deep learning based strategies is that they have the
ability to build claim highlights, during the preparation period of a model. It is very
attractive to chop down the prerequisite of any area mastery for preparing an exact
model [50].

1.2.1. Importance of Weather Forecasting

The beginning of weather forecasting started from the early civilizations and was
dependent upon the observation of reoccurrence of the astronomical and
meteorological processes. Presently, weather forecasts depend upon the collection of
data related to the atmosphere’s present state and the usage of scientific processes in
predicting the further evolution of atmosphere. The disordered atmosphere’s nature
and the immense computational power essential for solving the overall equations
describing the atmosphere leads to decrease in the accuracy of the weather predictions
and also becomes costly with the rise in the range of these predictions. This makes us
in brainstorming more novel processes for weather forecasting being less costly
and/or more efficient.
So, forecasting the weather is being a challenging and complex process for
researchers, demands a heterogeneous combination of expertise, observations, and
techniques adopted from multiple disciplines. Weather on the way is important to
many people. For example, farmers plan their seeding, planting, watering and
harvesting schedules based on the upcoming weather [51].The Aviation department
decides their flight take off, landing, flying based on the weather
report[52].Tournaments[53], Big events, Series matches depends on weather schedule
before fixing the dates[54] and for a common man, in his daily day to day events. It is
significant for so many reasons like,

6
  Agriculture and Crop production
 Cultivation, sowing, weeding, harvesting
 Aviation
 Water Transport
 Marine and Sea Food
 Fisheries
 Irrigation
 Hydro power projects
 Military operations
 Natural disasters
 Storage of food grains
 Protection of livestock

1.2.2. Traditional Weather Forecasting

Human beings started to predict the weather informally over stone ages, and formally
from the nineteenth century. The art of forecasting weather was started with the very
early civilizations and it was based on the various astronomical phenomenon,
meteorological events eventually assist them to track the seasonal variation in the
weather [55]. Throughout the centuries, people made forecasts based on their weather
knowledge and personal expertise [56]. By the end of the renaissance, it became the
order of the day for more advanced knowledge that led to the inventions of various
weather instruments.

Weather instruments were invented to measure the properties of moisture,

temperature [57], wind speed [58], direction and pressure. The meteorological
instruments were improved and refined through the centuries, and developments
based on observational, theoretical, and technological aspects also contributed to the
knowledge of the atmosphere [59].

I. Looking at the Sky:

In ancient times, the weather was predicted by observing the sky conditions. In 650
BC, Babylonians started weather prediction by close observation of the sky and the
clouds [60] then after in 340 BC, Aristotle, Greek philosopher wrote a book
"Meteorological" about the formation of clouds, rain, wind and thunder storm [61].

7
 II. Use of Various Instruments:
The growth of science, in all domains paved the way for the long and rich history of
developments of weather instruments too [62]. The advent of weather instruments
namely thermometer, hygrometer, barometer, anemometer etc. helped to lay down the
framework for modern astronomy, meteorology and in turn weather forecasting [63].
Nicholas Cusa, the German scientist started using hygrometer to measure the
humidity of air during 1401-1464. Condensation and evaporation produce water
vapour in the atmosphere, which is measured as humidity. It can be measured as
either absolute or relative humidity. Galileo Galilei, an Italian scientist started using
thermometer to measure the temperature during 1564-1642 [64]. After that during
1608- 1647, Evangelista Torricelli, an Italian scientist started using barometer for
measuring atmospheric pressure [65]. Weather stations were equipped with these
different weather instruments to observe the sky and the atmosphere and thus the
weather data were accumulated for further processing [66]. In addition, scientists
launch weather balloons and satellites to collect data from the atmosphere. In 1450,
Leon Battista Alberti invented the anemometer, which is used for wind speed
measurement, and in 1846, cup anemometer was invented by Dr. J.T.R. Robinson
invented the cup anemometer. In the 1960s and 1970s,anemometers used lasers or
sonars to measure wind speed [67].

III. Analog Technique:

It is a composite technique, requiring the forecaster to call upon the past climate
conditions for making an assumption. It is a helpful technique for watching
precipitation in spots, for example, seas, just as the determining of precipitation sums
and conveyance later on. A comparative procedure is utilized in medium range
estimating, which is known as teleconnections, when the frameworks in different
regions are used to assist pin with the use of the area of another framework by
bringing down inside the encompassing system.

IV. Ensemble Forecasting:

Meteorologists have created barometrical models to predict the atmospheric
conditions by utilizing group estimating to depict how air temperature, weight and
humidity will change after some time. The conditions are customized into a
computing machine and the information on the present climatic conditions is
8
sustained into it. The machine analyses the circumstances to determine how the many
atmospheric factors will change over the next several moments. The machine repeats
this process, this time using the yield from the previous cycle as a contribution to the
next cycle. The computing machine outputs its predetermined data at a later period.

V. Radar:
After World War II , the usage of meteorological radars comes in existence to locate
precipitation, calculate its motion, and estimate its type [68]. Radio Detection and
Ranging is what Radar stands for. A transmitter transmits radio waves in radar. Radio
waves locate the nearest item and then return to a receiver. Climate radar can detect a
wide range of precipitation characteristics, including area, movement, force, and the
likelihood of future precipitation [69]. The most common type of climatic radar is
Doppler radar, which can also track the rate at which precipitation falls [70].

VI. Weather Satellites:

Climate satellites have become increasingly important sources of climate information
[71] since the first was propelled in 1952 and in 1960's first meteorological satellites
named Tiros I was launched. Climate satellites are the most ideal approach to
compute the weather conditions, similar to storms, volcanic ejection, and
contamination. They can record long duration changes. In the electro-magnetic range
climate satellites may watch all vitalities from all wavelengths range. Most
considerable are the obvious light and the heat (infrared) frequencies [12].
These are of two types, first is Polar Orbiting Satellites which are launched parallel to
the meridian lines and orbiting over the north to the south poles on each Earth’s
revolution, also helps in capturing the images perfectly and second is Geostationary
Satellites which orbits the equator with the same speed as that of the earth, in a
geosynchronous orbit enabling for continuous monitoring of a specific region,
especially useful for the prediction of Storm, as it constantly watches and observes
over a period of time [72].

VII. Weather Maps:

Weather maps are visual representations of weather conditions in the environment..
Climate maps may demonstrate just one component of the air or numerous highlights.

9
 They can use data from mechanical models or from human perceptions. Climate maps
are mostly found in papers and on the Internet [73].
On a climate map, significant meteorological conditions will be represented for each
climate station. Tides, flow of environment, dew point, overcast spread, temperature
and wind speed are all factors to consider. On a climate map, meteorologists utilize a
wide range of descriptions. These images give them a speedy and simple approach to
put data onto the guide.

VIII. Latest Devices and Technologies:

Apart from the aforementioned weather forecasting instruments, today’s weather
prediction [74] procedures are being automated, by using various tools and devices
such as Weather Research and Forecasting (WRF) [75], , Weather Processing System
(WPS), Advanced Research WRF (ARW), Numerical Weather Prediction (NWP)
Models, Doppler Weather Radars [76], Storm / Cyclone Detection Radars, and
Cyclone warning dissemination system (CWDS) [77]. Also, the installation of many
Weather observatory centers and Automatic weather stations (AWS) has been done.

1.3 Classification of Weather Forecasting

Every weather forecast can be characterized based on following measures:

 Temporal range of validity after emission
 Dominant technology
 Characteristics of input and output time and space resolution
 Accuracy
 Broadcasting needs
Weather is one of the significant factor of our lives, and it tends to be alluded as the
one that can't be controlled whereas climate all the more regularly controls how and
what individuals do that is the place we live, what we wear, and even what we eat. It
is made out of the parameters like breeze, deceivability, precipitation, downpour, day
off, cloud, weight and stickiness. Climate marvels for the most part occur in the lower
some portion of air that is troposphere [78] which fundamentally happens because of
the pneumatic force, temperature and dampness contrasts between to each other.
These distinctions are caused seen because of the point of the sun that fluctuates with

10
the scope. The polar and tropical zones shift generally in temperature and this wide
temperature differentiate prompted barometrical courses. Emergent upon the period,
forecasting is sorted into five kinds that are given below:

Now Casting:
This type of forecasting gives details of the present weather and forecasts up to 3-4
hours ahead (less than 24 hours) [79]. It uses observational data and extrapolates the
information using the latest results of numerical weather prediction (NWP) models
into the future [80][79].

Short Range Forecasting:

This is the estimation of climate conditions on a regular routine that roughly predicts
for one to seven days [81]. Predictions are based on satellite pictures, maps climate
outlines. In this type of forecasting, constant and progression techniques are utilized.
In everyday life, short range climate forecasting assumes a significant job in
transportation [82].

Medium Range Forecasting:

Forecast of climate conditions around one day to one month ahead of time[83].
Medium range climate forecasting is made by the computations done by considering
significant time span of climate conditions [84]. It is done by the worldwide
environmental models which is dependent on deterministic techniques [85].

Long Range Forecasting:

Widen the range for estimation and spread periods between one month and a year
ahead of time [86]. Long range estimates are produced using fortnight to the period of
a year ahead of time [87]. It doesn’t contain itemized data and has least exactness. It
has utility during warmth and cold waves and during dry seasons [88].

Hazardous Weather Forecasting:

Climate risks are compromising climate events that are hazardous to life and
properties [89]. Climate dangers can resemble tornadoes, lightning, tempest, hails,
streak flooding and so forth [90]. The National climate administration has named

11
dangerous climate viewpoint, a sort of articulation that is given so as to give data of
perilous or extreme climate occasions coming up inside seven days.

1.4 General Methods of Weather Forecasting

The most common types of weather forecasting methods are classified into four types,
namely persistence method, physical method, statistical method (time series and
ANN), and hybrid methods.

1.4.1. Persistence Method (or) Naive Predictor

A persistence method is a simplest and most economical method to forecast the

weather. This approach is based on the assumptions that present and future
temperature values are strongly correlated [91]. If P(t) is the observed temperature on
time (t), then the expected temperature at t+ Δt could be expressed as a linear equation
and is as follows:
P(t + Δt) =P(t) (1.1)

The linear equation above shows that temperature at time ‘t+ Δt’ is supposed to be the
same as ’t’ at time. For very short-term temperature forecasting, this approach is more
precise than the most physical and statistical approaches. Hence, for new forecasting
techniques should be tested against persistence method to validate the performance of
this technique.
Limitation of the persistence method is that if the forecasting lead time gets increased
the accuracy of this method gets decreased [92].

1.4.2. Physical Method

The physical approach models are the dynamics of the atmosphere, parameterized by
the Planetary Boundary Layer (PBL) theory, also recognized as the Atmospheric
Boundary Layer (ABL) [93]. The lowest portion of the atmosphere that is in continual
contact with the earth's surface is ABL. Here, the physical numbers are turbulent and
vertical mixing is higher, such as wind/air velocity, temperature, and moisture. The
physical techniques consist
of certain physical equations to translate meteorological data from a certain time to
temperature forecasts at a location considered. For long-term forecasting, this

12
approach is more efficient and reliable. Numerical Weather Prediction Method
(NWP) is a type of Physical method.

Numerical Weather Prediction Method (NWP)

The atmosphere is stimulated by the numerical weather forecast model by the
numerical integration of motion equation beginning from the present atmospheric
conditions. Numerical Weather Prediction (NWP) is based on a physical equation
that uses a variety of weather parameters to solve a complex mathematical equation.
The Fifth Generation Mesoscale Mode (MM5), Regional Spectra Model (RSM),
Weather Research and Forecasting (WRF) model, Prediktor, HIRLAM, etc. are some
of the NWP models.
The limitations of NWP model is that it is complex, expensive, limited observation set
for calibration and takes high computational time.

1.4.3. Statistical Method

Based on training the model using a required sample of real data unique to that area, a
statistical approach is applied, taking a number of discrete periodic cycles. The
mathematical approach is based on training the model and uses the difference between
the predicted values and the real immediate values to change the model parameters to
minimize the forecast error. For short and medium term temperature forecasting, this
approach is efficient and most reliable.
Limitation of the statistical method is that as the forecasting time increases the
forecasting error also increases. Despite this limitation, this method is very simple,
low cost and any stages of modeling are possible. This method is based on patterns
rather than the predefined mathematical model.
The statistical method is further divided into two subdivisions:
(i) Time Series methods
(ii) Artificial Neural Network models.

(i) Time Series Method

The method of the Time Series attempts to model the stochastic mechanism that
generates the structure of an observable series of events that are observed at certain
intervals and to render future predictions via the observational values belonging to the

13
previous interval. Time Series method does not require any records beyond historical
wind data. This method accurately provides the timely forecasting, and it is easy to
model. Some models of Time Series method are Auto Regressive (AR), Auto
Regressive Exogenous (ARX), ARMA with Exogenous inputs (ARMAX), Auto
Regressive Integrated Moving Average (ARIMA), Auto Regressive Moving Average
(ARMA), Grey Predictor, Linear Predictor, Algebraic Curve Fitting (ACF),
Exponential Smoothing, and Bayesian Model Averaging (BMA).
Limitation of the time series method is that it cannot forecast more than a day ahead.

(ii) Artificial Neural Network (ANN) Models

Artificial Neural Network is one of the analysis paradigms that are roughly modeled
after the massive study on parallel structure of the brain. Artificial Neural Network
(ANN) deals with the nonlinear and complex problem in terms of classification (or)
prediction. ANN has an ability to perform nonlinear and complex modeling without a
prior knowledge of the relationship between input and variables. Based on previous
temperature measurement data taken for a long period of time, ANN is trained to
learn the relationship between input data and temperature output.
ANN has strong self-learning ability, fault tolerance ability, real-time operation,
adaptability, and cost-effectiveness (To learn the relationship of any mathematical
formulation between inputs and outputs).
Few types of ANN methods are feed-forward (BPN, MLP, RBFN), Feedback
(ELMAN, Recurrent), Support Vector Machine (SVM), ADALINE, Probability
Neural Network (PNN), and so on.
The limitations of ANN method includes, falling into local minimum, slow
convergence, difficult to confirm the structure of a network (or) system. Despite these
limitations, ANN method outperforms the time series method in all time scale.

1.4.4. Hybrid Methods of Forecasting

The hybrid approach is a blend of various methodologies that are used to forecast
precise wind speed and power over various time scales. The objective of the hybrid
method is to get benefited from the merits of each method and obtain a worldwide
best forecasting performance. Types of combinations are as follows:
 Combination of physical and statistical methods (time series)

14
  Combination of physical method along with statistical method (ANN)
 Combination of statistical method and novel method
 Combination of a novel method and a physical method
 Combination of statistical (time series) and statistical (ANN) method
Some of the hybrid methods are Evolutionary Computation (EC) + Fuzzy, Wavelet
transform + Fuzzy, EC+ANN, Fuzzy + time series, ANN+NWP, NWP + time series,
ANN + Fuzzy, and ANN + time series.
Hybrid methods advantages are that they avoids over training and high computation
cost, achieve the optimal forecasting accuracy by reducing the forecasting error,
avoids the local minima problem, and faster the convergence. Limitation of hybrid
method is in some case, the single method outperforms the hybrid method.

1.5 Objectives of the Thesis

The aim of research is to improve the network from perturbation effects.

 To develop a model, that will provide improved weather prediction.
 To achieve higher accuracy using deep learning models.
 To do the short period weather condition based rapid prediction, using deep
learning models.
 To evaluate our technique and compare it with several existing models for
climate prediction.

1.6 Structural Organization of Thesis

We can find that weather forecasting has been an ever-challenging research area and
if our prediction models keep in pace with the rapid changing environment we can
minimize the loss of life and properties. Also, we can make the public well informed
and updated with the natural calamities.
The thesis is divided into eight chapters. The contents of the thesis are organized as
follows:

Chapter 1 contains the introduction that is the background, classification, and general
methods of weather forecasting; it discusses the existing weather forecasting

15
instruments and models. It also highlights the objectives and need for the research and
its social significance.

Chapter 2 presents a broad literature review which includes the previous work related
to the study. A thorough survey is done with importance given to papers, which
narrates about implementation techniques and discusses about its experimental results
along with its limitations.

Chapter 3 describes the general methodology used for forecasting with different
models, which includes data collection, pre-processing of data, implementation setup,
statistical and deep learning models used for training and different parametric
evaluation metrics used for the evaluation process.

Chapter 4, 5 and 6 elaborates the construction and working of ARIMA, CRNN and
proposed Hybrid_Stacked Bi-LSTM models respectively and how weather prediction
model is designed using these models. For all the three models, their architecture is
briefly presented and the results, screen shots and graphs are produced and elaborately
discussed at the end of each model respectively.

Chapter 7 incorporates comparison and performance analysis of all three models on

different parameters. It compares all the three methods adopted for weather
forecasting, in terms of accuracy based on different error values that is MSE, RMSE,
MAE and the results are validated. Appropriate charts and graphs are included to
increase the quick interpretation of outputs.

Chapter 8 concludes this research works with the summary of the proposed models,
major findings, its outcomes and limitations. The scope for further research is also
discussed, which offers a great way to extend in different areas.

At last, list of references which are used for finalizing the work are cited.

16