Temperature trends in Hawaiʻi: A century of change, 1917–2016

Abstract

Based on a revised and extended multi‐station Hawaiʻi Temperature Index (HTI), the mean air temperature in the Hawaiian Islands has warmed significantly at 0.052°C/decade (p < 0.01) over the past 100 years (1917–2016). The year 2016 was the warmest year on record at 0.924°C above the 100‐year mean (0.202°C). During each of the last four decades, mean state‐wide positive air temperature anomalies were greater than those of any of the previous decades. Significant warming trends for the last 100 years are evident at low‐ (0.056°C/decade, p < 0.001) and high‐elevations (0.047°C/decade, p < 0.01). Warming in Hawai‘i is largely attributed to significant increases in minimum temperature (0.072°C/decade, p < 0.001) resulting in a corresponding downward trend in diurnal temperature range (−0.055°C/decade, p < 0.001) over the 100‐year period. Significant positive correlations were found between HTI, the Pacific Decadal Oscillation, and the Multivariate ENSO Index, indicating that natural climate variability has a significant impact on temperature in Hawaiʻi. Analysis of surface air temperatures from NCEP/NCAR reanalysis data for the region of Hawaiʻi over the last 69 years (1948–2016) and a mean atmospheric layer temperature time series calculated from radiosonde‐measured thickness (distance between constant pressure surfaces) data over the last 40 years (1977–2016) give results consistent with the HTI. Finally, we compare temperature trends for Hawaii's highest elevation station, Mauna Loa Observatory (3,397 m), to those on another mountainous subtropical island station in the Atlantic, Mt. Izaña Observatory (2,373 m), Tenerife, Canary Islands. Both stations sit above the local temperature inversion layer and have virtually identical significant warming trends of 0.19°C/decade (p < 0.001) between 1955 and 2016.

Full text

RESEARCH ARTICLE
Temperature trends in Hawaiʻi: A century of change, 1917–2016
Marie M. McKenzie | Thomas W. Giambelluca | Henry F. Diaz
Department of Geography and Environment,
University of HawaiʻiatM

anoa, Honolulu, Hawaii
Correspondence
Marie M. McKenzie, Department of Geography
and Environment, University of HawaiʻiatM

anoa,
2424 Maile Way, Saunders Hall 445, Honolulu, HI
96822.
Email: mariemm@hawaii.edu
Funding information
University of Hawai‘i at Hilo
Based on a revised and extended multi-station Hawaiʻi Temperature Index (HTI),
the mean air temperature in the Hawaiian Islands has warmed significantly at
0.052C/decade (p< 0.01) over the past 100 years (1917–2016). The year 2016
was the warmest year on record at 0.924C above the 100-year mean (0.202C).
During each of the last four decades, mean state-wide positive air temperature
anomalies were greater than those of any of the previous decades. Significant
warming trends for the last 100 years are evident at low- (0.056C/decade,
p< 0.001) and high-elevations (0.047C/decade, p< 0.01). Warming in Hawai‘i
is largely attributed to significant increases in minimum temperature (0.072C/
decade, p< 0.001) resulting in a corresponding downward trend in diurnal temper-
ature range (−0.055C/decade, p< 0.001) over the 100-year period. Significant
positive correlations were found between HTI, the Pacific Decadal Oscillation, and
the Multivariate ENSO Index, indicating that natural climate variability has a sig-
nificant impact on temperature in Hawaiʻi. Analysis of surface air temperatures
from NCEP/NCAR reanalysis data for the region of Hawaiʻi over the last 69 years
(1948–2016) and a mean atmospheric layer temperature time series calculated from
radiosonde-measured thickness (distance between constant pressure surfaces) data
over the last 40 years (1977–2016) give results consistent with the HTI. Finally,
we compare temperature trends for Hawaii's highest elevation station, Mauna Loa
Observatory (3,397 m), to those on another mountainous subtropical island station
in the Atlantic, Mt. Izaña Observatory (2,373 m), Tenerife, Canary Islands. Both
stations sit above the local temperature inversion layer and have virtually identical
significant warming trends of 0.19C/decade (p< 0.001) between 1955 and 2016.
KEYWORDS
climate change, El Niño-southern oscillation, Hawaiʻi, Pacific decadal
oscillation, radiosonde observations, temperature trends
1|INTRODUCTION
The Pacific Ocean region plays an important role in ongoing
global air and sea surface temperature changes. The Hawai-
ian Islands occupy a unique geographic position in the
North-Central Pacific and possess a dense, long-term histori-
cal climate data record at a range of elevations spanning a
mid-tropospheric temperature inversion produced by the des-
cending arm of the Hadley cell. As such, Hawai‘i is geo-
graphically well positioned to provide observations that
enhance understanding of regional and global temperature
trends, both at the surface and higher elevations in the atmo-
sphere. In addition to yielding insights for regional and
global climate, temperature change in the Hawaiian archipel-
ago can have critically important impacts on terrestrial eco-
systems, water supply, agriculture, and economic health
(Keener et al., 2012).
The Hawaiian temperature record also provides an
opportunity to examine the effects of large sources of inter-
nal climate variability—the El Niño–Southern Oscillation
(ENSO) and the Pacific Decadal Oscillation (PDO; Mantua
et al., 1997). Hawai‘i is close to the centres of action of
Received: 16 March 2018 Revised: 9 February 2019 Accepted: 19 February 2019 Published on: 28 March 2019
DOI: 10.1002/joc.6053
Int J Climatol. 2019;39:3987–4001. wileyonlinelibrary.com/journal/joc © 2019 Royal Meteorological Society 3987

ENSO, the strongest driver of natural global climate variabil-
ity on time scales of two to seven years (Salinger, 2005).
Whether PDO is a dynamic mode of climate variability inde-
pendent of ENSO has been questioned (Schneider and Cor-
nuelle, 2005). Nevertheless, the PDO Index (Zhang et al.,
1997; Mantua and Hare, 2002) provides a convenient metric
for basin-wide climate variation at the decadal scale, while
an ENSO index such as the Multivariate ENSO Index (MEI)
provides a statistic representing interannual variations. PDO
phase changes have been associated with oceanic circulation
changes linked to variations in the rate of global temperature
increase, including the recently ended period of slow global
warming (Dai et al., 2015).
In support of a state-wide surface air temperature analy-
sis, Giambelluca et al. (2008) ; hereafter referred to as GDL,
developed the Hawaiʻi Temperature Index (HTI) to evaluate
mean (T
mean
), minimum (T
min
), and maximum (T
max
) tem-
perature trends in the islands. Observations from 21 stations
across five of the main Hawaiian Islands were included in
the index, calculated as the mean of monthly station anoma-
lies, averaged to annual values. GDL showed increases in
T
mean
,T
min
, and T
max
at rates of 0.043, 0.094, and
0.005C/decade state-wide, respectively, over the 88-year
period, 1919–2006. The more rapidly rising T
min
and the
consequent reduction in the diurnal temperature range
(DTR) were key findings of GDL. In the more recent
decades of their study (1975–2006), mean warming rates
were much steeper state-wide (0.164C/decade), approach-
ing the global mean trend of 0.177C/decade (1976–2005;
Pachauri et al., 2014), with enhanced warming at high-
elevation stations (0.268C/decade). Mountain temperatures
in Hawaiʻi vary over short distances due to factors, such as
the vertical structure of the atmosphere and the interaction of
winds with the topography of the islands. GDL found posi-
tive temperature trends, at both low- and high-elevations,
comparable to changes observed in other tropical and sub-
tropical oceanic islands (Huber et al., 2006; Martín et al.,
2012; Diaz et al., 2014), and examined fluctuations in the
rate of warming over the 1919–2006 period. While previous
work has linked ENSO- and PDO-related natural variability
with fluctuations in temperature (Giambelluca et al., 2008 ;
Diaz et al., 2011) and precipitation (Diaz and Giambelluca,
2012; Frazier et al., 2018) in Hawai‘i, apparent shifts in
these teleconnections are not well understood. GDL noted,
for example, that in the most recent decades of their study,
temperature appeared to be less sensitive to phase changes in
the PDO.
With GDL as the starting point for this analysis, we
sought to extend the 88-year HTI record to 100 years. The
original HTI network of 21 stations actually had available
between 8 and 18 stations in any given year, and the number
of stations in the index had declined steadily from its peak in
the 1950s due to station closings. With further declines in
recent years, the homogeneity of the HTI, based on its
changing network, became questionable. Hence, it is neces-
sary to re-evaluate and refine the index to address problems
associated with the network. To do so, we develop a new
HTI by merging temperature anomaly time series from two
networks, spanning the early and late portions of the
100-year study period, respectively. Extending the time
series even by a few years is significant because the addition
of recent years captures both the period of a widely observed
global warming slowdown, which has been linked to PDO-
related changes in Pacific Ocean circulation patterns (Dai
et al., 2015), and the 2015–2016 El Niño, ranked among the
three strongest El Niño events in the historical record
(L'Heureux et al., 2017). Including these events in the time
series adds significantly to the available information on
regional temperature effects of ENSO and PDO.
To further expand on the work of GDL, in this study, we
have conducted additional analyses to validate the HTI
trends, by comparing the station-based index with time
series derived from NCEP/NCAR reanalysis data and a
mean atmospheric layer temperature time series derived
from the difference in geopotential height observations at
two pressure levels. We also re-examined the GDL conclu-
sions on the difference in warming trends between low- and
high-elevations in Hawai‘i, considering possible spurious
effects of small sample size and data gaps. Finally, in light
of the phenomenon of enhanced warming with elevation
more generally (Pepin et al., 2015), we examine temperature
trends for the highest elevation station in the islands, Mauna
Loa Observatory (19.54N, 155.58W, 3397 m), and com-
pared results with that of a high-elevation subtropical island
station in the Atlantic, Mt. Izaña Observatory (28.30N,
16.50W, 2373 m), Tenerife, Canary Islands. Both of these
stations are located above the local trade wind inversion
level and, hence, subject to effects of changes in Hadley cell
circulation. The similarity of long-term temperature trends in
the Canary Islands, as shown by Martín et al. (2012), to
those previously found for Hawai‘i led us to compare tem-
perature changes, particularly at higher elevations, at these
two widely separated, but similarly situated sites.
2|METHODS
GDL used a temperature index (1919–2006) derived from
21 stations (Table 1) distributed across the main Hawaiian
Islands. In this analysis, we added 12 additional years of data
from the National Oceanographic and Atmospheric Adminis-
tration's (NOAA's) National Centers for Environmental Infor-
mation (NCEI, http://www.ncdc.noaa.gov/cdo-web/) and the
Western Regional Climate Center (WRCC, https://wrcc.dri.
edu/summary/Climsmhi.html) from 1917–1918 and 2007–
2016 to the GDL HTI time series. Monthly T
mean
,T
min
,and
T
max
surface air temperature (SAT) are analysed, and station
annual temperature anomalies were calculated as the mean of
the monthly anomalies in each year. Stations missing more
3988 MCKENZIE ET AL.

than one monthly value in a given year were excluded.
Monthly anomalies were calculated as departures from the
baseline period (1944–1980) mean monthly values at each
station. The baseline period corresponds to the period with
the greatest number of index stations available. For each sta-
tion, the annual temperature anomaly time series was calcu-
lated as the departure relative to the station annual mean over
the baseline period. To address outliers, any values outside of
the 99% confidence interval (set as ±2.326 standard devia-
tions derived from a running 31-year sample), were excluded
from the analysis. Following GDL, temperature indices were
computed separately for state-wide, low- (below 800 m), and
high-elevation stations (above 800 m). Because of the scarcity
of high-elevation stations, the limit was set in GDL at 800 m
to include four stations, and retained here for the same reason.
The HTI anomalies were calculated as the weighted mean of
the low- and high-elevation indices with weights of 0.575 and
0.425 corresponding to the relative proportion of land area
below and above 800 m elevation, respectively.
Due to the various establishment and discontinuation
dates of stations (Table 1) the use of a single GDL network
for this extended analysis would have led to increasingly
greater effects of changes in network membership over time
(Figure 1a). Homogeneity is questionable if the number of
stations in the index changes significantly over the period
under study. Therefore, we re-evaluated the reference
network for this study. Many long-term stations were avail-
able early in the period but discontinued in the middle of
the study period, while other stations started in the middle
and continued toward the end of the period. To address this
problem, while trying to maximize the spatial coverage
throughout the period, we established two networks to rep-
resent the early and late parts of the record (Figure 1b, c).
The time series of the early and late networks were merged
with a total of 14 stations (Figure 2), four of which were
common to the two networks (Table 1), after adjustment of
the earlier network index that is based on a linear regression
between the two networks during the period of overlap
(1952–1983) to achieve a homogeneous 100-year index
time series. This process was done separately for T
mean
,
T
min
,andT
max
time series, state-wide, and for low- and
high-elevation stations. Subsequently, the F-ratio test was
used to assess the significance of the difference in the vari-
ances between the regression-adjusted temperature time
series of the earlier period and that of the original time
series of the later period.
As a measure of temperature variation in the broader
region, mean SAT, and mean sea surface temperatures (SST)
based on NCEP/NCAR reanalysis data (Kistler et al., 2001;
NCEP reanalysis data provided by the NOAA/OAR/ESRL
PSD, Boulder, Colorado, USA, from their website at http://
www.esrl.noaa.gov/psd/) were analysed. We also examined
TABLE 1 Locations and elevations for the 21 climate stations used in the study by Giambelluca et al. (2008) and the 14 of which (marked) were used in
this study
SKN Station name Status Elevation (m) Latitude (N) Longitude (W)
39 Mauna Loa Slope Observatory
a
1955 to present 3,341 19.54 155.58
338 Haleakal
a Ranger Station
a
1939 to present 2,143 20.76 156.25
54 Hawaiʻi Volcanoes Nat'l Park HQ
a,b
1949–2014 1,211 19.43 155.26
267 Kula Sanatorium/Hospital
a,b
1916 to present 916 20.70 156.36
436 Kaʻiliʻili 1925–2012 744 20.51 156.16
434 Haleakal
a BES/Farm Exp 4 1939–1992 640 20.51 156.18
672 L
anaʻi City 1930–2014 494 20.50 156.55
91 Mountain View 1906–1985 466 19.33 155.07
714 Tantalus 1936–1957 427 21.21 157.49
446 Kailua
b
1905 to present 213 20.89 156.22
73.2 Kainaliu 1939–2013 457 19.54 155.93
870 ʻŌpaeʻula
b
1949 to present 288 21.57 158.04
21 P
ahala
b
1905–1985 264 19.12 155.29
175.1 Kohala Mission
b
1905–1978 164 20.14 155.48
223 ʻŌʻ
okala
b
1949–1993 131 20.10 155.17
1,020 Līhuʻe
b
1905–1963 63 21.59 159.22
781 K
aneʻohe Mauka 1928–1998 60 21.25 157.49
1,020.1 Līhuʻe Airport
a
1950 to present 30 21.98 159.34
86 Hilo
a,b
1905 to 1966 12 19.44 155.05
87 Hilo Airport
a,b
1941 to present 12 19.72 155.06
703 Honolulu WB Airport
a
1946 to present 2 21.32 157.93
SKN: state key number used to identify climate stations.
a
Included in the 8-station network.
b
Included in the 10-station network.
MCKENZIE ET AL.3989

the NOAA Extended Reconstructed Sea Surface Tempera-
ture v5 dataset provided by the NOAA/OAR/ESRL PSD,
Boulder, Colorado, USA, from their website at https://www.
esrl.noaa.gov/psd (ERSST v5, Huang et al., 2017) around
Hawaiʻi for comparison with the HTI results. The datasets
were obtained for the HTI period (1917–2016).
The multivariate ENSO Index (MEI; Wolter and Timlin,
1998) was used to represent ENSO status in this study. The
MEI includes six observed fields (SLP, zonal and meridional
surface wind, SST, and total cloudiness) in the classification of
ENSO modes. MEI data were obtained from NOAA (http://
www.esrl.noaa.gov/psd/enso/mei/table.html). PDO index
values were obtained for the HTI period of record from the
Joint Institution for the Study of the Atmosphere and Ocean
(http://research.jisao.washington.edu/pdo/PDO.latest.txt).
Toumi et al. (1999) showed that observed pressure varia-
tions at high-elevation could be used to estimate temperature
change in the underlying air layer. Here, we followed a simi-
lar approach and analysed the difference in geopotential
height observations at two pressure levels (700 and
1,000 hPa) from Hawaiian radiosonde data to derive the
time series of mean 1,000–700 hPa (approximately the
lowest 3,000 m) air layer temperature (MLT). The MLT
time series (1977–2016) was calculated as a function of the
geopotential height difference or layer thickness based on
the assumption of the hydrostatic equilibrium:
dp
dz=−ρg,ð1Þ
where dp/dzis the vertical pressure gradient, ρis air density,
and gis acceleration due to gravity 9.81 m s
−2
. Applying this
relationship to a specific layer and substituting for ρusing the
equation of state (ρ=p
Rd

Tv, where R
d
is the gas constant for
dry air, 287 J kg
−1
K
−1
, and

Tvis the mean virtual tempera-
ture of the air layer) results in (Wallace and Hobbs, 1977):
z2−z1
Rd

Tv
gln p1
p2
 ð2Þ
where z
1
and z
2
are the geopotential heights
(m) corresponding to pressure levels p
1
, and p
2
(hPa). Re-
arranging,

Tvis obtained as:

Tv=z2−z1
ðÞ
Rd
gln p1
p2
 ð3Þ
This equation was applied using height observations at
the 1,000 hPa (~150 m a.s.l.) and 700 hPa (~3,200 m a.s.l.)
levels measured twice daily 0:00 UTC (2:00 p.m. HST) and
12:00 UTC (2:00 a.m. HST) from an atmospheric sounding
station at Hilo (19.72N, 155.05W) on Hawaiʻi Island
(http://weather.uwyo.edu/upperair/sounding.html).
Statistical analyses were performed in the R statistical
programming environment (R core Team, 2014). Secular
trends were calculated through linear regression using the
gls function in the nlme package (Pinheiro et al., 2017) to
explicitly account for temporal autocorrelation and fitted
the regression equation using restricted maximum likeli-
hood (REML). We assessed statistical significance at levels
of 0.001, 0.01, and 0.05, for the null hypothesis that the
trend is zero. In addition to the secular trend analyses, the
FIGURE 2 Elevation map of the temperature stations (red triangles) across
the Hawaiian islands used in this study [Colour figure can be viewed at
wileyonlinelibrary.com]
FIGURE 1 (a) Number of stations available in the 21-station network (GDL) during 1917–2016, (b) total number of stations in each year for the early
10-station subset network, and (c) total number of stations in each year for the later 8-station subset network. Vertical dashed line indicates the period (1977)
where the two networks were merged
3990 MCKENZIE ET AL.

Pearson product–moment correlation coefficients were
computed for assessing the relationships between atmo-
spheric variables and temperature indices.
3|RESULTS
3.1 |Network testing
Two temperature subset station networks were evaluated to
represent the early and late parts of the study period. A
10-station network was selected with nearly continuously
available data at all stations during 1917–1983. Similarly, an
eight-station network was identified with nearly complete data
during 1952–2016. Data from these network stations were
screened, and anomalies outside of the 99th percentile were
removed from the analysis. For each of T
mean
,T
min
,andT
max
and for state-wide, low- and high-elevation stations, linear
regressions between the annual time series of the two net-
works during the overlapping period, 1952–1983, were per-
formed (Figure 3). Results from the linear regressions showed
good relationships and were used to adjust the early time
series to correct for any systematic heterogeneity arising from
merging the two records. The state-wide mean bias errors
(MBE) between the two time series for T
mean,
T
min
,andT
max
during the 1952–1983 period before the adjustments were
0.012, −0.001, and −0.005C, respectively. After the adjust-
ments, state-wide MBE values were reduced to −0.007,
−0.000001, and 0.00003C, respectively, for T
mean,
T
min
,and
T
max
(see Table 2 for MBE values for all time series before
and after adjustments). The F-ratio test was used to compare
the variances of the time series derived from the two networks
during the overlapping period after the adjustments and found
no statistically significant differences. Therefore, for state-
FIGURE 3 Linear regression plots for annual mean temperature anomaly time series for the 10-station (xvariable) and 8-station (yvariable) networks during
the overlap period (1952–1983). The 10-station time series were subsequently adjusted using the linear regression equations shown here [Colour figure can be
viewed at wileyonlinelibrary.com]
TABLE 2 Mean bias error (MBE) values (C) for all time series during
overlapping periods of the earlier (10-station) and original later (8-station)
networks before and after linear regression adjustments
Annual series T
mean
T
min
T
max
State-wide Before adjustment 0.01180 −0.00114 −0.00497
After adjustment −0.00692 −0.0000014 0.00003
Low-elevation Before adjustment 0.01238 0.03379 −0.02158
After adjustment 0.00002 −0.00002 −0.00002
High-elevation Before adjustment −0.00748 −0.00002 0.01750
After adjustment −0.00003 −0.04841 −0.0000098
MCKENZIE ET AL.3991

wide, low-, and high-elevation stations, the two time series
were merged at the end of 1977, approximately in the middle
of the period of overlap, to create 100-year time series of
T
mean,
T
min
,andT
max
. The new HTI (T
mean,
T
min
,andT
max
for
state-wide, low- and high-elevation stations) have strong rela-
tionships with the previously defined HTI of GDL (Figure 4).
The new anomaly time series for state-wide T
mean
,T
min
,and
T
max
temperatures were tested against the corresponding GDL
time series, yielding coefficients of determination (r
2
)of0.92,
0.94 and 0.87, respectively.
3.2 |Station surface air temperature trends
The trends in near-surface temperature for state-wide
(Figure 5) and both low- and high- elevations stations
(Figure 6) are summarized in Table 3. Based on the long-
term (1917–2016) data, including the additional 12 years
included in this analysis, Hawaii's 100-year T
mean
has a sta-
tistically significant positive trend of 0.052C/decade
(p< 0.01; Figure 5a). As demonstrated in GDL, warming is
largely attributed to an increase in T
min
, which has seen a
significant long-term warming trend of 0.072C/decade
(p< 0.001) state-wide (Figure 5b). T
max
(Figure 5c), on the
other hand, was found to have no significant long-term
trend. The increase in T
min
and the consistency in T
max
imply
a significant reduction in DTR over the long-term period
(Figure 5d) with a trend of −0.054C/decade (p< 0.001).
T
mean
at low-elevation stations over the 100-year period
(1917–2016) shows a positive trend of 0.056C/decade
(p< 0.001; Figure 6a). At high-elevation stations, T
mean
over the long-term period was found to have a significant
positive trend of 0.047C/decade (p< 0.01; Figure 6b). A
significant rate of warming for T
min
(0.090C/decade; p
< 0.001) is also found at low-elevation stations (Figure 6c).
T
min
at high-elevation stations did not show a significant
trend over the long-term (Figure 6d). T
max
, for both low- and
high-elevation stations, shows no significant trend between
1917 and 2016 (Figure 7e–f).
Trends during subsets of the 100-year study period,
defined according to apparent periods of warming (+) and
cooling (−), 1917–1939 (+), 1940–1957 (−), 1958–1999
(+), and 2000–2013 (−), are also summarized in Table 3.
While temperature has fluctuated over the past 100 years,
the long-term trend is clearly upward. Although the period
FIGURE 4 Comparison of annual mean temperature anomalies used in the GDL HTI (yvariable) and the revised and extended index presented in this study
(xvariable) from 1919 to 2006, which have good to strong relationships [Colour figure can be viewed at wileyonlinelibrary.com]
3992 MCKENZIE ET AL.

2000–2013 had no significant changes in T
mean
,T
min
,or
T
max
state-wide, the last 3 years of the analysis, with 2015
and 2016 being the warmest years on record, strongly sug-
gest that rapid warming has resumed. The years 2014–2016
were exceptionally warm. However, even with 2015–2016
removed from the time series, the long-term (1917–2014)
trends in T
mean
(0.042C/decade; p< 0.01) and T
min
(0.066C/decade; p< 0.001) remain significant.
During the last four decades, state-wide decadal mean
T
mean
and T
min
anomalies were all higher than those of any
of the previous decades (Table 4). A particularly interesting
feature is the accelerated rate of increase in the last few
decades for T
min
state-wide and at low-elevation stations. At
both low- and high-elevation stations, T
mean
and T
min
in the
three most recent decades (excluding the recent decade for
high-elevation stations) were higher than in all previous
decades. Decadal means for T
max
, on the other hand, have
stabilized since 1987.
To assess the effect of adjusting the time series based on
the 10-station network (representing the earlier part of the
study period), trends were also calculated from the unad-
justed time series. Long-term trends for the merged time
series without adjustments (for T
mean
,T
min
, and T
max
, tem-
peratures state-wide and at both elevation ranges) are
slightly lower than the long-term trends derived from the
regression-adjusted time series, excluding the long-term
trend at low-elevation, which remains the same (Table 5).
3.3 |Temperature trends derived from other data sets
An alternative regional temperature time series was derived
using NCEP/NCAR reanalysis data (Kistler et al., 2001) for
comparison with our HTI results. Annual SATs over the
Hawaiʻi region, 15–25N, 170–140W, was analysed during
1948–2016. While not entirely independent of the station
data, the mean of gridded estimates for the region provide a
metric of the broader regional temperature variation with
assimilation of all available observations. The resulting tem-
perature time series (Figure 7) has a positive trend of
0.103C/decade (p< 0.01) over this 69-year period, similar
to the positive HTI trend for the same period
(0.122C/decade; p< 0.001). The two time series are also
well correlated (r = 0.84; p<0.001).
Geopotential height observations for the period
1977–2016 from radiosonde data provided an additional
temperature estimate for comparison with the HTI results.
The 700–1,000 hPa atmospheric MLT corresponds well
with the state-wide HTI over this period (Figure 8). While
the correlation between the two temperature estimates is rea-
sonably high (r= 0.64, p< 0.001), neither time series had a
significant trend over the 40-year overlap period. MLT has a
non-significant negative trend of −0.015C/decade. Please
note that estimation based on the geopotential height differ-
ence gives virtual temperature of the air layer between 1,000
and 700 hPa (

Tv), which can be affected by changes in spe-
cific humidity within the layer over time. Diaz et al. (2011)
FIGURE 5 Temperature anomalies for HTI in Hawaiʻi are shown as the annual mean relative to 1944–1980, with individual years shown as grey bars. The
black curve represents the 7-year running mean, and the lines correspond to the linear regressions for the 100-year period (1917–2016). The red lines indicate
significant trends. Non-significant trend is marked n.s. (p> 0.05) [Colour figure can be viewed at wileyonlinelibrary.com]
MCKENZIE ET AL.3993

showed that specific humidity in the 1,000–700 hPa layer
increased by about 0.15 g kg
−1
during 1958–2009. Assum-
ing the specific humidity increased during 1977–2016 at the
same rate as during 1958–2009, the humidity change would
have accounted for an increase in

Tvof about 0.02C (Elliott
et al., 1994, Equation 3).
The HTI time series is compared with the two large-scale
modes of natural internal climate variability, PDO and
ENSO, for the 100-year study period (1917–2016) in
Figure 9. HTI has a reasonably strong relationship with PDO
(Figure 9a) (r= 0.53, p< 0.001), whereas the HTI and
MEI (Figure 9b) have a weaker, but significant positive cor-
relation (r= 0.34, p< 0.001). GDL and Diaz et al. (2011)
showed that Hawai‘i temperature variations are strongly
coupled to the PDO and SST, but that in recent decades air
temperature in Hawai‘i had increased relative to the
variations in the PDO index. After revising and extending
the HTI time series, temperature still appears to be rising rel-
ative to the variations driven by the PDO (Figure 9a). Com-
paring recent HTI time series with NCEP/NCAR SST
anomalies averaged over 16–24N and 160–140W for the
period 1948–2016 (Figure 10a) shows that while air temper-
ature in the islands is positively correlated with SST of the
surrounding area (r= 0.53; p< 0.001), the SST trend of
0.032C/decade is much lower that the HTI trend,
0.122C/decade, for the same period. We also compared
HTI with another SST data set available for a longer period,
NOAA ERSST v5, averaged over 15–30N and
170–140W. Over the whole study period (1917–2016), it
also showed a positive correlation (r= 0.65; p< 0.001;
Figure 10b), and, in this case, a trend very similar to the HTI
trend (ERSST: 0.056C/decade vs. HTI: 0.052C/decade).
FIGURE 6 Temperature anomalies for HTI in Hawaiʻi for low-elevations (left column; (a), (c), and (e) and high-elevations (right column; (b), (d), and (f )
shown as the annual anomaly relative to 1944–1980, with individual years shown as grey bars. The black curve represents the 7-year running mean, and the
lines correspond to linear regressions for the 100-year period (1917–2016). The red lines indicate significant trends. Non-significant trends are marked
n.s. (p> 0.05) [Colour figure can be viewed at wileyonlinelibrary.com]
3994 MCKENZIE ET AL.

In comparison with either of the SST data sets, HTI
increases faster than SST starting around 1950, as was also
noted by GDL. It remains unclear why air temperature and
SST have similar centennial scale trends, but apparently dif-
fer on shorter time scales. Possible explanations include cli-
mate variations that affect land areas differently than the
surrounding ocean areas (affecting for example night time
cloud cover over the islands), and land cover changes. We
note that irrigated sugarcane acreage expanded rapidly
before around 1940 (Juvik and Juvik, 1998, p. 246), which
might have suppressed the air temperature trend relative to
the SST trend during that time. Hawaii's urbanized area has
experienced rapid expansion since the time of statehood
(1959) coincident with a decline in irrigated sugarcane, per-
haps contributing to enhanced air temperature warming rela-
tive to SST since that time.
Of particular interest is the apparent tuning of the
Hawai‘i air temperature time series to the global warming
signal. The long-term (1917–2016) global land-ocean tem-
perature index (GLOTI) in Figure 11a and the global ocean-
only temperature index (GOTI) in Figure 11b time series
(NOAA/NCEI, 2017; http://www.ncdc.noaa.gov/cag/) have
somewhat stronger warming trends than Hawaii's at
0.098C/decade and 0.081C/decade, respectively. The
20-year running trends are determined and compared
between the HTI, GLOTI, GOTI, and lowess-smoothed
PDO time series in Figure 11. Prior to around 1970, Hawaii's
temperature responses to natural climate variability appear to
have been more pronounced than those of the global indices.
Following 1970, however, HTI tracks GLOTI and GOTI
closely, lending confidence to the apparent link between
regional and global warming.
Figure 12 shows mean surface temperature from
Mt. Izaña Observatory (2,373 m), Tenerife, Canary Islands
(source of data: Emilio Cuevas, AEMET, Canary Islands,
Spain) for comparison with Mauna Loa Observatory (MLO)
in Hawai‘i, elevation 3,397 m. Mt. Izaña, shown in
Figure 12a, has a significant positive T
mean
of 0.15C/decade
(p< 0.001) over 1917–2016, and 0.19C/decade (p
< 0.001) for the 62-year period (1955–2016). MLO, in
Figure 12b, also has a mean warming trend of
0.19C/decade (p< 0.001) for the 62-year period
TABLE 3 Linear air temperatures trends for elevation ranges in Hawaiʻi and the whole state, for the entire period of record and for the periods with apparent
warming and cooling
Annual series T
mean
(C/decade) T
min
(C/decade) T
max
(C/decade) DTR (C/decade)
State-wide 1917–2016 0.052** 0.072*** 0.017 −0.054***
1917–1939 0.115 0.137* 0.025 −0.161
1940–1957 −0.428*** −0.323*** −0.589*** −0.264*
1958–1999 0.119** 0.197*** 0.059 −0.144*
2000–2013 −0.271 −0.420 −0.053 0.560***
Low-elevation 1917–2016 0.056*** 0.090*** 0.019
1917–1939 0.102 0.117 0.020
1940–1957 −0.410*** −0.256*** −0.541**
1958–1999 0.120* 0.177*** 0.053
2000–2013 −0.217 −0.252 −0.196
High-elevation 1917–2016 0.047** 0.051 0.016
1917–1939 0.073 0.085 −0.081
1940–1957 −0.266** −0.264 * −0.610 ***
1958–1999 0.120*** 0.220** 0.060
2000–2013 −0.387 −0.768 0.059
Symbols following fitted coefficient indicate statistical significance level at alpha: ***0.001; **0.01; *0.05; other values not significant.
FIGURE 7 Mean annual SAT anomalies from NCEP/NCAR reanalysis
data (black solid line) for Hawaiʻi region (15–25N, 170–140W) from
1948 to 2016 compared to the HTI (black dotted line). Linear trend for
NCEP/NCAR reanalysis data temperature is 0.103C/decade (red solid line)
and the HTI trend is 0.122C/decade (red dotted line) over the 69-year
period, 1948–2016 (both significant, p< 0.001.). There is a strong
correlation between two the data sets (r= 0.84; p< 0.001) [Colour figure
can be viewed at wileyonlinelibrary.com]
MCKENZIE ET AL.3995

(1955–2016). MLO and Mt. Izaña's temperatures anomalies
are plotted together in Figure 12c to show the similarities
and are significantly correlated with each other (r= 0.51, p
< 0.001; Figure 12d).
4|DISCUSSION
The result presented here of a mean long-term period
(1917–2016) warming rate for Hawai‘i of 0.052C/decade
(p< 0.01) updates the previously analysed temperature
change in Hawaiʻi for 1919–2006 (GDL; 0.043C/decade,
p= 0.05). The warming trend in Hawai‘i amounts to more
than half the global rate over the past century of
0.098C/decade (NOAA/NCEI, 2017). HTI time series and
the GLOTI time series show substantial coherency in the
timing of periods of accelerated and reduced warming
(Figure 11a).
GDL found that the rate of warming was much greater at
high-elevations than low-elevations during 1975–2006.
Their finding is consistent with expectations and observa-
tions of amplified warming with elevation (Pepin et al.,
2015). Diaz et al. (2011) analysed free atmosphere tempera-
ture trends along the vertical profile over Hawai‘i using
NCEP reanalysis data. They found warming throughout the
profile with a distinct peak at the 850 hPa level, which dif-
fers from the station-based findings of GDL. The extended
HTI analysis shows (see Table 3) that T
mean
at low-elevation
stations has a significant warming trend of 0.056C/decade
(p< 0.001) compared to 0.043C/decade in the GDL series.
At high-elevation stations (n= 4), T
mean
over 1917–2016
increased at 0.047C/decade (p< 0.01), slightly higher than
for the 85-year study period analysed by GDL
(0.044C/decade, p= 0.05), but lower than the low-
TABLE 4 Decadal mean surface mean, minimum, and maximum
temperature anomalies (C) 1917–2016 state-wide and for high and low
elevation stations
Annual series T
mean
T
min
T
max
State-wide 1917–1926 0.058 0.040 0.178
1927–1936 0.197 0.213 0.276
1937–1946 0.250 0.203 0.281
1947–1956 −0.189 −0.143 −0.247
1957–1966 −0.041 −0.082 0.059
1967–1976 0.106 0.172 0.041
1977–1986 0.300 0.249 0.482
1987–1996 0.458 0.707 0.314
1997–2006 0.415 0.720 0.189
2007–2016 0.470 0.476 0.295
Low-elevation 1917–1926 0.055 −0.037 0.201
1927–1936 0.174 0.032 0.349
1937–1946 0.231 0.153 0.278
1947–1956 −0.225 −0.095 −0.309
1957–1966 −0.052 −0.084 0.046
1967–1976 0.169 0.078 0.214
1977–1986 0.345 0.246 0.449
1987–1996 0.510 0.579 0.461
1997–2006 0.400 0.641 0.176
2007–2016 0.491 0.704 0.281
High-elevation 1917–1926 0.116 0.125 0.138
1927–1936 0.206 0.313 0.162
1937–1946 0.195 0.170 0.268
1947–1956 −0.122 −0.088 −0.142
1957–1966 0.008 −0.008 0.076
1967–1976 0.063 0.204 −0.192
1977–1986 0.233 0.219 0.522
1987–1996 0.388 0.880 0.116
1997–2006 0.475 0.826 0.207
2007–2016 0.442 0.167 0.314
TABLE 5 Surface temperature trend anomalies for merged time series without linear regression-adjustments for 1917–2016 state-wide, and for high- and
low-elevation stations
Annual series T
mean
(C/decade) T
min
(C/decade) T
max
(C/decade) DTR (C/decade)
State-wide 1917–2016 0.046** 0.067** 0.016 −0.050**
1917–1939 0.145 0.202* −0.027 −0.225
1940–1957 −0.580*** −0.476*** −0.644*** −0.161
1958–1999 0.114** 0.205*** 0.060 −0.160*
2000–2013 −0.272 −0.419 0.053 0.560***
Low-elevation 1917–2016 0.032 0.090*** 0.021
1917–1939 0.120 0.211 0.023
1940–1957 −0.481*** −0.461*** −0.610**
1958–1999 0.118* 0.184*** 0.060
2000–2013 −0.217 −0.252 −0.196
High-elevation 1917–2016 0.038 0.033 0.011
1917–1939 0.123 0.200 −0.091
1940–1957 −0.446** −0.611* −0.688***
1958–1999 0.118*** 0.242** 0.054
2000–2013 −0.388 −0.768 0.150
Symbols following fitted coefficient indicate statistical significance level at alpha: ***0.001; **0.01; *0.05; other values are not significant.
3996 MCKENZIE ET AL.

elevation trend. This result is similar to that of GDL for
1919–2006, that is, the long-term trends do not indicate
warming enhancement with elevation. But in contrast to
GDL, high-elevation trends are also not higher than low-ele-
vation trends in recent decades (Table 3). We note, however,
that the highest station in Hawai‘i (MLO) has been rapidly
warming much faster (0.19C/decade) than the high-eleva-
tion average (0.12C/decade, p< 0.001) for the 1955–2016
time period. This suggests that the selected elevation range
is too wide to properly see enhanced high-elevation warm-
ing. Indeed, an analysis of trends in air temperature lapse
rates in Hawai‘i (Kagawa-Viviani and Giambelluca, in
review) suggests that air temperature trends in the elevation
range of approximately 800–1,600 m have been very low or
even negative during the past 50 years. Thus, if enhanced
warming is occurring at the high-elevations, it is being can-
celled by flat or negative trends in the lower part of “high-
elevation”network used in our analysis. However, it also
might be an artefact of the noisy signal produced by the
small sample of high-elevation stations. While a higher limit
would suffer greater uncertainty due to an even smaller sam-
ple of stations, it might better capture the elevation effect.
Because it infers a more stable atmosphere, faster warming
at high-elevations is consistent with the observed drier con-
ditions during the past several decades in Hawaii (Frazier
and Giambelluca, 2017). Greater scrutiny of elevation-
related differences in Hawaii's air temperature trends is
ongoing. In a separate study, Kagawa-Viviani and Giambel-
luca (in review) focus specifically on the spatial patterns of
temperature change in Hawaiʻi.
To gain more insight into the possible systematic temper-
ature change for high-elevation tropical island sites, we com-
pared temperature change at MLO (3,397 m), Hawai‘i and
the Mt. Izaña Observatory (2,373 m), Tenerife, Canary
Islands, Spain. Mt. Izaña Observatory is the only high-eleva-
tion subtropical island site other than MLO with a long sur-
face air temperature record, as far as we know. At these two
stations, which differ in elevation by nearly 1,000 m, but are
both located above the typical local trade wind inversion
(TWI) height, the warming rates are virtually identical and
quite high (0.19C/decade, p< 0.001) for the period of
overlapping records (1955–2016; Figure 12). While a sam-
ple of only two stations is not adequate to make sweeping
generalizations, the rapid warming observed at these two
sites suggests that they are affected in a similar way by
ongoing large-scale climate warming. Furthermore, they are
similarly located latitudinally, in the descending branch of
the northern Hadley cell, albeit in different oceans, implying
that the ongoing warming of the layer above the TWI might
be systematic and widespread. It also indicates that the real
FIGURE 8 Mean atmospheric layer temperature (MLT) anomalies from
Hilo radiosonde height (m) between the 1,000 and 700 hPa surfaces relative
to 1985–2005 base period and mean surface temperature anomalies in
Hawaiʻi (HTI). Linear trend for MLT ( )is−0.015C/decade over the
40-year period, 1977–2016 (not significant; p= 0.90) and the linear trend
for temperature ( ) is 0.079C/decade over the 40-year period,
1977–2016 (not significant; p= 0.17). There is a fairly good correlation
between the MLT from radiosonde data and HTI (r= 0.641, p< 0.001)
FIGURE 9 Annual and smoothed (with a 7-year running mean) HTI and
both PDO trends and MEI trends for the study period. (a) Significant
correlation are found between PDO and HTI (r= 0.53, p< 0.001), ( )
PDO index, ( ) PDO smoothed, ( ) HTI, and ( ) HTI smoothed
and (b) there is a positive correlation between MEI and HTI (r= 0.34,
p< 0.001), ( ) MEI index, ( ) MEI smoothed, ( ) HTI, and
() HTI smoothed [Colour figure can be viewed at
wileyonlinelibrary.com]
MCKENZIE ET AL.3997

break in lowland/highland warming rates is not at 800 m,
but rather near or above the inversion level, which is at a
much higher altitude –around 2,000–2,300 m in Hawai‘i
(Longman et al., 2015) and around 1,500 m in the Canary
Islands (Sanroma et al., 2010; Martín et al., 2012).
Although rapid warming in the 800 m and above layer
found by GDL has apparently not persisted based on the
extended HTI, enhanced warming at the highest elevations
(MLO) in Hawai‘i comports with the abundant evidence of
elevation dependent warming in mountains throughout the
world (Pepin et al., 2015). While several possible mecha-
nisms have been identified (Rangwala and Miller, 2012;
Pepin et al., 2015), none point specifically to the zone above
the TWI. We note that, areas above the TWI are warmer than
they would be without the TWI. Hence, a more persistent
TWI would enhance warming above the mean TWI level.
Subsiding air in the descending branch of the regional Had-
ley circulation, which maintains the TWI, influences both
Hawaiʻi and the Canary Islands. Longman et al. (2015)
found consistent increases in the intensity and persistence of
the TWI over the 1973–2013 period and found a good corre-
lation between the TWI frequency of occurrence and the
strength of Hadley cell subsidence in Hawaiʻi. Should the
apparent uptick in Hadley cell overturning and consequent
increase in TWI frequency of occurrence prove to be linked
FIGURE 11 (a) HTI (base period 1917–2000), PDO and GLOTI time
series (base period 1901–2000) comparison for the 1917–2016 time period.
The GLOTI trend is 0.098C/decade for the 100-year period. ( ) HTI, linear
trend: 1917–2016: 0.052C/decade, ( ) GLOTI, linear trend: 1917–2016:
0.098C/decade, ( ) HTI 20-year running mean, ( ) GLOTI 20-year
running mean, ( ) PDO LOWEES-smoothed. (b) HTI (base period
1917–2000), PDO and GOTI time series (base period 1901–2000)
comparison for the 1917–2016 time period. The global ocean trend is
0.081C/decade for the 100-year period. ( ) HTI, linear trend: 1917–2016:
0.052C/decade, ( ) GOTI, linear trend: 1917–2016: 0.081C/decade,
() HTI 20-year running mean, ( ) GLOTI 20-year running mean,
() PDO LOWEES-smoothed [Colour figure can be viewed at
wileyonlinelibrary.com]
FIGURE 10 (a) SST anomaly for 16–24N, 160–140W using
NCEP/NCAR reanalysis data in comparison with HTI for the 67-year
period (1948–2016). SST trend is 0.032C/decade (not significant, p= 0.4),
and the warming trend of 0.122C/decade for HTI between 1948 and 2016
(p< 0.001). There is positive correlation between HTI and SST (r= 0.53,
p< 0.001). ( ) SST, ( ) SST smoothed, ( ) SST, linear trend:
1948–2016: 0.032C/decade, ( ) HTI, ( ) HTI smoothed, ( )
HTI, linear trend: 1948–2016: 0.122C/decade. (b) ERSST v5 for 15–30N,
170–140W in comparison with HTI over the 100-year (1917–2016) record
resulted in a warming trend of 0.052C/decade (p< 0.001) and strong
correlation between the two variables (r= 65, p< 0.001). ( ) ERSST,
() ERSST smoothed, ( ) ERSST, linear trend: 1917–2016:
0.056C/decade, ( ) HTI, ( ) HTI smoothed, ( ) HTI, linear
trend: 1917–2016: 0.052C/decade [Colour figure can be viewed at
wileyonlinelibrary.com]
3998 MCKENZIE ET AL.

to global climate change, as suggested by downscaled cli-
mate projections for Hawai‘i (Lauer et al., 2013), rapid
warming at high-elevations in Hawai‘i is likely to continue.
As the global climate warms, Hawaiʻi is likely to see
increasing atmospheric stability, due in part to a more persis-
tent TWI (Cao et al., 2007; Lauer et al., 2013). In the Canary
Islands, an intensification of the trade wind circulation in the
northern windward slopes of the Canary Islands is expected
with increases in wet and humid air below the inversion, but
above the inversion, with the strengthening of the Hadley
cell, the air is dry and clear (Sanroma et al., 2010). Although
mean temperature and precipitation are known to be strongly
modulated by ENSO in Hawaiʻi, interannual to decadal
drivers of climatic variability in Canary Islands are generally
different. Nonetheless, the similarity in the observed rate of
warming in response to global climate change at these two
widely separated sites is striking.
T
min
in Hawai‘i has increased over the long-term period,
while T
max
has not increased significantly. These results
translate to a considerable drop in the DTR. Vose et al.
(2005) found a global trend in DTR of −0.066C/decade
over land areas between 1950 and 2004 as a result of
increased rapid T
min
rise. Malamud et al. (2011) showed that
greater increases in T
min
compared with T
max
at Mauna Loa
Observatory had caused DTR to decrease by
−0.50C/decade over the period 1977–2006. Martín et al.
(2012) also found greater night time warming for high-eleva-
tion areas of Tenerife.
Throughout the 100-year record, temperature anomalies
have experienced periods of both positive and negative
FIGURE 12 (a) and (b) annual mean surface temperature anomalies (grey bars), 7-year running means (black curve) and trends (red lines) at Mt. Izaña
(2,373 m), Canary Islands, Spain, and Mauna Loa Observatory (MLO; 3,397 m), Hawaiʻi island, relative to a 1961–1990 baseline period. (a) Mt. Izaña has a
significant warming rate of 0.15C/decade (solid red line) for the century (p< 0.001; 1917–2016), and a significant trend of 0.19C/decade (p< 0.001;
dashed red line) over the recent 62-year period (1955–2016). (b) MLO also has a warming rate of 0.19C/decade (p< 0.001; solid red line) over the
1955–2016 period. (c) Comparison of annual mean surface temperature anomalies on Mt. Izaña and MLO. (d) Scatterplot of MLO vs. Mt. Izaña annual
temperature time series during 1955–2016 (r= 0.51, p< 0.001) [Colour figure can be viewed at wileyonlinelibrary.com]
MCKENZIE ET AL.3999

trends. However, the long-term trend is positive, with new
record highs in recent years. Based on the HTI, 2015 and
2016, which were influenced by one of the strongest El Niño
events on record, were the warmest years throughout the
100-year study period at 0.788 and 0.924C above the
100-year mean, respectively. The extreme positive tempera-
ture anomaly associated with the 2015–2016 event reached
1.5C during July–October 2015 (Zhu and Li, 2017).
5|CONCLUSION
We refined a previous HTI index and extended the record by
12 years from the earlier reported 88-year time series
(1919–2006, GDL 2008) to a 100-year trend series
(1917–2016). To minimize the effects of discontinuous sta-
tion data records, an improved HTI was developed. We com-
pared the station-based results with other datasets
(e.g., radiosonde and Reanalysis data). The various resulting
temperature series were consistent with the HTI results, both
well correlated and exhibiting similar trends. The long-term
trends for T
mean
,T
min
,orT
max
anomalies for state-wide, low-
, and high-elevation stations were assessed. T
mean
and T
min
for the 100-year study period state-wide showed significant
rates of warming (0.052C/decade and 0.072C/decade,
respectively). T
mean
trends by elevation showed higher rates
of warming at low-elevations (0.056C/decade) compared to
the high-elevation stations (0.047C/decade) over the last
century. However, much higher rates of warming are evident
above the TWI at Hawaii's highest elevation station, MLO,
and at Mt. Izaña Observatory in the Canary Islands. The
regional effects of the global warming slowdown beginning
in the early 2000s coinciding with a phase change in the
PDO continue into the additional recent years added to the
HTI record. However, rapid warming appears to have
resumed in Hawai‘i early in 2014.
ACKNOWLEDGMENTS
This research was made possible through funding by the
Office of Maunakea Management (OMKM). We thank Emi-
lio Cuevas of AEMET, Canary Islands, Spain for providing
the Mt. Izaña temperature data and John Barnes, Research
Scientist, NOAA/ESRL, for access to MLO data. We thank
Aurora Kagawa-Viviani for providing insightful comments
and edits. We also thank Alexandra Hedgpeth for help at the
start of this project, and Camilo Mora for the radiosonde
data excel macro code. Special thanks to Jim Juvik of the
University of Hawaiʻi at Hilo and Jessica Kirkpatrick
of OMKM.
ORCID
Marie M. McKenzie https://orcid.org/0000-0002-9365-
4286
REFERENCES
Cao, G., Giambelluca, T.W., Stevens, D.E. and Schroeder, T.A. (2007) Inversion
variability in the Hawaiian trade wind regime. Journal of Climate, 20,
1145–1160. https://doi.org/10.1175/JCLI4033.1.
Dai, A., Fyfe, J.C., Xie, S.-P. and Dai, X. (2015) Decadal modulation of global
surface temperature by internal climate variability. Nature Climate Change,
5, 555–559.
Diaz, H.F. and Giambelluca, T.W. (2012) Changes in atmospheric circulation
patterns associated with high and low rainfall regimes in the Hawaiian
islands region on multiple time scales. Global and Planetary Change, 98-99,
97–108.
Diaz, H.F., Giambelluca, T.W. and Eischeid, J.K. (2011) Changes in the vertical
profiles of mean temperature and humidity in the Hawaiian islands. Global
and Planetary Change, 77(1–2), 21–25. https://doi.org/10.1016/j.gloplacha.
2011.02.007.
Diaz, H.F., Bradley, R.S. and Ning, L. (2014) Climatic changes in mountain
regions of the American cordillera and the tropics. Arctic, Antarctic, and
Alpine Research, 46, 735–743. https://doi.org/10.1657/1938-4246-46.4.735.
Elliott, W.P., Gaffen, D.J., Angell, J.K. and Kahl, J.D.W. (1994) The effect of
moisture on layer thicknesses used to monitor global temperatures. Journal
of Climate, 7(2), 304–308. https://doi.org/10.1175/1520-0442(1994)007<03
04:TEOMOL>2.0.CO;2.
Frazier, A.G. and Giambelluca, T.W. (2017) Spatial trend analysis of Hawaiian
rainfall from 1920 to 2012. International Journal of Climatology, 37,
2522–2531. https://doi.org/10.1002/joc.4862.
Frazier, A.G., Elison Timm, O., Giambelluca, T.W. and Diaz, H.F. (2018) The
influence of ENSO, PDO and PNA on secular rainfall variations in Hawai‘i.
Climate Dynamics, 51, 2127–2140. https://doi.org/10.1007/s00382-017-
4003-4.
Giambelluca, T.W., Diaz, H.F. and Luke, M.S.A. (2008) Secular temperature
changes in Hawai‘i. Geophysical Research Letters, 35(12), L12702. https://
doi.org/10.1029/2008GL034377.
Huang, B., Thorne, P.W., Banzon, V.F., Boyer, T., Chepurin, G., Lawrimore, J.
H., Menne, M.J., Smith, T.M., Vose, R.S. and Zhang, H.-M. (2017) NOAA
extended reconstructed sea surface temperature (ERSST), Version 5. [SST/-
Skin T, 1916–2016]. NOAA, National Centers for Environmental Informa-
tion. https://doi.org/10.7289/V5T72FNM accessed August 14, 2018.
Huber, U.M., Bugmann, H.K. and Reasoner, M.A. (2006) Global change and
mountain regions: an overview of current knowledge. Dordrecht, The
Netherlands: Springer.
Juvik, S.P. and Juvik, J.O. (1998) Atlas of Hawai‘i, 3rd edition. Honolulu,
Hawai‘i: University of Hawai‘i Press.
Keener, V.W., Marra, J.J., Finucane, M.L., Spooner, D. and Smith, M.H. (Eds.).
(2012) Climate change and pacific islands: indicators and impacts: Report
for the 2012 Pacific Islands Regional climate assessment. Washington, DC:
Island Press.
Kistler, R., Collins, W., Saha, S., White, G., Woollen, J., Kalnay, E., et al.
(2001) The NCEP-NCAR 50-year reanalysis: monthly means CD-ROM and
documentation. Bulletin of the American Meteorological Society, 82(2),
247–267.
Lauer, A., Zhang, C., Elison-Timm, O., Wang, Y. and Hamilton, K. (2013)
Downscaling of climate change in the Hawaii region using CMIP5 results:
on the choice of the forcing fields. Journal of Climate, 26(24),
10006–10030. https://doi.org/10.1175/JCLI-D-13-00126.1.
L'Heureux, M.L., et al. (2017) Observing and predicting the 2015/16 El Niño.
Bulletin of the American Meteorological Society, 98(7), 1363–1382. https://
doi.org/10.1175/BAMS-D-16-0009.1.
Longman, R.J., Diaz, H.F. and Giambelluca, T.W. (2015) Sustained increases in
lower-tropospheric subsidence over the central tropical North Pacific drive a
decline in high-elevation rainfall in Hawaii. Journal of Climate, 28(22),
8743–8759. https://doi.org/10.1175/JCLI-D-15-0006.1.
Malamud, B.D., Turcotte, D.L. and Grimmond, C.S.B. (2011) Temperature
trends at the Mauna Loa observatory. Hawaii. Climate Past, 7(3), 975–983.
https://doi.org/10.5194/cp-7-975-2011.
Mantua, N.J. and Hare, S. (2002) Pacific-Decadal Oscillation (PDO). Encyclope-
dia of global environmental change, 1, 592–594.
Mantua, N.J., Hare, S.R., Zhang, Y., Wallace, J.M. and Francis, R.C. (1997) A
Pacific Interdecadal climate oscillation with impacts on Salmon production.
Bulletin of the American Meteorological Society, 78(6), 1069–1079. https://
doi.org/10.1175/1520-0477(1997)078<1069:APICOW>2.0.CO;2.
4000 MCKENZIE ET AL.

Martín, J.L., Bethencourt, J. and Cuevas-Agulló, E. (2012) Assessment of global
warming on the Island of Tenerife, Canary Islands (Spain). Trends in mini-
mum, maximum and mean temperatures since 1944. Climatic Change, 114
(2), 343–355. https://doi.org/10.1007/s10584-012-0407-7.
National Centers for Environmental Information (NOAA), Climate at a Glance:
U.S. Time Series, 2017, retrieved on February 16, 2017, from http://www.
ncdc.noaa.gov/cag/
Pachauri, R.K., Allen, M.R., Barros, V.R., Broome, J., Cramer, W., Christ, R.,
et al. (2014) Climate Change 2014: Synthesis Report. In: Pachauri, R. and
Meyer, L. (Eds.) Contribution of Working Groups I, II and III to the Fifth
Assessment Report of the Intergovernmental Panel on Climate Change.
Geneva, Switzerland: IPCC, p. 151. isbn:978–92–9169-143-2.
Pepin, N., Bradley, R.S., Diaz, H.F., Baraer, M., Caceres, E.B., Forsythe, N.,
Greenwood, G., et al. (2015) Elevation-dependent warming in mountain
regions of the world. Nature Climate Change, 5, 424–430. https://doi.org/10.
1038/NCLIMATE2563.
Pinheiro J, Bates D, DebRoy S, Sarkar D and R Core Team (2017). Nlme: linear
and nonlinear mixed effects models. R package version 3.1-131, URL:
https://CRAN.R-project.org/package=nlme.
R Core Team. (2014) R: a language and environment for statistical computing.
Vienna, Austria: R Foundation for Statistical Computing URL http://www.
R-project.org/.
Rangwala, I. and Miller, J.R. (2012) Climate change in mountains: a review of
elevation-dependent warming and its possible causes. Climatic Change, 114,
527–547. https://doi.org/10.1007/s10584-012-0419-3.
Salinger, M.J. (2005) Climate variability and change: past, present and future —
an overview. In: Salinger, J. , Sivakumar, M.V.K. and Motha, R.P. (Eds.)
Increasing climate variability and change: Reducing the vulnerability of
agriculture and forestry. Dordrecht: Springer Netherlands, pp. 9–29. https://
doi.org/10.1007/1-4020-4166-7_3.
Sanroma, E., Palle, E. and Sanchez-Lorenzo, A. (2010) Long-term changes in
insolation and temperatures at different altitudes. Environmental Research
Letters, 5(2), 024006. https://doi.org/10.1088/1748-9326/5/2/024006.
Schneider, N. and Cornuelle, B.D. (2005) The forcing of the Pacific decadal
oscillation. Journal of Climate, 18(21), 4355–4373. https://doi.org/10.1175/
JCLI3527.1.
Toumi, R., Hartell, N. and Bignell, K. (1999) Mountain Station pressure as an
indicator of climate change. Geophysical Research Letters, 26(12),
1751–1754. https://doi.org/10.1029/1999GL900351.
Vose, R.S., Easterling, D.R. and Gleason, B. (2005) Maximum and minimum
temperature trends for the globe: an update through 2004. Geophysical
Research Letters, 32(23), L23822. https://doi.org/10.1029/2005GL02
4379.
Wallace, J.M. and Hobbs, P.V. (1977) Atmospheric science, an introductory sur-
vey. New York: Academic Press.
Wolter, K. and Timlin, M.S. (1998) Measuring the strength of ENSO events:
how does 1997/98 rank? Weather, 53(9), 315–324. https://doi.org/10.1002/j.
1477-8696.1998.tb06408.x.
Zhang, Y., Wallace, J.M. and Battisti, D.S. (1997) ENSO-like Interdecadal Vari-
ability: 1900-93. Journal of Climate, 10(5), 1004–1020. https://doi.org/10.
1175/1520-0442(1997)010<1004:ELIV>2.0.CO;2.
Zhu, Z. and Li, T. (2017) The record-breaking hot summer in 2015 over Hawaii
and its physical causes. Journal of Climate, 30(11), 4253–4266. https://doi.
org/10.1175/JCLI-D-16-0438.1.
How to cite this article: McKenzie MM,
Giambelluca TW, Diaz HF. Temperature trends in
Hawaiʻi: A century of change, 1917–2016. Int
J Climatol. 2019;39:3987–4001. https://doi.org/10.
1002/joc.6053
MCKENZIE ET AL.4001