CHAPTER II
LITERATURE REVIEW
Time series forecasting models are applicable in various areas of daily life,
including finance, agriculture, marketing, health, energy, and meteorology (Tugal et
al., 2023). Time series analysis and forecasting are major tools used by scientists in
meteorology and environmental fields to study meteorological phenomena such as
rainfall, temperature, and humidity (Hafeez et al.,2022). Prediction has consistently
been a prominent topic in academic circles and a crucial factor in practical
applications. Determining the appropriate forecasting method has become a central
challenge in the field of forecasting. The ARIMA model and the exponential
smoothing method are classical and widely used techniques for analyzing time series
(Liu,2022). Globally, numerous researchers have made efforts to accurately forecast
temperature using various techniques. However, due to the nonlinear nature of
weather, the prediction accuracy achieved by these techniques remains unsatisfactory
(Saxena et al.,2013). In this chapter, a literature review of previous research papers
and articles relevant to temperature prediction was provided.
2.1 Temperature
Temperature is a fundamental physical quantity that indicates the degree of
hotness or coldness of an object or environment. It is a crucial parameter in various
scientific disciplines, including meteorology, where it significantly influences weather
patterns and climatic conditions. Accurate measurement of temperature is essential for
predicting weather, studying climate change, and understanding various
environmental phenomena (Bolton, 2020). Temperature is measured using different
scales, with Celsius (°C), Fahrenheit (°F), and Kelvin (K) being the most commonly
used. The Celsius scale, also known as the centigrade scale, is widely used in
scientific research and most countries around the world. The Fahrenheit scale is
primarily used in the United States, while the Kelvin scale is used in scientific
contexts, particularly in physics, as it begins at absolute zero, the theoretically lowest
possible temperature (Bolton, 2020). Various instruments are used to measure
temperature, including thermometers, thermocouples, and infrared sensors.
Traditional liquid-in-glass thermometers, which use mercury or alcohol, have largely
been replaced by digital thermometers due to their higher accuracy and safety. In
meteorology, temperature data is collected from weather stations, satellites, and
remote sensing technologies, providing valuable information for weather forecasting
and climate monitoring (Smith & Jones, 2019).
2.2 Effects on Temperature Prediction
Temperature prediction is a critical component in various fields and has
significant impacts on both human activities and natural ecosystems. The ability to
accurately predict temperature is fundamental to a wide range of human activities and
environmental processes. From ensuring food security and protecting public health to
managing energy resources and understanding climate change, temperature prediction
plays a vital role in fostering sustainable development and resilience against
environmental challenges. As such, ongoing research and advancements in
temperature modeling and forecasting are essential for the well-being of societies and
ecosystems worldwide.
2.3 Some Empirical Reviews of Holt-Winters Model
Gundalia (2012) predicted maximum and minimum temperatures using the
Holt-Winters method. In this paper, the Holt-Winters method was proposed to predict
the maximum and minimum daily temperature time series of the Junagadh region
using Excel spreadsheets. The performance of the method was measured by three
standard error metrics: MSE, MAPE, and MAD, all of which were found to be well
within acceptable limits. The results showed that the maximum temperature time
series exhibited less fluctuation and provided better results compared to the minimum
temperature time series.
Heydari et al. (2019) studied Holt-Winters models to forecast monthly climatic
data, including temperature, precipitation, and sunshine hours, for one year in
advance. As a case study, monthly environmental data (i.e., precipitation, maximum
temperature, minimum temperature, and sunshine hours) collected over 30 years
(from 1981 to 2010) from Robat Gharah-Bil Station located in Golestan, Iran, were
used. After modeling the data through additive and multiplicative procedures, the
three main smoothing parameters of the model were optimized using a nonlinear
optimization method. Based on this study, using the multiplicative form of the Holt-
Winters time series resulted in an overall 4% lower mean absolute percentage error
(MAPE) compared to the additive form. The results showed that this model is more
efficient in predicting and modeling climate parameters, which exhibit stable patterns
of cycles and seasonality.
2.4 Some Empirical Reviews of SARIMA Model
Chang et al. (2012) applied Seasonal Autoregressive Integrated Moving
Average (SARIMA) model to time series precipitation data from 1961 to 2011 to
forecast monthly precipitation in Yantai, China. The auhors found that the Seasonal
ARIMA (1, 0, 1) (0, 1, 1)12 model fitted the historical data well and could be used
successfully for forecasting.
Dwivedi, Sharma, and Wandre (2017) developed a Box-Jenkins Seasonal
Autoregressive Integrated Moving Average (SARIMA) model using temperature data
recorded from 1984 to 2015 for Junagadh city. The data were obtained from the Agro-
meteorology department at Junagadh Agricultural University, located in Junagadh
city, Gujarat. The model with the lowest Akaike Information Criterion (AIC) value
was selected as the appropriate model for forecasting mean temperature. SARIMA (1,
0, 1) (1, 1, 1)12 was selected as the best model for mean temperature forecasting.
Akinbobola et al. (2018) investigated the statistical modeling of monthly
rainfall in selected stations in the forest and savannah eco-climatic regions of Nigeria
using Autoregressive Integrated Moving Average (ARIMA) and Seasonal
Autoregressive Integrated Moving Average (SARIMA) models. The results showed
that the models fitted the data well and successfully modeled the stochastic seasonal
fluctuations. They concluded that the SARIMA model was a suitable method for
modeling and predicting monthly rainfall. The results are useful for forecasting the
pattern of rainfall in the study area and provide information that would help decision-
makers formulate policies to mitigate the problems of water resources management,
soil erosion, flooding, and drought.
Hafeez et al. (2022) studied the monthly average maximum surface air
temperature in Multan over a 10-year period from January 2012 to December 2021.
The authors employed the Box-Jenkins Seasonal Autoregressive Integrated Moving
Average (SARIMA) model, using data from January 2012 to December 2019 as the
training dataset and data from January 2020 to December 2021 as the test dataset to
evaluate the model's performance. Based on the Akaike Information Criterion (AIC)
and Bayesian Information Criterion (BIC), they found two competing SARIMA
models: SARIMA (1,0,0) (1,1,1)12 and SARIMA (1,0,1) (1,1,1)12. Among these,
SARIMA (1,0,0) (1,1,1)12 was selected as the best forecasting model based on
forecasting accuracy measures such as RMSE, MAE, and MAPE.
Upul et al. (2022) analyzed the monthly average rainfall and temperature from
January 2005 to December 2021 in Minot, ND, USA, to develop a seasonal ARIMA
model. Among the competing tentative models, SARIMA (2, 0, 0) (2, 0, 1) 12 and
SARIMA (1, 0, 1) (2, 0, 1)12 were found to be the best time series forecasting models
that captured the existing patterns of the rainfall and temperature data, respectively.
Tugal et al. (2023) researched average daily temperature data from the
province of Mus, Turkey, between 2010 and 2022, covering a total of 4,748 days. The
data were obtained from the Mus Provincial Directorate of Meteorology. Initially, they
used 4,523 days of data for training and 225 days (4.74%) for testing. Later, they used
4,162 days of data for training and 586 days (12.34%) for testing. In this study, three
different models were used to forecast future temperature values: ARIMA,
PROPHET, and LSTM. Each model was chosen for its unique strengths and
weaknesses: ARIMA excels at short-term forecasting, PROPHET is effective for
forecasting trends and seasonality, and LSTM is capable of capturing long-term
dependencies. By utilizing all three models, a more comprehensive forecast of future
temperature values was achieved. The study also found that the LSTM model was the
most accurate for temperature forecasting in the province of Mus.
Jafarian et al. (2024) predicted monthly temperature using data from January
1951 to December 2008. The researchers employed the Box-Jenkins time series
SARIMA model and the ANN-PSO model. A comparison was made between the
SARIMA (0, 0, 1) (0, 1, 1)12 model and the ANN-PSO model. Ultimately, SARIMA
(0, 0, 1) (0, 1, 1)12 was selected for its ability to capture the autocorrelative structure in
the temperature data and provided the most reliable forecasts within the reported
uncertainty.
2.5 Some Empirical Reviews Comparing the Holt-Winters and SARIMA
Models
Xiangpeng (2017) described three different models—Triple Exponential
Smoothing, ARIMA, and SARIMA—using 29 years of monthly maximum
temperature data from Duluth, covering 1986 to 2015. The author then predicted the
temperature for 2016 using these models and compared the results with the actual
temperatures of 2016. Among these models, the Seasonal ARIMA (2, 0, 1) (0, 1, 1) 12
was found to be the best prediction model for this case study. Furthermore, many
researchers compared between Seasonal ARIMA (SARIMA) and Holt-Winters (HW)
in other fields.
Maurice et.al (2013) tried to examine the most appropriate short-term
forecasting method for Ghana’s inflation. Monthly inflation data spanning from
January 1971 to October 2012 was obtained from the Ghana Statistical Service. The
data was divided into two sets: the first set was used for modeling and forecasting,
while the second was used as a test set. Seasonal ARIMA and Holt-Winters
approaches were used to achieve short-term out-of-sample forecasts. The accuracy of
the out-of-sample forecasts was measured using MAE, RMSE, MAPE, and MASE.
Empirical results from the study indicated that the Seasonal-ARIMA forecast from
ARIMA (2,1,1) (0,0,1)12 recorded MAE, RMSE, MAPE and MASE of 0.1787,
0.2104, 1.9123 and 0.0073 respectively; that of the Seasonal Additive HW was
1.8329, 2.0176, 19.996, 0.0745; and the Seasonal Multiplicative HW forecast
recorded 2.2305, 2.4274, 24.000, 0.0911 respectively. Based on these results, the
authors concluded that the Seasonal ARIMA process is the most appropriate short-
term forecasting method for Ghana's inflation.
Makatjane et al. (2016) compared the Seasonal ARIMA and Holt-Winters
models to predict monthly car sales in South Africa using data from January 1994 to
December 2013. The purpose of this study was to choose the optimal model suited for
the sector. The error metrics—mean absolute error (MAE), mean absolute percentage
error (MAPE), and root mean square error (RMSE)—were used for this evaluation.
The findings showed that Holt-Winters triple exponential smoothing was more
effective compared to SARIMA.
Wael and Bahar (2020) analyzed the number of tourists visiting Turkey using
two models: Holt-Winters and Seasonal ARIMA. The study used monthly data of
foreigners visiting Turkey from 2007 to 2018. The authors analyzed the Holt-Winters
models (both multiplicative and additive) and the SARIMA model to achieve a high
level of forecasting accuracy. The multiplicative Holt-Winters (HW) model
demonstrated superior performance compared to both the additive HW and SARIMA
models based on the error measures.
Alemu et al. (2023) presented a study on forecasting monthly ground-level
ozone by comparing the Holt-Winters model and the SARIMA model. The authors
used data from the period 2007–2019 for model formulation and parameter
estimation, and one year of data from 2020 to test model forecasts. More than 60
SARIMA models were generated for the time series. The goodness of fit of these
models was assessed using the Akaike Information Criterion (AIC) and Bayesian
Information Criterion (BIC). Model evaluation metrics, such as root mean square
error (RMSE), mean absolute error (MAE), and mean absolute percentage error
(MAPE), were applied to compare the accuracy of the SARIMA and Holt-Winters
models. Their findings show that the best model for forecasting surface-level ozone is
ARIMA (2,0,4) (1,1,2)12 for the Addis Ababa station and ARIMA (3,1,0) (2,0,0) 12 for
the Ras Dashen station. However, for the Danakil Depression and Bahir Dar stations,
the Holt-Winters model with parameters α = 0.346, β = 0.023, γ = 0.36 and α = 0.302,
β = 0.019, γ = 0.266, respectively, were found to be better than the SARIMA model.
The maximum MAPE values were found to be less than 7.86%, indicating that all the
forecasts are acceptable. The study emphasized that the appropriate model depends on
the specific data examined. Therefore, it is important to investigate multiple
forecasting methods to determine the most suitable one for the data, ensuring a high
level of forecasting accuracy.
