Journal Pre-proof

Spatial forecasting of solar radiation using ARIMA model

Ahzam Shadab, Shamshad Ahmad, Saif Said

PII: S2352-9385(20)30273-1
DOI: [Link]
Reference: RSASE 100427

To appear in: Remote Sensing Applications: Society and Environment

Received Date: 2 June 2020

Revised Date: 6 October 2020
Accepted Date: 15 October 2020

Please cite this article as: Shadab, A., Ahmad, S., Said, S., Spatial forecasting of solar radiation
using ARIMA model, Remote Sensing Applications: Society and Environment (2020), doi: https://
[Link]/10.1016/[Link].2020.100427.

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2020 Published by Elsevier B.V.

Spatial Forecasting of Solar Radiation using ARIMA Model

Ahzam Shadaba*, Shamshad Ahmada, Saif Saidb

a
Jamia Millia Islamia, Faculty of Engineering and Technology, Department of Civil
Engineering, New Delhi, India, 110025
b
Aligarh Muslim University (AMU), Department of Civil Engineering, Aligarh, U.P., India,
202002

of
Corresponding Author Name and Email Address
*Ahzam Shadab

ro
ahzamshadab@[Link]

-p
re
lP
na
ur
Jo

1
 Abstract

This paper is an attempt to forecast monthly solar radiation using remote sensing data on a
region. Seasonal ARIMA (SARIMA) models are used for simulating and forecasting time
series of insolation data from NASA’s POWER (Prediction of Worldwide Energy Resources)
data archive. Remotely sensed modelled insolation data for 34 years (i.e. January 1984 to
December 2017) has been retrieved and analysed for forecasting. Monthly average insolation
forecasts of the region around India’s capital Delhi have been generated for next four years
(i.e. January 2018 to December 2021) and presented in the form of contours obtained using
marching square algorithm. The overall accuracy of forecasts in terms of R2 (0.9293), Root
Mean Square Error (0.3529), Mean Absolute Error (0.2659) and Mean Absolute Percentage

of
Error (6.556) was obtained. The ARIMA model forecasted the maximum insolation values in

ro
the months of May (6.52-6.76 KwH/m2/day in year 2018, 6.56-6.8 KwH/m2/day in 2019,
-p
6.6-6.8 KwH/m2/day in 2020 and 6.6-6.84 KwH/m2/day in 2021) and Minimum in the
months of January and December (3.2-3.7 KwH/m2/day in January 2018, 3.28-3.52
re
KwH/m2/day in December 2018). Insolation contours were analysed for identification of
lP

potential regions receiving maximum insolation as well as high average annual values of
insolation for implementing efficient solar power generation projects. Parts of Haryana and
na

Rajasthan region in study area were found most suitable for such projects.
ur
Jo

Keywords: Solar radiation, ARIMA, Spatial forecasting, Prediction, Insolation, Marching

square

2
1 Introduction

Global demand for energy is increasing rapidly in an unprecedented manner due to

population and economic growth. However, conventional source of fossil fuel is limited and

depleting very fast. Besides they leave significant carbon footprints which contribute to

greenhouse effect and global warming (Atique et al. 2019). Such influences and technological

innovations have led to growing use of renewable energy resources. Solar energy is one of

the most promising sustainable energy sources due to its accessibility (Kabir et al. 2018). Due

to growing interest in green energy forecasting solar radiation is currently focus of many

of
researchers as designing, planning and operational management of solar energy systems

ro
depends on accurate solar radiation data (Alsharif et al. 2019). Solar radiation data is
-p
measured on site using specially designed sophisticated instruments which are costly and
re
placed sparsely at different locations across the world (Sengupta et al. 2018).
lP

A major demand therefore lies for predicting solar radiation at different locations where
na

measuring equipment cannot be mounted. With the help of remote sensing data, the
ur

assessment solar energy potentials become easier than before. With remote sensing data we
Jo

can evaluate the contemporary potentials of solar energy with the specified spatial resolution

(Wang and Koch 2010). Additionally, by virtue of remote sensing data we can also map the

spatial distribution of solar energy. As a result, it is interesting to integrate remote sensing

data into the solar energy forecasting model, especially the models focusing on the solar

radiation. However, solar energy, like any other renewable resources, is fundamentally

unpredictable, since solar irradiance depends on other environmental factors such as cloud

cover, humidity, temperature and geographical locations (Hassan et al. 2017). As the amount

of solar radiation is uncertain, it can be modelled as a stochastic time series model (Wu and

Chan 2012). A time series is characterized as a set of data that is sequentially collected over

time. It has the property that neighbouring values are correlated. This tendency is called

3
autocorrelation (Etuk 2012). The time series forecasting approach is useful for predicting

future solar radiation status of a region. The forecasting method analyses the sequence of

historical data in a period of time to establish the forecasting model (Mohd et al. 2009). The

Autoregressive Integrated Moving Average (ARIMA) method has been extensively studied

and used in previous research proven to be effective in the forecasting field (Akincilar et al.

2011; Asamoah et al. 2012; Eni 2015; Etuk 2012; Mohd et al. 2009; Rahman and Hasan

2017). Available literature reveals that ARIMA models have been used extensively for

forecasting different time series data. There have been few attempts to forecast solar radiation

of
data using ARIMA also but they have been limited to short term hourly or daily forecasting

ro
of insolation (Ghofrani and Suherli 2017a; Guermoui et al. 2020; Paulescu et al. 2017). In a
-p
previous study, (Shadab et al. 2019) have summarized some of these recent studies and
re
presented a method for long term monthly prediction of solar radiation utilizing ARIMA
lP

model. However, they have used the seasonal ARIMA model for forecasting monthly
na

insolation data on a single point location. In this paper the aim is to extend the ARIMA

models further for spatial forecasting of monthly average insolation and to find the best
ur

suitable location for solar power projects based on forecasts.

Jo

2 Methodology

2.1 ARIMA model

An Autoregressive Integrated Moving Average (ARIMA) process is a mathematical model

which is used for time series modelling and forecasting. ARIMA model is based on Box

Jenkins approach and composed of Autoregressive (AR) and Moving average (MA) models.

AR model establishes that present time series observation is a linear combination of the

previous time series observations plus some noise term. AR process of the order p or AR(p)

is defined by the equation

=∑ ∅ + (1)

4
Where is forecast error, ∅ = (∅ , ∅ , ∅ , ……∅ ) is the vector of AR model coefficients

and p is a non-negative integer. In case of moving average model present time series value is

a function of past errors. A moving average forecasting model uses lagged values of the

forecast error to improve the current forecast (Mohd et al. 2009). Moving average process of

the order q or MA(q) is defined as

= −∑ (2)

Where ε ~ N(0, σ2) : normally distributed sequence of random white noise having zero mean

and constant variance and θ =( ,

of
, , …… ) is the vector of MA coefficients. The AR

ro
and MA process can be combined together to form autoregressive and moving average

ARMA model. ARMA model of the order p and q or ARMA(p, q) can be written as
-p
=∑ ∅ + −∑ (3)
re
lP

ARMA model can be used for stationary time series only where the condition of stationarity

is satisfied i.e. mean and variance is constant in time. In case of non-stationary time series
na

differencing is required and ARIMA model is used which is generalised form of ARMA
ur

model (Box et al. 2015; Ghofrani and Suherli 2017b; Hassan 2014; Zaharim et al. 2009).
Jo

ARIMA models are therefore established by combining the autoregressive (AR), integration

(I) - indicating to the process of differencing, and the moving average (MA) operations (Box

et al. 2015). Generally the ARIMA model is written as ARIMA (p,d,q), where ‘p’ describes

the auto regressive process order, ‘d’ defines the stationary data order and ‘q’ represents the

order of the moving average process (Shadab et al. 2019). The mathematical formulation for

ARIMA model is as follows

( )
(1 − B) = ∅( )ɛ (4)

5
Where, ‘t’ represents the time indices, ‘B’ is the backshift operator, i.e., BXt = Xt-1. and the

terms ∅ (B) and θ (B) are the Autoregressive and Moving Average operators respectively

which can be expressed as

∅ (B) = 1 − ∅ − ∅ −……….∅ (5)

θ (B) = 1 − θ − θ −……….θ (6)

2.2 Seasonal ARIMA model (SARIMA)

Seasonal ARIMA models are used when data include the seasonality and time series is

of
seasonal. A time series is said to be of seasonal order d if there is propensity for the series to

ro
show periodic behaviour after each interval of time d (Etuk 2012). In a time series,
-p
seasonality refers to predictable and persistent trends over a given time frame. Seasonality is
re
a stable pattern of variability that replicates the time frame, where the number of time periods
lP

is indicated before the pattern replicates again. Seasonality can be interpreted in a dataset by

evaluating time series data’s autocorrelation function (ACF), which should differ
na

significantly from zero (Shadab et al. 2019). The ACF or correlogram for a seasonal sequence
ur

shows a spike at seasonal lag. The monthly insolation data collection for present study
Jo

addresses seasonality with duration of 12 months. Seasonal differencing is required for

seasonal data which is accomplished by evaluating the difference between the current and

corresponding past seasonal steps. Seasonal ARIMA model is designated as

ARIMA(p,d,q)(P,D,Q)s where P, D and Q represents the orders of seasonal autoregression,

differencing and moving average terms respectively and ‘s’ as the number of seasons.

Seasonal ARIMA (p,d,q) (P,D,Q)s model may be written as follows:

ϕ Φ" (B# )(1 − B)$ X = θ& (B)Θ( (B# )E (7)

where, ϕ and θq are nonseasonal AR and MA parameters respectively; ΦP and ΘQ are

seasonal AR and MA parameters respectively; ‘d’ and ‘D’ are non-seasonal and seasonal

6
differences; ‘B’ is the backshift or lag operator, ‘Et’ represents random error with normal

distribution N(0,σ2) and ‘s’ is seasonal lag (for present study s =12 months).

2.3 Box-Jenkins procedure

The Box-Jenkins approach to modelling ARIMA processes was applied for forecasting

insolation. Box-Jenkins modelling involves recognizing a suitable ARIMA model, fitting it to

the data, and then using the fitted model for forecasting. One of the attractive features of the

Box-Jenkins approach to forecasting is that ARIMA processes are a very rich class of

of
possible models and it is usually possible to find a process which provides an adequate

ro
description to the data. The Box-Jenkins modelling procedure is an iterative process as shown

in Fig. 1. Following are the steps involved:

-p
re
Data preparation involves transformation of data using differencing. Stationarity of time
lP

series is checked and if found non stationary, data are differenced until there are no
na

noticeable patterns such as trend or seasonality left in the data. “Differencing” means taking

the difference between consecutive observations, or between observations a year/season

ur

apart. Seasonal differencing is required to remove the seasonality in data.

Jo

Model Selection is based on the transformed and differenced data to try to identify AR and

MA terms of potential ARIMA processes which might provide a good fit to the data.

Parameter estimation means finding the values of the model coefficients on the basis of

maximum likelihood or minimum least square methods which provide the best fit to the data.

Most preferred models are selected on the basis of minimum AIC (Akaike information

criterion)(Akaike 1974) or BIC (Bayesian information criterion)(Schwarz 1978). In this study

minimum AIC value is used for selection of the best model.

AIC = −2 log (maximumlikelyhood) + 2k (8)

7
 KLMN O
<= = −2 >?@ (ABCDAEA>DFG>Hℎ??J) + (9)
O

Where k is independently adjusted number of parameters and n is total number of data points.

Model checking involves testing the adequacy of the model by checking the residuals (error

between actual observation and model prediction). Residuals should be random and bear no

auto correlation. In order to reject the hypothesis that there is no autocorrelation among the

model residuals, the p-value, obtained by performing Ljung-Box test (Box and Pierce 1970)

on the residuals, needs to be more than 0.05 for a 95% significance level. If the model is

of
found to be inadequate, go back to initial step of model selection and try to identify a better

ro
model.
-p
Forecasting is done finally, once the best fit model has been selected, estimated and checked.
re
lP

Postulate a general class of

ARIMA model
na

Identification of the model to

ur

be tentatively entertained
Jo

Parameter estimation in the

No
tentatively entertained model

Diagnostic checking
(Is the model adequate?)

Yes

Use this model for

forecasting

Fig. 1 Box Jenkins methodology for insolation forecasting

8
Accuracy of time series models can be assessed in terms of Root Mean Square Error

(RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean

Absolute Scaled Error (MASE), Akaike Information criterion (AIC) and Bayesian

information criterion (BIC).

2.4 Contour Generation for spatial forecasts

Contour plots are generated for forecasted values in the study area using plotly pakage in R

which utilizes marching square algorithm with linear interpolation. (de Queiroz Neto et al.

of
2016; Maple 2003). Marching square is a very efficient computer graphic algorithm for

ro
generating contour lines for a two-dimensional scalar field represented in a regular grid. Its

-p
approach is similar to 3D marching cube algorithm, therefore its 2D version of 3D marching
re
cube algorithm (Lorensen and Cline 1987). Application of marching square is well known for

generation of contour lines on topographic maps and isobars on weather maps. Algorithm
lP

simply works by marching and processing each cell in the grid independently. The data
na

values of scalar field at the cell corners are analysed and a cell index is calculated for
ur

different cases of contour geometry. Then a pre-built lookup table, keyed on cell index, is
Jo

used to compute the geometry of the cell. The final exact position of the contour is obtained

by linear interpolation along the boundaries of cell using data values at the vertices. (de

Queiroz Neto et al. 2016) have described the complete details of marching square algorithm

including lookup table and the code.

3 Data and study area

Solar Insolation data incident on a horizontal surface, in all sky condition, available from

NASA POWER (Prediction of Worldwide Energy Resource) data archive

([Link] is used in this study. POWER project datasets are solar and

meteorological data sets from NASA research for support of renewable energy, building

9
energy efficiency and agricultural needs. NASA POWER has a repository containing long

time series of various solar and meteorological parameters including insolation. These

parameters have been produced from satellite remote sensing based observations of solar

radiation and meteorological data from various assimilation models. Data is accessible on a

grid of 0.5o latitude x 0.5o longitude spatial resolution. In POWER data, Solar radiation data

for a time span from 1st July, 1983 to 31st December , 2007 are acquired from NASA’s

Global Energy and Water Cycle Experiment : Surface Radiation Budget Project Release 3.0

archive (NASA/GEWEX SRB 3.0; see [Link] &

of
[Link] (Stackhouse Jr et al. 2018) and data for

ro
time span from 1st January , 2008 to within 1st week of near real time are taken from Fast

Longwave And Shortwave Radiative

-p
Flux projects ( FLASHFlux: see
re
[Link] from NASA (Stackhouse Jr et al. 2018). In this study monthly
lP

average insolation data, under all sky condition, has been used for a time duration of 36 years
na

(i. e. From January 1984 to December 2019). Data is segregated in into training and

validation datasets. Monthly Insolation Data of 34 years (January 1984 to December 2017)
ur

acquired from POWER data archive is used for ARIMA models generation for forecasting
Jo

and data from January 2018 to December 2019 is used to validate the performance of the

ARIMA models. The time series of mean monthly insolation data over New Delhi, used by

(Shadab et al. 2019) for generating ARIMA models, from January 1984 to December 2015, is

shown in Fig. 2. In our previous study, the most appropriate ARIMA model was identified as

seasonal ARIMA (1,0,1)(0,1,1)12, based on Box Jenkins methodology, which was used to

simulate and forecast monthly insolation data for a single point location (Shadab et al. 2019).

10
 of
Fig.2 Monthly mean insolation (KwH/m2/day) data at New Delhi from year 1984 to 2015.

ro
3.1 Study Area
-p
For spatial forecast of Insolation values a study area around India’s capital Delhi and nearby
re
locations (covering total area of ∼18904 Square Miles), as seen in Fig. 3, is selected from
lP

NASA Power Access Viewer between following coordinates:

na

Latitude 29.25o N, Longitude 75.75o E Latitude 29.25o N, Longitude 78.75o E

Latitude 27.75o N, Longitude 75.75o E Latitude 27.75o N, Longitude 78.75o E

ur
Jo

The study area in Fig. 3 shows the Google earth imagery of study area and location of

available insolation grid of 0.5º x 0.5ºfrom NASA’s POWER data. In the study area the

NASA POWER grid is having 28 points with 4 rows and 7 columns. Grid points are selected

as point feature class for spatial Insolation forecast analysis. Coordinates of point number 1 at

the bottom left side of study area are latitude 27.75o N, longitude 75.75o E and the coordinates

of other points are according to the grid (see Fig. 3). Google earth image was first

georeferenced and then overlapping of the grid points selected for the study is performed in

Arc GIS v.10.3.

11
 of
Fig.3: Study area selected for spatial insolation forecasts with NASA POWER grid point locations

ro
4 Results and Discussion
-p
re
4.1 Insolation time series seasonal ARIMA models The best fit seasonal ARIMA models are
lP

determined using Box-Jenkins methodology for spatial forecasting of Insolation values at

different grid point locations in study area. It is implemented in R package using [Link]()
na

function which select the best model using minimum AIC value. The accuracy and
ur

performance metrics along with the coefficients of these selected models are shown in Table
Jo

1. It is evident from error metrics that all the selected models are adequate and fits well for

forecasting at their respective point locations on grid. The validation of forecasts is done

using the test dataset for years 2018 and 2019 from NASA POWER data. Fig. 4 shows the

accuracy of insolation forecast values from ARIMA models with respect to actual NASA

POWER data for the validation in year 2018 and 2019. Model shows excellent forecasting

performance with R2 value as 0.9293 along with RMSE, MAE and MAPE values as 0.3524,

0.2659 and 6.5567 respectively.

12
 8
y = 0.9224x + 0.5559
R² = 0.9293
7
Predicted ARIMA Insolation

6 RMSE = 0.3524
MAE = 0.2659
MAPE = 6.5567
(Kwh/m2/Day)

4 Best Fit Line

3 1:1 Line

of
1

ro
0
0 1 2 3 4 5 6 7 8
(Kwh/m2/Day)
-p
Actual NASA POWER Insolation data
re
lP

Fig.4 Validation of monthly insolation forecasts for years 2018 and 2019 in the study area

4.2 Mean monthly Insolation Spatial Forecast contours for selected area
na

The Selected ARIMA models are extended for spatial forecasting of Insolation values at the
ur

selected region for next 4 years (i.e. January 2018 to December 2021). Monthly average
Jo

insolation contour plots for forecasted insolation values were obtained using marching square

algorithm for each month in these years. The maximum and minimum insolation values for

each year were found in the months of January, May and December only. Fig. 5 shows

contour plots for January, May and December along with the average annual insolation

forecast contour for the forecast years 2018-2021. It can be observed from contour plots that

the expected monthly average insolation values are well within the range of 3-7 KwH/m2/day.

The month wise variation trend in Insolation forecast values can be explained by changes in

weather condition throughout the year. In year 2018, insolation values are increasing from

north east part of study area towards southwest direction whereas Insolation forecast values

are found to be maximum in the months of May (6.52-6.76 KwH/m2/day) and lowest values

13
are found to be in the months of January (3.2-3.7 KwH/m2/day) and December (3.28-3.52

KwH/m2/day). Similarly same trends were seen in year 2019 where the maximum value of

insolation forecast were found in the months of May (6.56- 6.8 KwH/m2/day) and minimum

values were found in the months of January (3.3- 3.8 KwH/m2/day) and December (3.32-3.52

KwH/m2/day). Not very significant differences were found in the trends of insolation forecast

in the year 2019 and 2018. In continuation to our observation we have found the forecasts in

the year 2020 and year 2021. For these years, the maximum insolation forecasts were found

in the months of May, 2020 (6.6-6.8 KwH/m2/day) and May 2021 (6.6-6.84 KwH/m2/day)

of
respectively. Minimum forecasts of insolation values were seen in the months of January

ro
2020 (3.3-3.8 KwH/m2/day), December 2020 (3.3-3.56 KwH/m2/day), January 2021 (3.4-3.8
-p
KwH/m2/day) and December 2021(3.3-3.56 KwH/m2/day). It can be seen from various
re
contour plots that the insolation forecast values are increasing from the months of January to
lP

the month of May and then decreasing from May to December as a general trend in each year
na

of forecast. This trend can be attributed to the change in weather conditions throughout the

year and the astronomical seasons. While observing the insolation trend carefully each year it
ur

has been noticed that there is a slight increase in the insolation values in respective months
Jo

each year, which could be possible because of rising levels of UV radiation due to climate

change and ozone layer depletion (Häder et al. 2015). In previous studies researchers have

applied ARIMA and other models of solar radiation forecast to predict solar radiation but

most of them have done short term hourly or day ahead prediction and their application was

limited to a single point forecast (Alsharif et al. 2019; Ghofrani and Suherli 2017a; Guermoui

et al. 2020; Shadab et al. 2019). In this study first time ARIMA models has been extended for

spatial forecasts of insolation. It is also evident from all the plots that Insolation values are

increasing from North-East region towards South West direction in the selected region. The

potential regions identified with maximum values of insolation are situated in the south

14
western area of the study area covering Rajasthan and parts of Haryana state that includes

small villages and towns listed in Table 2. The average annual insolation values are also

found to be highest in these regions with least variation in insolation values in the year and

between different years. Therefore, these regions, with maximum average annual insolation

value are considered as the most suitable locations for setting up of solar power projects in

the study area.

of
Table 2 Region identified with highest Insolation values in study area

ro
Average
Max Min
S. Potential regions Annual
Insolation Insolation Coordinates
No. identified -p Insolation
(KwH/m2/day) 2
(KwH/m /day)
(KwH/m2/day)
Kerwali Village,
re
Neem ka Thana 27.7604° N,
1 6.84 5.14 3.52
Town, District Sikar, 75.7866° E
lP

Rajasthan.
Mandholi Town,
27.7746°N,
2 District Sikar, 6.82 5.12 3.52
na

75.7996°E
Rajasthan.
Singhana Town,
28.0984°N,
ur

3 District Jhunjhunu, 6.76 5.10 3.48

75.8389°E
Rajasthan
Jo

Khetri Nagar Town,

28.0724°N,
4 District Jhunjhunu, 6.78 5.10 3.48
75.8229°E
Rajasthan.
Dabla village,
27.9004°N,
5 District Jhunjhunu, 6.78 5.08 3.48
75.9333°E
Rajasthan.
Banethi Village,
27.7954°N,
6 District Jaipur, 6.76 5.10 3.48
76.1019°E
Rajasthan.
Narnaul city, 28.0688°N,
7 6.74 5.06 3.46
Haryana. 76.1068°E
Village Nangal
Sirohi, District 28.1916°N,
8 6.72 5.06 3.46
Mahendragarh, 76.1291°E
Haryana.

15
 January, 2018

of
ro
-p
re
lP
na

May, 2018
ur
Jo

16
 December, 2018

of
ro
-p
re
lP

January, 2019
na
ur
Jo

17
 May, 2019

of
ro
-p
re
lP
na

December, 2019
ur
Jo

18
 January, 2020

of
ro
-p
re
lP
na

May, 2020
ur
Jo

19
 December, 2020

of
ro
-p
re
lP
na

January, 2021
ur
Jo

20
 May, 2021

of
ro
-p
re
lP
na

December, 2021
ur
Jo

21
 Annual average forecast Insolation

of
ro
-p
re
lP

Fig.5 Monthly average insolation forecast contours in the months of January, May and December and annual
average insolation forecast contour from ARIMA models for the years 2018-2021
na

5 Conclusion
ur

The study assessed the ARIMA models as potential method for predicting the mean monthly
Jo

insolation data over an area of 18904 square miles around India’s capital Delhi and concluded that

seasonal ARIMA models can be used successfully for forecasting spatial time series solar radiation

data. Different seasonal ARIMA models were developed using 34 year’s time series of NASA

POWER insolation data on a spatial grid. The statistical parameters like RMSE, MAE, MAPE,

MASE, AIC and BIC were estimated for testing the validity of the models used in forecasting

insolation time series data at different point locations. These models were found adequate and

extended further for spatial forecasting of insolation data for next 4 years. The overall accuracy of

spatial forecasts was found to be satisfactory with R2 = 0.9293, RMSE = 0.3529, MAE = 0.2659 and

MAPE = 6.556. Spatial Insolation forecasts were obtained in the form of monthly contour plots of

Insolation. After carefully examining the contour plots it has been concluded that the south west part

of the study area was exposed to highest insolation forecast values. Therefore this region comprising

22
of several small villages and towns covering Rajasthan and parts of Haryana state are best suited

for setting up of solar power projects. The outcome of this study is a robust methodology for tapping

solar energy in an effective manner in future. In future work, the method presented can be extended

for spatial solar radiation prediction for longer time duration on a large area. The research can provide

renewable energy project investors and planners with scientific information regarding solar energy

potential and suggestions for solar energy development in the near future. The spatial solar insolation

forecasts results presented in this study for north India region near Delhi would certainly benefit

policy makers to make sound decisions in future for site selection, maintenance and operation,

of
planning of solar parks, and installation of suitable solar power technologies at the best suitable sites.

ro
Ethical statement
-p
All ethical practices have been followed in relation to the development, data analysis,
writing, and publication of this research article.
re
Acknowledgements
lP

The data were obtained from the NASA Langley Research Center (LaRC) POWER Project funded
through the NASA Earth Science/Applied Science Program
na
ur

Disclosure: Authors declare no conflict of interests.

Jo

References

Akaike, H. (1974). A new look at the statistical model identification. IEEE transactions on automatic
control, 19, 716-723
Akincilar, A., Temiz, İ., & Şahin, E. (2011). An application of exchange rate forecasting in Turkey. Gazi
University Journal of Science, 24, 817-828
Alsharif, M.H., Younes, M.K., & Kim, J. (2019). Time series ARIMA model for prediction of daily and
monthly average global solar radiation: the case study of Seoul, South Korea. Symmetry, 11, 240
Asamoah, D., Annan, J., & Arthur, Y.D. (2012). Time Series Analysis of Electricity Meter Supply in
Ghana
Atique, S., Noureen, S., Roy, V., Subburaj, V., Bayne, S., & Macfie, J. (2019). Forecasting of total daily
solar energy generation using ARIMA: A case study. In, 2019 IEEE 9th annual computing and
communication workshop and conference (CCWC) (pp. 0114-0119): IEEE
Box, G.E., Jenkins, G.M., Reinsel, G.C., & Ljung, G.M. (2015). Time series analysis: forecasting and
control. John Wiley & Sons
Box, G.E., & Pierce, D.A. (1970). Distribution of residual autocorrelations in autoregressive-
integrated moving average time series models. Journal of the American statistical Association, 65,
1509-1526

23
de Queiroz Neto, J.F., dos Santos, E.M., & Vidal, C.A. (2016). Mskde-using marching squares to
quickly make high quality crime hotspot maps. In, 2016 29th SIBGRAPI Conference on Graphics,
Patterns and Images (SIBGRAPI) (pp. 305-312): IEEE
Eni, D. (2015). Seasonal ARIMA modeling and forecasting of rainfall in Warri Town, Nigeria. Journal
of Geoscience and Environment Protection, 3, 91
Etuk, E.H. (2012). A multiplicative seasonal ARIMA model for Nigerian unemployment rates. Bulletin
of Society for Mathematical Services and Standards, 1, 57-67
Ghofrani, M., & Suherli, A. (2017a). Time series and renewable energy forecasting. Time series
analysis and applications, 77-92
Ghofrani, M., & Suherli, A. (2017b). Time Series and Renewable Energy Forecasting. Dependability
Engineering: IntechOpen
Guermoui, M., Melgani, F., Gairaa, K., & Mekhalfi, M.L. (2020). A comprehensive review of hybrid
models for solar radiation forecasting. Journal of Cleaner Production, 120357
Häder, D.-P., Williamson, C.E., Wängberg, S.-Å., Rautio, M., Rose, K.C., Gao, K., Helbling, E.W., Sinha,
R.P., & Worrest, R. (2015). Effects of UV radiation on aquatic ecosystems and interactions with other

of
environmental factors. Photochemical & Photobiological Sciences, 14, 108-126
Hassan, J. (2014). ARIMA and regression models for prediction of daily and monthly clearness index.
Renewable energy, 68, 421-427

ro
Hassan, M.Z., Ali, M.E.K., Ali, A.S., & Kumar, J. (2017). Forecasting day-ahead solar radiation using
machine learning approach. In, 2017 4th Asia-Pacific World Congress on Computer Science and
Engineering (APWC on CSE) (pp. 252-258): IEEE
-p
Kabir, E., Kumar, P., Kumar, S., Adelodun, A.A., & Kim, K.-H. (2018). Solar energy: Potential and
re
future prospects. Renewable and Sustainable Energy Reviews, 82, 894-900
Lorensen, W.E., & Cline, H.E. (1987). Marching cubes: A high resolution 3D surface construction
lP

algorithm. ACM siggraph computer graphics, 21, 163-169

Maple, C. (2003). Geometric design and space planning using the marching squares and marching
cube algorithms. In, 2003 international conference on geometric modeling and graphics, 2003.
na

Proceedings (pp. 90-95): IEEE

Mohd, Z.I., Roziah, Z., Marzuki, I., & Muhd, S.L. (2009). Forecasting and time series analysis of air
pollutants in several area of Malaysia. American Journal of Environmental Sciences, 5, 625-632
ur

Paulescu, M., Pop, N., Stefu, N., Paulescu, E., Boata, R., & Calinoiu, D. (2017). Seasonal modeling of
hourly solar irradiation series. Romanian Journal of Physics, 62, 58-71
Jo

Rahman, A., & Hasan, M.M. (2017). Modeling and forecasting of carbon dioxide emissions in
Bangladesh using Autoregressive Integrated Moving Average (ARIMA) models. Open Journal of
Statistics, 7, 560-566
Schwarz, G. (1978). Estimating the dimension of a model. The annals of statistics, 6, 461-464
Sengupta, M., Xie, Y., Lopez, A., Habte, A., Maclaurin, G., & Shelby, J. (2018). The national solar
radiation data base (NSRDB). Renewable and Sustainable Energy Reviews, 89, 51-60
Shadab, A., Said, S., & Ahmad, S. (2019). Box–Jenkins multiplicative ARIMA modeling for prediction
of solar radiation: a case study. International Journal of Energy and Water Resources, 3, 305-318
Stackhouse Jr, P.W., Zhang, T., Westberg, D., Barnett, A.J., Bristow, T., Macpherson, B., Hoell, J.M., &
Hamilton, B.A. (2018). POWER Release 8 (with GIS Applications) Methodology (Data Parameters,
Sources, & Validation) Documentation Date May 1, 2018 (All previous versions are obsolete)(Data
Version 8.0. 1)
Wang, S., & Koch, B. (2010). Determining profits for solar energy with remote sensing data. Energy,
35, 2934-2938
Wu, J., & Chan, C.K. (2012). The prediction of monthly average solar radiation with TDNN and
ARIMA. In, 2012 11th International Conference on Machine Learning and Applications (pp. 469-474):
IEEE
Zaharim, A., Razali, A.M., Gim, T.P., & Sopian, K. (2009). Time series analysis of solar radiation data in
the tropics. European Journal of Scientific Research, 25, 672-678

24
Table 1 Best fit selected seasonal ARIMA models with their coefficients and performance accuracy

Point Model Coefficients Performance error metrics

ARIMA Model
No AR1 AR2 AR3 AR4 MA1 MA2 MA3 SAR1 SMA1 RMSE MAE MAPE MASE AIC BIC
1 ARIMA(1,0,1)(0,1,1)12 0.926 - - - -0.684 - - - -0.863 0.3266 0.2448 4.7846 0.7201 273.54 289.46
2 ARIMA(1,0,1)(0,1,1)12 0.950 - - - -0.751 - - - -0.855 0.3378 0.2466 4.9691 0.7521 299.16 315.08
3 ARIMA(1,0,1)(0,1,1)12 0.950 - - - -0.751 - - - -0.855 0.3378 0.2466 4.9691 0.7521 299.16 315.08
4 ARIMA(1,0,1)(0,1,1)12 0.948 - - - -0.771 - - - -0.827 0.3533 0.2606 5.4491 0.7736 332.97 348.89
5 ARIMA(1,0,1)(0,1,1)12 0.948 - - - -0.771 - - - -0.827 0.3533 0.2606 5.4491 0.7736 332.97 348.89

o f
6 ARIMA(2,0,1)(0,1,1)12 1.085 -0.129 - - -0.818 - - - -0.847 0.3620 0.2645 5.5718 0.7335 355.41 375.32

ro
7 ARIMA(2,0,1)(0,1,1)12 1.085 -0.129 - - -0.818 - - - -0.847 0.3620 0.2645 5.5718 0.7335 355.41 375.32
8 ARIMA(1,0,1)(0,1,1)12 0.941 - - - -0.734 - - - -0.850 0.3330 0.2483 4.9421 0.7303 287.6 303.52

-p
9 ARIMA(3,0,3)(0,1,1)12 0.021 0.322 0.538 - 0.217 -0.116 -0.576 - -0.843 0.3408 0.2583 5.2165 0.7683 313.37 345.22

re
10 ARIMA(3,0,3)(0,1,1)12 0.021 0.322 0.538 - 0.217 -0.116 -0.576 - -0.843 0.3408 0.2583 5.2165 0.7683 313.37 345.22
11 ARIMA(1,0,1)(1,1,1)12 0.899 - - - -0.731 - - 0.098 -0.887 0.3791 0.2802 5.6912 0.7545 395.67 419.56

lP
12 ARIMA(1,0,1)(1,1,1)12 0.899 - - - -0.731 - - 0.098 -0.887 0.3791 0.2802 5.6912 0.7545 395.67 419.56
13 ARIMA(2,0,0)(0,1,1)12 0.231 0.118 - - - - - - -0.865 0.4211 0.3080 6.3707 0.7421 475.32 491.24

na
14 ARIMA(2,0,0)(0,1,1)12 0.231 0.118 - - - - - - -0.865 0.4211 0.3080 6.3707 0.7421 475.32 491.24
15 ARIMA(1,0,1)(0,1,1)12 0.941 - - -0.734 - - - -0.850 0.3330 0.2483 4.9421 0.7303 287.6 303.52
16
17
ARIMA(3,0,3)(0,1,1)12
ARIMA(3,0,3)(0,1,1)12
0.021
0.021
0.322
0.322
0.538
0.538 ur-
-
0.217 -0.116
0.217 -0.116
-0.576
-0.576
-
-
-0.843
-0.843
0.3408
0.3408
0.2583 5.2165 0.7683 313.37
0.2583 5.2165 0.7683 313.37
345.22
345.22
Jo
18 ARIMA(1,0,1)(1,1,1)12 0.899 - - - -0.731 - - 0.098 -0.887 0.3791 0.2802 5.6912 0.7545 395.67 419.56
19 ARIMA(1,0,1)(1,1,1)12 0.899 - - - -0.731 - - 0.098 -0.887 0.3791 0.2802 5.6912 0.7545 395.67 419.56
20 ARIMA(2,0,0)(0,1,1)12 0.231 0.118 - - - - - - -0.865 0.4211 0.3080 6.3707 0.7421 475.32 491.24
21 ARIMA(2,0,0)(0,1,1)12 0.231 0.118 - - - - - - -0.865 0.4211 0.3080 6.3707 0.7421 475.32 491.24
22 ARIMA(4,0,1)(0,1,1)12 0.988 -0.032 -0.155 0.146 0.716 - - - -0.880 0.3210 0.2496 5.1127 0.7356 266.85 294.72
23 ARIMA(1,0,1)(0,1,1)12 0.943 - - - -0.764 - - - -0.860 0.3392 0.2540 5.2299 0.7443 302.93 318.86
24 ARIMA(1,0,1)(0,1,1)12 0.943 - - - -0.764 - - - -0.860 0.3392 0.2540 5.2299 0.7443 302.93 318.86
25 ARIMA(1,0,2)(0,1,1)12 -0.634 - - - 0.844 0.243 - - -0.855 0.4048 0.3023 6.0282 0.7562 445.3 465.21
26 ARIMA(1,0,2)(0,1,1)12 -0.634 - - - 0.844 0.243 - - -0.855 0.4048 0.3023 6.0282 0.7562 445.3 465.21
27 ARIMA(2,0,2)(1,1,1)12 -0.612 -0.288 - - 0.773 0.465 - -0.056 -0.884 0.4110 0.3066 6.0864 0.7120 466.89 498.75
28 ARIMA(2,0,2)(1,1,1)12 -0.612 -0.288 - - 0.773 0.465 - -0.056 -0.884 0.4110 0.3066 6.0864 0.7120 466.89 498.75

25
HIGHLIGHTS

• Time series ARIMA modeling was performed for solar radiation prediction.
• ARIMA models are extended for monthly spatial insolation forecasts.
• Spatial Insolation forecasts were obtained in the form of contour plots of Insolation using
marching square algorithm.
• Best locations for setting up of solar power projects were identified within the study area
based on maximum and average annual insolation.

of
ro
-p
re
lP
na
ur
Jo
Author Statement

Ahzam Shadab: Methodology, Investigation, Software, Validation, Writing- Original draft, Writing-
Reviewing and Editing. Saif Said: Supervision, Visualization, Data curation, Project administration.
Shamshad Ahmad: Supervision, Resources, Conceptualization, Project administration.

of
ro
-p
re
lP
na
ur
Jo
Ethical statement
All ethical practices have been followed in relation to the development, data analysis,
writing, and publication of this research article.

of
ro
-p
re
lP
na
ur
Jo
Conflict of Interest

Authors declare no conflict of interest.

Acknowledgment

The data was obtained from the NASA Langley Research Center (LaRC) POWER Project funded
through the NASA Earth Science/Applied Science Program.

of
ro
-p
re
lP
na
ur
Jo