S
1
Supplementary Information for
“
Oahu as a Case Study for Island Electricity Systems
Relying on Wind and Solar Generation Instead of Imported
Petroleum
Fuel
s
”
Table of Contents:
1.
Comment on Significant Figures
2.
Land Suitable for Development of Solar
Photovoltaics and Onshore Wind
3.
Processing of Electricity Demand Data
4.
Wind,
S
olar, and
D
emand
R
esource
Fi
le
G
eneration
5.
Weighted Average Wind and Solar
Resource Availability
for
Developable Portions of
Oahu
6.
Mean Daily Wind and Solar Resource
Availability for Example Year 2006
7.
Supplementary Analyses of Wind and Solar Resource Availability
8.
Calculation of Current Electricity System Cost per kWh
9.
Impact
of Island Generation Assets on Residential Rates
10.
Yearly Energy Storage of 100% Renewable Electricity Systems with 100% Resource
Adequacy
11.
Dispatch Curves of 100% Renewable Electricity Systems with 100% Resource
Adequacy
12.
Cost and Technology Inputs
13.
Effects of Load Shifting
14.
Effects of Lost Load
15.
Effect of Land Use Reductions on Cost
16.
Project Sunroof Data
17.
Land Use of Technologies
18.
Sensitivity Analysis Relating to Hydrogen Storage
19.
Effect of Land Area on Cost of Least
-
Cost 100% Renewable Electricity Systems with
100% Resource Adequacy
20.
Tables Supporting Plots
21.
Reference
s
S
2
1.
Comment on Significant Figures
T
he significant figures presented
throughout
represent the
values of the
inputs utilized within the
model, rather than empirical results derived from measurements.
T
hese
values
do
not reflect
the precise level of certainty associated with the
costs
, as they are, by nature, subject to
inherent uncertainties. The primary purpose of presenting these inputs with such precision is to
ensure transparency and facilitate reproducibility without
truncating
values obtained from
databases or equations.
Use
of a substantial number of significant figures
did
not fundamentally
impact the integrity or accuracy of the model's
output or the results discussed herein
.
S
3
2.
Land Suitable for Development of Solar PV and Onshore Wind
Figure S1: Land Suitable for Development of Solar P
hotovoltaics
and Onshore Wind
Image S1a shows
the Digital Elevation Model of Oahu based on LiDAR data. Image S1b shows
the slope surface of Oahu. Darker color indicates a steeper slope. Image S1c shows critical
habitats according to Hawaii's Office of Planning and Sustainable Development. The
gridded
spaces in image S1d indicate the available land for development of solar
photovoltaics
and
onshore wind based on restrictions that: 1) the land’s slope is less than 20%
;
2) the land does
not cover a critical habitat
;
and 3) the land is outside of restricted flooding areas.
764 km
2
of
Oahu is suitable for siting solar panels and on
-
shore wind projects.
It is possible that the excluded areas with slopes > 20% would be sites with high wind
resources. However, installation of a turbine at a steep slope is difficult due to construction
challenges and foundation stability concerns
[1]
. As such, limiting the construction of turbines to
land with a slope smaller than some grade, such as 20%, is common
[2]
.
S
4
3.
Processing of Electricity Demand Data
Figure
S
2: Scaling and Imputation of Electricity Demand Data
Figure S2 shows the electricity demand data before and after processing
for the example year
2008
. Processing included both scaling by a factor of (
1614.5 MW / 2168.1 MW) to account for
Oahu
-
specific electricity demand
and imputation of outlier demand values. Outlier demand
values can be seen in late December
in the
‘Before Processing’ line of Figure S2.
Table S1: Regression Imputation of Outlier Demand Values
Outlier
Index*
Original
Value
Imputed
Value
Outlier
Index*
Original
Value
Imputed
Value
273
0
716.06966
26178
32.68318
686.48546
6895
0
708.50716
26179
26.87482
686.48432
6896
0
708.50601
26180
0
686.48318
6897
0
708.50487
26181
0
686.48204
6898
0
708.50373
26182
0
686.48090
6899
3.723306
708.50259
26183
13.55283
686.47975
6900
10.42526
708.50145
26184
24.55148
686.47861
6901
21.59517
708.50031
26185
29.18327
686.47747
S
5
6902
66.27485
708.49916
26186
81.66699
686.47633
6903
113.9332
708.49802
26187
123.1893
686.47519
6904
181.6973
708.49688
26188
212.1242
686.47404
*
Indicates
the index value in the demand data
set for each
outlier. The index starts at a value of
0, which corresponds to the first value of the data
set: hour 1 on January 1
st
, 2006.
S
6
4.
Wind,
S
olar, and
D
emand
R
esource
F
ile
G
eneration
Hourly
averaged
demand data
were
retrieved from the Federal Energy Regulatory
Commission (FERC). Specifically, electronic filing data from FERC’s Form No. 714
were
obtained
for
2006
-
2020, the years for which electronic filing data
are
available
[3]
. This hourly
averaged
demand represents the demand of all customers of Hawaiian Electric across the state
of Hawaii. To
estimate
the demand for
only
Oahu, the extracted demand values
were multiplied
by the total firm generation capacity of Oahu divided by the total firm generation capacity of all
of Hawaiian Electric’s assets statewide, as reported in their 2022 Power Facts document
[4]
.
R
eported electricity demand data commonly
have
missing
values
or otherwise identified outliers
[5]
. After formulating demand for Oahu, outliers in the data
were identified
by computing the first
quartile (Q1), third quartile (Q3) and interquartile range (IQR) of the data values,
and
then
assigning
as
outliers data points that were less than Q1
-
1.5
[IQR] or greater than Q3 + 1.5
[IQR].
Twenty
-
two
values were identified as outliers o
ut of the 122,712 values in
the
dataset
.
Outliers are commonly observed in
hourly electricity demand datasets and are
attributed to
disruption
s
(outages, technical malfunctions,
maintenance activities, etc.) and
/or
errors
(malfunctioning
recording
equipment, software issues, etc.)
[6]
.
R
egression imputation
was then
performed on
these values
to produce the data set that was
used for further
analysis
. Outlier
index numbers, original values, and imputed values are shown in Table S1.
Hourly wind data
were
retrieved from NREL’s
2023 National Offshore Wind data set
(NOW
-
23)
[7]
. Full datasets for Hawaii were downloaded from the Open Energy Data Initiative.
The U.S. Wind Turbine Database
[8]
lists that Oahu has 12 turbines with an 80 meter hub
height, 30 turbines with a 99.5 meter hub height, and 8 turbines with a 105 meter hub height.
W
ind speed values
were thus extracted from
the
NOW
-
23
calculated at a hub height of 100
meters.
Each one
-
year
NOW
-
23
data set contains a total of 306,112 points in a 2 km
-
by
-
2 km
grid of Hawaii and its surrounding waters. 478 points covering Oahu were extracted from each
yearly dataset for use in calculating wind
resource availability
.
Wind speed values in each data
set are provided in 5
-
minute intervals. H
ourly
wind speed values
for each point for each year
w
ere
determined by averaging sets of 12 values (1
-
12, 13
-
24, 25
-
36, and so on) to produce
hourly averaged values
from the 5
-
minute values
.
Files containing hourly averaged
wind
speeds
for one year at one point were added together for that point to generate a file with complete
hourly
wind
speeds
for that point for the 2006
-
2019 timeframe.
S
7
Then
, to calculate
wind resource availability
from the wind speed values,
each hourly
value was processed according to a piecewise function describing a normalized power curve
described by the function
f(x)
:
푓
(
푥
)
=
{
0
푥
3
12
3
1
0
푥
<
3
3
≤
푥
<
12
12
≤
푥
≤
25
푥
>
25
where values in each condition have units of m/s
[9,10]
.
The piecewise function is
plotted in
Figure S
3
.
Figure S
3
:
Power Curve of Wind Resource Availability
The
horizontal dashed line at x = 3 represents the cut
-
in speed, the horizontal dashed line at x =
12 represents the rated speed, and the horizontal dashed line at x = 25 represents the cut
-
out
speed.
To determine the weighting factor for each point,
the
478 points covering Oahu were
overlaid onto the shapefile of Oahu
with an area value centered around the point (Figure S
4
a).
The percentage of each point’s area value that overlapped with land on Oahu suitable for
development was used to determine a weighting factor for each point (Figure S
4
b). The
hourly
wind
resource availability
for each point w
as
then multiplied by the point’s weighting factor
, and
the
wind resource availability
per hour from 2006 through the end of 2019 w
as
averaged to
generate one file with the weighted average hourly wind
resource availability
over the full
timeframe for the developable portions of Oahu.
S
8
Hourly solar data were retrieved from the National Solar Radiation Database (NSRDB)
[11]
. Data were collected for points in a grid covering Oahu and its surrounding waters. The
resolution of this grid was 4 km by 4 km, representing the most detailed grid available for the
dataset from 2006 to 2019. Solar
resource availability
for each value at each point w
as
calculated by dividing the global horizontal irradiance (GHI) by 1,000 W/m
2
, an industry standard
used to rate current solar cells and modules
[12]
. The data files contained GHI values for one
year, so once hourly
resource availability
w
as
calculated, the files for each point from 2006
through 2019 were added to one master file of hourly
resource availability
for one point for the
full timeframe.
To generate one file with weighted average hourly solar
resource availability
, t
he
NSRDB points
were
subjected to
the same
weighting
process
that was
used for
the
wind
resource data
points (
Figures
S
4
c
and S
4
d).
Hourly w
ind
resource availability
for CONUS (used in Figure 1 and Figure S
8
) w
as
retrieved from the GitHub repository for Tong et al.
[13]
. That source calculated capacity factors
from the Modern
-
Era Retrospective analysis for Research and Application, Version
-
2 (MERRA
-
2) reanalysis satellite weather data
[14]
. The article’s methodology was as follows:
To calculate the wind and solar
[resource availability]
for this study, we first obtained the
hourly climatology data from the Modern
-
Era Retrospective analysis for Research and
Application, Version
-
2 (MERRA
-
2) reanalysis product, which spans 39 years (1980
–
2018) and has a horizontal resolution of 0.5° by lati
tude [−90
–
90°] and 0.625° by
longitude [−180
–
179.375°] with 361 × 576 grid cells worldwide. Here we used the surface
incoming shortwave flux [W m
−
2
] (variable name: SWGDN), top
-
of
-
atmosphere incoming
shortwave flux [W m
−
2
] (variable name: SWTDN), and surface air temperature [K]
(variable name: T) for deriving solar
[resource availability]
; and wind speed at 100 m
[m s
−
1
], estimated based on wind speed at 10 m and 50 m (variable names: U10M,
V10M, U50M, and V50M) and a power
-
law relationship, to derive wind
[resource
availability ]
.
[
The w
ind resource availability was calculated as described in the
piecewise function plotted in Figure S
3
]
Hourly s
olar
resource availability
for CONUS (used in Figure 1 and Figure S
6
) w
as
retrieved from the GitHub repository for Shaner et al.
[12]
. That source calculated
resource
availability
from the Modern
-
Era Retrospective analysis for Research and Application, Version
-
2
(MERRA
-
2) reanalysis satellite weather data. The article’s methodology was as follows:
Time
-
averaged hourly resource data were taken from the NASA developed MERRA
-
2
reanalysis product, which spans 36 years (1980
–
2015) and has a resolution of 0.5°
S
9
latitude by 0.625° longitude. We used the surface incoming shortwave flux [W m
−
2
]
(variable name: SWGDN) and eastward wind and northward wind at 50 m [m s
−
1
]
(variable names: U50M, V50M)...Each raw data point (an hourly energy density (solar)
or wind speed (wind) value at a specific location and time) was converted into a
[resource availability]
based on a pre
-
determined power capacity rating for solar and
wind generators. The
[resource availability]
describes the actual energy output as
compared to the systems* rated energy output (power capacity multiplied by one hour).
For solar, the raw data were divided by 1000 W m
−
2
, which is the industry standard used
to rate current solar cells and modules.
For use in Figure S1
9
, e
lectricity demand data
and
w
ind and solar
resource availability
for CONUS
, the Western Electricity Coordinating Council (WECC), and California
were obtained
from the
GitHub
repository for
Rinaldi et al.
[15]
.
The electricity demand data from this source
was also used in Figure 1.
Th
is
source
calculate
d
resource availability
from the
Modern
-
Era
Retrospective analysis for Research and Application, Version
-
2 (MERRA
-
2) reanalysis satellite
weather data
.
This
article’s methodology was as follows:
To calculate solar
[resource availability]
, we determine the solar zenith angle and
incidence angle based on the location and local hour and then estimate the in
-
panel
radiation. We use an empirical piecewise model that takes into account both ratios of
surface to top
-
of
-
atmosphere solar radiation
(the clearness index) and the local time to
separate the direct and diffuse solar components. We assume a horizontal single
-
axis
tracking system, with a tilt of 0° and a maximum tuning angle of 45°. The use of a
tracking system minimizes variability, as c
ompared to flat plate solar panels, and
moreover improves solar availability. Our use of single
-
axis trackers primarily excludes
rooftop solar installations from this study but produces less variability and increased
potential for solar electricity generat
ion. We use a performance model, which considers
both the surrounding temperature and the effect of irradiance, to calculate the power
output from a given panel.
[
The
Wind resource
availability was calculated as described in the piecewise function
plotted in Figure S
3
]
The solar and wind
[resource availability]
are estimated with the same resolution as
MERRA
-
2 for each grid cell in CONUS, WECC including CA, WECC excluding CA, and
CA. ... We then selected grid cells over land where the annual mean
[resource
availability]
is larger than a threshold value of 26% for both solar and wind. The resulting
S
10
average
[resource availability]
over the 39
-
year period were similar to those reported for
utility scale generation of wind and solar in the U.S.
[15]
.
S
11
5.
Weighted Average Wind and Solar
Resource Availability
for
Developable Portions
of Oahu
:
Figure S4: Weighted Average Wind and Solar Resource Availability For Developable
Portions of Oahu
Image S4a shows the extracted NOW
-
23 data points overlaid onto a shapefile of Oahu. Each
point was fitted at the center of a 2 km by 2 km square. Image S4b shows the overlap of each
square with developable portions of Oahu. Green areas indicate developable
land. The amount
of developable land in each square was divided by the total developable land to determine the
weighting of each point in the extracted NOW
-
23 data. Image S4c shows the extracted NSRDB
data overlaid onto a shapefile of Oahu. Each point was
fitted at the center of a 4 km by 4 km
square. Image S4d shows the overlap of each square with developable portions of Oahu. The
weighting for each point in the same manner was determined using the same process that was
used for the NOW
-
23 points. In both
cases, some points received a weighting of zero and were
excluded from further calculation.
S
12
6.
Mean Daily Wind and Solar Resource Availability for Example Year 200
6
Figure S5: Mean Daily Wind and Solar Resource Availability for Example Year 2006
Figure S5 shows the mean daily wind and solar resource availability for the example year 2006.
The value for the resource availability o
n
any given day is the mean of the 24 hourly
-
resource
availability values for that day.