Time Series Weather Forecasting Techniques: Literature Survey

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
p-ISSN: 2395-0072
Volume: 09 Issue: 06 | Jun 2022 www.irjet.net
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3097
Time Series Weather Forecasting Techniques: Literature Survey
Janhavi Patil1, Prof. Nirmala Shinde2
1MTech Computer Engineering Student, K J Somaiya College of Engineering, Mumbai, Maharashtra, India
2Professor, Dept. of Computer Engineering, K J Somaiya College of Engineering, Mumbai, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract -Forecasting is a part of statistical modellingthat
is widely used in several fields because of its benefits in
decision-making. The purpose of forecasting is to predict the
future values of certain variables that range with time using
its previous values. Time Forecasting is related to the
formation of models and methods that can be used to produce
a good forecast. This research is survey paper research that
used a systematic mapping study and systematic literature
review. Generally, time series forecasting uses linear time
series models, specifically the ARIMA model andLSTMthathas
long been used because it has good forecasting accuracy. The
goal of this research is to review time series forecasting
methods such as ARIMA, Prophet and LSTM and analyze the
working of time series forecasting methods. It also discusses
the approaches of different methods used in time series
forecasting. Its goal is to increase the amount of awareness
regarding time series forecasting and its methods.
Key Words: Time Series models, ARIMA, LSTM, Prophet,
Accuracy, Forecasting
1.INTRODUCTION
Time series data forecasting is a part of statistical modelling
that is widely used in various departments such as weather
stations, Finance, banking, healthcare departments such as
covid-19 data analysis because of its benefits in decision-
making. Time series forecasting analysis has several
objectives, namely, forecasting, modelling, and manage.
Forecasting is an element that is important in managing
activities because whether or not an effective decision is
made depends on several factors that influence, although
hidden, when a decision is taken.
The purpose of time series forecasting model is to predict
the upcoming values of certain variables that range with
time using its previous values. Forecasting is related to the
generation of models and methods that can be used to
construct a good forecast. In time series data, the doings of
past events can be used for forecasting because itisexpected
that, in the future, theimpact of the doings of pasteventswill
still occur. The advantages of forecasting can be felt in many
fields, including production, marketing, economics and
finance. Generally, time series research uses linear time
series models, specifically the autoregressive integrated
moving average (ARIMA) model, Prophet and LSTM.
2 Methodology
The method used structured mapping study and structured
literature review conducted by recognizing and interpreting
the clarifying in the literature review in accordance with the
topic time series forecasting raised in this paper. The
univariate time series made up of a single result over a time
period. The multivariate time series made up of more than
one result collected over time. Multivariate time series
analysis examination is more challenging compared to
univariate time series analysis.
Fig -1: Time Series forecasting Methods and Models
Literature review related to the use and development of
time series forecasting models from various studies in
various departments then calculated to find timelessness
and the latest developments from each method used.
3 Forecasting Model
The research methodology was studied to assess the
accuracy of different types of time series models for rainfall,
covid data, real- estate, Bit-coin forecasting. Initially, a
comprehensive literature survey was carried out to study
related research conducted to identify the techniques,
datasets and observations of the different methodologies
implemented worldwide. Often used time series forecasting
models were identified from the literature survey and the
models were developed to forecast the rainfall Bit-coin
forecasting, covid-19 data, real-estate.
3.1 Forecasting with ARIMA
The first phaseinapplying ARIMA model is tocheck whether
the time seriesisstationary or not. AutoregressiveIntegrated
Moving Average works best when data has a fixed design

p-ISSN: 2395-0072
overtime, meaning that the variance and mean of the data
have to remainconstant overtime. Thus, when the data has a
trend of going upwards or downwards and has a particular
pattern (seasonality), then the data is not stationary.[1]
Three parameters constitute the Autoregressive Integrated
Moving Average model (AR, I, MA) that have essence on the
model accuracy: (p, d, q) indicate the autoregressive,
difference order, and moving average window size,
respectively. To identify these parameters, we first apply
differencing lag-1fora moving trend orseasonaldifferencing,
then wefit the ARIMA modelto thevarious series. Theclassic
method for determining these parameters is using
perceptible examinations of the time series to detect trends
as well as looking at the correlation and partial correlation
charts.
The forecasting equation is constructed as follows. First, y
denotes the dth difference of Y, which describes:
If d1=0: yt1 = Yt (1)
If d1=1: yt1 = Yt - Yt-1 (2)
If d1=2: yt1 = (Yt - Yt-1) - (Yt-1 - Yt-2) = Yt - 2Yt-1 +
Yt-2 (3)
The general forecasting model equation is, In terms of y isas
follows:
ŷt1 = μ + ϕ1 yt-1 +…+ ϕp yt-p - θ1et-1 -…- θqet-q (4)
Frequently the parameters are denoted there by AR(1),
AR(2) and MA(1), MA(2),etc. To identify the appropriate
Autoregressive Integrated Moving Average model for Y, we
start by identifying the order of differencing needing to
stationaries the series and cancel the gross features of
seasonality [5]
3.2 Forecasting with Prophet
PROPHET is a software which is open-source that is
available in R and Python for forecasting time series model
data. PROPHET is published by Facebook’steamofCoreData
Science. It is based on a contribution of model where trends
are nonlinear are fit with yearly and weekly holidays and
plus seasonality. PROPHET is well built to missing data,
capturing the shifts in the trend and large outliers. In
summation, it gets a suitable estimate of the mixed data
without spending manual work [7]. PROPHET is effectivefor
business forecast that are observed on Facebook. For
example, time, weekly, daily observations of history,withina
year, missing observation, trend changes, large outliers and
trends that are non-linear growth curves [7]
Prophet parameters consist of changepoints, capacities,
seasonality, holiday and smoothing parameters that can be
interpretablyapplied to improve the model performance.[8]
Prophet can be considered a nonlinear regression model, of
the form:
yt=g(t)+s(t)+h(t)+εt, yt=g(t)+s(t)+h(t)+εt, (1)
describes a piecewise-linear trend (or “growth term”),
s(t)s(t) describes the various seasonal patterns, h(t)h(t)
captures the holiday effects, and εtεt is a white noise error
term.
3.3 Forecasting with LSTM
Extension of recurrent neural network are LongShort–Term
Memory (LSTM), which essentially deepen their memory.
Along these lines it is significant to acquire from vital
encounters that have long situations slacks in the middle.
The units of Long-Short Term Memory networks are usedas
building units for the layers of a recurrent neural network,
which is then often known as a Long Short–Term Memory
network. Long Short–Term Memory allow recurrent neural
networks to recall their inputs over a long period of time.
This is because recurrent neural network contains theirdata
in memory that is much similar to the memory ofacomputer
in the light of the fact LSTM can erase, read and write data
from its memory. [9]
The cell situation in LSTM helps the knowledge to flow
through the units without being modified by allowing only
a few linear exchanges. Each unit has an input, output and
a forget gate which can count or eliminate the information
to the cell state. The forget gate chooses which
information from the previous cell state should be
obliterated for which it uses a sigmoid function. The input
gate rules the information flow to the current cell state
using a point-wise multiplication operation of ‘tanh’ and
‘sigmoid’ respectively. Finally, the output gate determines
which information should be passed on to the next
unrevealed state
4 Literature Survey
A list of research areas in time series forecasting using
ARIMA, Prophet and LSTM. The summary of methods used
in literature is given in following table.

p-ISSN: 2395-0072
Reference Method Observation Dataset
Salehin,
Imrus, et al.
2020 [7]
SARIMA
,
Facebo
ok
Prophet
The result
indicates that
Facebook
Prophet,with
the lowest
Mean Squared
Error (MSE)
and Root Mean
Squared Error
(RMSE), is the
fittest model
to predict the
monthly
rainfall in
Central Jakarta
The datasets
used in the
study consist of
rainfall time
series datasets
from January
1st,2008, to
August 31,
2020
Salehin,
Imrus, et al.
2020 [8]
LSTM,
Neural
Networ
k
They got 76%
accuracy in
LSTMmodel
Rainfall
Dataset in
Dhaka city
from 2000
to 2014
Fente, Dires
Negash, and
Dheeraj
Kumar
Singh 2018
[9]
LSTM In this model
the data is
trained using
LSTM
algorithm.
From
experimental
result, it is
observed that
Long-Short
Term Memory
neural
network gives
substantial
resultswith
high accuracy
among the
other weather
forecasting
techniques.
The historical
weather dataset
is taken from
national climate
datacentre
(NCDC)
from
November
2007 to
October
2017.The
dataset contain
many weather
attributes like
temperature
, humidity,
dew point,
pressure
5 Conclusion
This paper has discussed different Time Series Weather
Prediction techniques like ARIMA, Prophet, LSTM used to
develop data-driven weather forecasting systems. While
there aremany promising DL applications inother fields and
numerous attempts were made to replace the widely used
Numerical Weather Prediction (NWP), there are specific
properties of weather data that require the development of
new approaches beyond theclassical. New concepts derived
Reference Method Observation Dataset
Satrio,
Christophor
us
Beneditto
Aditya, et
al.2021 [1]
ARIMA
and
PROPH
ET
PROPHET has
good accuracy in
predicting the
confirmed cases
with 91%
precision, while
ARIMA did not
even
pass through half
precision. Both
models also have
negative R2 values.
The dataset
of Covid- 19
consists of
27618 rows
and 8
columns.
Alghamdi,
Taghreed,
et al.2019
[2]
ARIMA ARIMA model
achieves excellent
performance with
a confidence level
at 95%.
Dataset
contains
2175
observation
s that been
obtained in
different 3
months and
13
attributes
with
different
data types
Raymond,
Y. C. 1997
[3]
ARIMA Results strongly
show that there
exist cyclical
trends in the office
and industrial
property prices in
Hong Kong.
Real‐estate
prices
Geetha, A.,
and G. M.
Nasira 2016
[4]
ARIMA,
Statistic
al
measur
es
Results obtained
through this model
are well acceptable
with the prediction
accuracy range of
80%.
Rainfall of a
coastal
region, five-
year dataset
(2009-
2013)
consisting of
temperatur
edew point
Yenidoğan,
Işil, et al.
2018 [6]
Prophet While the
PROPHET model
makes predictions
quite close to
reality, that is up
to 94.5% precision,
the ARIMA model
provided only 68%
precision.
The dataset
selected for
this study
starts from
May 2016
and ends in
March 2018,
Bitcoin
value
Table -1: Literature Survey
from computer vision, speech recognition, or DL
solutions for many of these issues are still under
development.

p-ISSN: 2395-0072
REFERENCES
[1] Satrio, Christophorus Beneditto Aditya, et al. "Time
series analysis and forecasting of coronavirus diseasein
Indonesia using ARIMA model and PROPHET." Procedia
Computer Science 179 (2021): 524-532.
[2] Alghamdi, Taghreed, et al. "Forecasting traffic
congestion using ARIMA modeling." 2019 15th
International Wireless Communications & Mobile
Computing Conference (IWCMC). IEEE, 2019.
[3] Raymond, Y. C. "An application of the ARIMA model to
real‐estate prices in Hong Kong." Journal of Property
Finance (1997).
[4] Geetha, A., and G. M. Nasira. "Time-series modelling and
forecasting: Modelling of rainfall prediction using
ARIMA model." International Journal of Society Systems
Science 8.4 (2016): 361-372.
[5] Duke “https://people.duke.edu/~rnau/411arim.html”
[8] Salehin, Imrus, et al. "An artificial intelligence-based
rainfall prediction using LSTM and neural network."
2020 IEEE International Women in Engineering (WIE)
Conference on Electrical and Computer Engineering
(WIECON-ECE). IEEE, 2020.
[9] Fente, Dires Negash, and Dheeraj Kumar Singh.
"Weather forecasting using artificial neural network."
2018 second international conference on inventive
communication and computational technologies
(ICICCT). IEEE, 2018.
[7] Sulasikin, Andi, et al. "Monthly Rainfall Prediction
Using the Facebook Prophet Model for Flood
Mitigation in Central Jakarta." 2021 International
Conference on ICT for Smart Society (ICISS). IEEE,
2021.
[6] Yenidoğan, Işil, et al. "Bitcoin forecasting using ARIMA
and PROPHET." 2018 3rd international conference on
computer science and engineering (UBMK). IEEE,
2018.

Time Series Weather Forecasting Techniques: Literature Survey

More Related Content

Time Series Weather Forecasting Techniques: Literature Survey