You are on page 1of 40

74 The Open Atmospheric Science Journal, 2011, 5, 74-86

Open Access
Climate Change Attribution Using Empirical Decomposition of Climatic
Data
Craig Loehle1 and Nicola Scafetta*,2

1
National Council for Air and Stream Improvement, Inc., 552 S Washington Street, Suite 224, Naperville, Illinois 60540,
USA
2
Active Cavity Radiometer Irradiance Monitor (ACRIM) & Duke University, Durham, North Carolina 27708, USA

Abstract: The climate change attribution problem is addressed using empirical decomposition. Cycles in solar motion and
activity of 60 and 20 years were used to develop an empirical model of Earth temperature variations. The model was fit to
the Hadley global temperature data up to 1950 (time period before anthropogenic emissions became the dominant forcing
mechanism), and then extrapolated from 1951 to 2010. The residuals showed an approximate linear upward trend of about
0.66C/century from 1942 to 2010. Herein we assume that this residual upward warming has been mostly induced by
anthropogenic emissions, urbanization and land use change. The warming observed before 1942 is relatively small and is
assumed to have been mostly naturally induced. The resulting full natural plus anthropogenic model fits the entire 160
year record very well. Residual analysis does not provide any evidence for a substantial cooling effect due to sulfate
aerosols from 1940 to 1970. The cooling observed during that period may be due to a natural 60-year cycle, which is
visible in the global temperature since 1850 and has been observed also in numerous multisecular climatic records. New
solar activity proxy models are developed that suggest a mechanism for both the 60-year climate cycle and a portion of
the long-term warming trend. Our results suggest that because current models underestimate the strength of natural
multidecadal cycles in the temperature records, the anthropogenic contribution to climate change since 1850 should be
less than half of that previously claimed by the IPCC. About 60% of the warming observed from 1970 to 2000 was very
likely caused by the above natural 60-year climatic cycle during its warming phase. A 21st Century forecast suggests that
climate may remain approximately steady until 2030-2040, and may at most warm 0.5-1.0C by 2100 at the estimated
0.66C/century anthropogenic warming rate, which is about 3.5 times smaller than the average 2.3C/century
anthropogenic warming rate projected by the IPCC up to the first decades of the 21st century. However, additional
multisecular natural cycles may cool the climate further.
Keywords: Attribution, climate change, ENSO, LULC change, solar activity, UHI effect.

1. INTRODUCTION computations are poorly constrained. Each factor considered


is uncertain both in terms of forcing and in terms of data, as
There is little doubt that the Earth has warmed since 1850,
documented by Scafetta [8]. For example, while greenhouse
the time when global instrumental temperature estimates are
gases such as CO2 and CH4 no doubt have a warming effect
first available (Brohan et al. [1]). A significant increase in the
and sulfate aerosols produced by volcanoes or industrial
warming has been observed since 1970 relative to the period emissions no doubt have a cooling effect, the IPCC [2] itself
from 1940 to 1970: see Fig. (1A). This sudden change in the
acknowledges that there are large uncertainties as to the
warming trend has suggested an alarming anthropogenic effect
magnitude of their effects on climate. The same can be said
on climate (IPCC [2]). However, partitioning causation has
of brown clouds (warming), black soot (warming), solar
proven to be problematic. It is important to establish the relative
activity, and other factors. In none of these cases is a precise
importance of forcing factors in order to properly calibrate
global dataset for the past 160 years available to provide
climate models used to project future climate scenarios and, in input to an attribution study (Kiehl [12]), nor is the precise
particular, it is necessary to determine whether multidecadal
forcing known independently of attribution studies (e.g.,
climatic patterns may be induced by natural multidecadal cycles.
Visser et al. [13]). For example, the IPCCs best estimate of
Attribution studies are based on either statistical or the climate sensitivity to CO2 doubling is 3 +/- 1.5 oC,
simulation approaches, or a combination of these. Several although the distribution is skewed and has a tail to 10 oC
studies (e.g., Andronova and Schlesinger [3]; Crowley [4]; (see figure 9.20 of IPCC [2]). Other climate mechanisms in
Damon and Jirikowic [5]; Hoyt and Schatten [6]; Lean et al. climate models are also either poorly modeled (cloud cover,
[7]; Scafetta [8]; Serreze et al. [9]; Stott et al. [10]; Tett et al. water vapor, ocean oscillations) or highly uncertain (land
[11]) have arrived at divergent estimates of anthropogenic cover). More spatially resolved climate simulations (e.g.,
forcing. One reason for these divergent results is that the Wyant et al. [14]) produce cloud and water vapor patterns
that differ from those in operational climate models. These
differences are difficult to resolve within the traditional
*Address correspondence to this author at the Active Cavity Radiometer attribution study framework.
Irradiance Monitor (ACRIM) & Duke University, Durham, North Carolina
27708, USA; Tel: 919-660-5252; E-mail: nicola.scafetta@gmail.com

1874-2823/11 2011 Bentham Open


Climate Change Attribution Using Empirical Decomposition of Climatic Data The Open Atmospheric Science Journal, 2011, Volume 5 75

To overcome this problem we apply an empirical an opportunity to detect the anthropogenic signal as a
decomposition method that does not depend on assumptions residual component not explained by natural cycles.
about forcing magnitudes, on detailed historical inputs of the
In what follows, we empirically estimate the structure of
various factors, nor on climate models. Our approach is to
the temperature history data over a 160 year interval and
empirically characterize the pattern of global temperature evaluate the components of the empirical model as well as
change and to relate this pattern to the historical timing of
the residuals to assess natural vs anthropogenic forcings. A
anthropogenic greenhouse gas forcing.
key to the analysis is the assumption that anthropogenic
2. DECOMPOSITION ANALYSIS forcings become dominant only during the second half of the
20th century with a net forcing of about 1.6 W/m2 since 1950
Long-term climate trends and cycles have been detected (e.g., Hegerl et al. [23]; Thompson et al. [24]). This
in many geologic datasets. The most regular of these are the assumption is based on figure 1A in Hansen et al. [25] which
repeated cycles of glaciations evident in ice core data (e.g., shows that before 1970 the effective positive forcing due to a
Rasmussen et al. [15]), though the precise factors governing natural plus anthropogenic increase of greenhouse gases is
this cycle are not fully worked out. A number of studies mostly compensated by the aerosol indirect and tropospheric
claim to have found decadal to centennial periodic behaviors cooling effects. Before about 1950 (although we estimate a
in climate records that can be associated to solar cycles more precise date) the climate effect of elevated greenhouse
(discussed later). As well, the 11 and 22 year sunspot cycles gases was no doubt small (IPCC [2]).
appear to result from the effect of planetary tidal forces on
the sun (e.g., Bendandi [16]; Hung [17]). These findings The approach we take is illustrated by the work of
offer the possibility that decadal to centennial scale climate Klyashtorin and Lyubushin [26] who identified a periodicity
could have some structure in the absence of human activity, of about 60 years in long-term climate data and fit a model to
rather than being featureless. This possibility is not ruled out the global anomaly data, with good results. A problem with
by the IPCC [2] because no scientific consensus exists on the their model is that it is fit to the period in which we wish to
importance of solar forcing and the current level of detect a difference from any natural cycles of climate. A
understanding associated with it is stated by IPCC to be low. better approach would be to fit the model to the period prior
to elevated greenhouse gases and then extrapolate it forward
Using the pattern of perturbations of the suns location and determine whether the forecast climate pattern
from the center of mass of the solar system as a measure of corresponds to the observed one. In fact, in the latter case the
the oscillations of the sun-planets system due to gravitational model would be tested on its forecasting capability.
and magnetic interactions, which can be back-calculated by
orbital calculations to any desired length of time, Scafetta As explained above, we follow Thompson et al. [24] and
[18] showed via spectral analysis and other means that two IPCC [2] in identifying 1950 as the approximate year after
particularly strong periodic signals occur with the periods of which an anthropogenic signal appears to dominate climate
about 60 and 20 years. These oscillations are synchronous in forcings. Hegerl et al. [23] identified 1960 as the cutoff year,
both astronomical records and numerous global surface but their analysis might have found an earlier year if sulfate
temperature and climate records. Spectral decomposition of aerosols were not included in it. Given the strong periodicity
the Hadley climate data showed spectra similar to the in the solar signal detected by Scafetta [18], we used this
astronomical record, with a spectral coherence test being model, with two cycles of lengths 60 and 20 years, plus a
highly significant with a 96% confidence level. A model linear term. A free fit of the Hadley HadCRUT3 global
based on these astronomical cycles fit the global temperature surface temperature data found cycles quite close to this, but
data well and fit the ocean temperature data even better. On we use the Scafetta solar periodicities plus a linear trend. The
the contrary, the spectral patterns of climate model linear trend would approximately extrapolate a natural
simulations did not match those found in the climate at all warming trend due to solar and volcano effects that is known
(just a 16% confidence level), suggesting that current general to have occurred since the Little Ice Age, a period that
circulation models adopted by the IPCC do not reproduce the encompassed the Maunder and Dalton solar minima of the
climate oscillations at the decadal and multidecadal scales 17-19th centuries (Eddy [27]; Scafetta [8]).
(Scafetta [18]). As explained in Scafetta and West [28] and in Scafetta
The physical mechanisms responsible for the cycles are [8] most of this warming is likely due to the increased solar
not understood yet and, therefore, are not included in the activity with the volcano effect playing only a minor effect if
current climate models. However, it is reasonable to recent paleoclimatic reconstructions of the global climate,
speculate that planetary tides and solar angular motion affect which show a large pre-industrial variability (e.g. Moberg
solar activity (Scafetta [18]; Wolff and Patrone [19]). et al. [29]), are adopted. On the contrary, analysis using only
Changes of solar activity then influence the Earth's climate Hockey Stick temperature graphs, which do not show a
through changes of direct total solar irradiance, the suns significant pre-industrial variability (e.g. Mann et al.
magnetic field and the solar wind. Very likely, solar activity [30]), suggest that volcanic and solar effects are equally
changes influence also the intensity of cosmic rays reaching important in explaining the Little Ice Age (Crowley [4];
the Earth, the electrical physical properties of the terrestrial Scafetta and West [28]).
magnetosphere and of the ionosphere. The latter influences The model was fit by nonlinear least squares estimation
the generation of clouds which modulate terrestrial albedo using Mathematica functions, with phase and amplitude free
and, consequently, the climate (Kirkby [20]; Tinsley [21]; but period fixed as above. The validity of this approach for
Svensmark et al. [22]). The above empirical relationship, estimating parameters for a time series model with data of
even if the exact mechanisms are not understood yet, offers
76 The Open Atmospheric Science Journal, 2011, Volume 5 Loehle and Scafetta

Fig. (1). A) Model for entire period utilizing HadleyCRUT3 global surface temperature April 8, 2010 dataset. B) Residuals showing clear
model mis-specification.

this length is demonstrated in Supplemental Information a. using the data only up to 1950 (Fig. 2A). See Table 1 for the
When fit to the entire 160 year period, the model evaluated regression coefficients in the two cases.
y (t) = A cos[2 (t-T1)/60] + B cos[2 (t-T2)/20] + C (t-1900) Just as in the solar data, the 60-year cycle has about three
+D (1) times the peak-to-trough amplitude (~0.24C) of the 20 year
cycle (~0.08 C). The model fit is R2 = 0.53. Comparing the
performs only decently (Fig. 1A), as the nonrandom residual
residuals over the entire period depicted in Fig. (2B), they
signal over the entire period (Fig. 1B) reveals that the model
appear stationary up to 1950, but there is a visible positive
is deficient. Since this mis-specification could result from
not including anthropogenic effects that could have trend from 1950 to 2010. The residual warming observed
since 1950 is mostly induced by an increasing anthropogenic
accelerated the warming since 1950, the model was refit
Climate Change Attribution Using Empirical Decomposition of Climatic Data The Open Atmospheric Science Journal, 2011, Volume 5 77

Fig. (2). A) As in Fig. (1A), with model fit to pre-1950 data. B) Residuals. Before about 1950 residuals are stationary around the zero level.
After about 1942 there is a clear upward linear trend which may be associated to anthropogenic warming.

warming signal due to global industrial development, land trend is the result of all positive and negative additional
usage change and urban heat island (UHI) effects since 1950. climatic forcings (e.g., the anthropogenic ones) and other
A linear fit to the residuals since 1950 (Fig. 2B) has a possible contributions (e.g., poor adjustment of the
temperature data) that are not implicit in our model
slope of 0.66 +/- 0.08C/century (R2 = 0.59) and begins (>0)
beginning in 1942.
in 1942. Note that given a roughly exponential rate of CO2
increase (Loehle [31]) and a logarithmic saturation effect of Fig. (2A) shows that the model fit from 1850 to 1950
GHG concentration on forcing, a quasi-linear climatic effect reproduces the modulation of the temperature up to 1942 and
of rising GHG could be expected. In any case, this upward has random residuals over this period (Fig. 2B). It is very
78 The Open Atmospheric Science Journal, 2011, Volume 5 Loehle and Scafetta

unlikely that the 60 and 20-year cycles would match the maxima around 2000, as Fig. (3B) shows. These cycles
pattern of temperature over this period by chance; thus, the would imply a combined warming of about 0.32oC on the
empirical model calibrated on the temperature data before total observed warming of about 0.5oC since 1970. Thus,
1950 has the capability of forecasting the multidecadal about 60% of the warming observed since 1970 can be
oscillations of the temperature after 1942, which we do next. associated with natural 20- and 60-year multidecadal cycles.
Table 1. Regression Coefficients of the Harmonic Model Plus 3. MULTI-SECULAR EVIDENCE OF A QUASI 60-
Linear Trend Depicted in Eq. (1) Used in Fig. (1, fit YEAR CLIMATIC CYCLE
1850-2010) and in Fig. (2, 1850-1950), Respectively
Here we list empirical studies supporting the existence of
a quasi 60-year cycle in the climate system. A 20-year cycle
Case 1 Case 2
with smaller amplitude will necessarily be harder to detect in
A 0.121 +/- 0.015 C 0.121 +/- 0.016 C
geologic records, and is not evaluated here. Also, note that
climate may be characterized by other cycles with periods
B 0.034 +/- 0.016 C 0.041 +/- 0.016 C longer than 60-year (e.g., Loehle and Singer [32]). Because
C 0.0042 +/- 0.0002 C/year 0.0016 +/- 0.0004 C/year complex interference patterns may emerge through the
superposition of all cycles and volcano eruptions can further
D -0.297 +/- 0.013 C -0.317 +/- 0.011 C disrupt the signal, perfect 60-year harmonics may not be
T1 1998.62 +/- 1.2 year 1998.58 +/- 1.3 year easily visible in the records. Noisy geologic data may not
show a signal even if one is present.
T2 2000.98 +/- 1.5 year 1999.65 +/- 1.3 year
Note that the harmonic coefficients A, B, T1 and T2 are compatible within the error of While spectral coherence of the Hadley historical data
measure. Parameter A is the 60-year cycle amplitude and B the 20-year cycle and solar data has been established (Scafetta [18]), does this
amplitude.
effect extend before 1850? Before 1850 global instrumental
temperature records are not available. Thus, it is possible to
The above finding is also confirmed by two recent use only a few historical European temperature records
studies that show a similar linear anthropogenic warming dating back since the 17th century and several paleoclimatic
trend after about 1950 (Thompson et al. [24]) and about records. These records may present several problems and
1960 (Hegerl et al. [23]. Both studies also find a small may be a poor proxy for the real global climate.
residual warming in the pre-1950 or 1960 period but do not It has been recently noted that a 500 year temperature
attribute this trend to human influences. The magnitude of reconstruction in the Mediterranean basin (Spain, France,
the Thompson et al. [24] residual post-1950 trend (i.e., the Italy) by means of documentary data and instrumental
anthropogenic component) is similar to our result. These observations suggests a dominant ~60 year oscillation
studies attempt to filter out sources of extraneous noise, such (Camuffo et al. [33]). Moreover, there are multidecadal
as ENSO events and volcanoes. They make many periods of high and low ENSO-type activity evident in
assumptions, which are not needed in the present study in historical records going back several hundred years (e.g.,
which short-term perturbations are simply noise relative to Biondi et al. [34]; Mantua and Hare [35]; Mantua et al. [36];
the 60 year cycle and the long-term natural trend. Minobe [37-39]; Patterson et al. [40]; Shabalova and Weber
To verify the combined effect of natural plus [41]; Wiles et al. [42]; figure 5 in Gergis and Fowler [43]).
anthropogenic forcings, a full model was constructed, with The long-term ENSO-type signal has been shown to be
the anthropogenic linear trend obtained by fitting the residual quasi-periodic, with period of 50 to 70 years, in records such
from 1950 to 2010 assumed to be in effect since 1942. This as the Greenland ice core, bristlecone pine ring widths, and
model (Fig. 3A) fits the entire 160-year record better than sea sediment records of fish abundance going back
any model we have seen. Fig (3B) depicts the two harmonics thousands of years (Klyashtorin and Lyubushin [26, 44];
at 60 and 20-year period and the linear natural trend (fit Klyashtorin et al. [45]). Over these long periods, cycles of
1850-1950) added to the anthropogenic linear trend (fit of 50 to 70 years were shown, using moving window spectral
the residual in 1950-2010) since 1942. decomposition, to increase in strength over the last 600 to
1000 years, reaching a peak strength in the 20th century.
It would thus appear that the modeling approach used
Klyashtorin and Lyubushin [26] and Klyashtorin et al. [45]
here has captured the essential features of the climate from
also showed that over the past 100+ years an average 60 year
1850 to 2010. A model based on 100 years of data (from
global temperature cycle is strongly coherent with the Pacific
1850 to 1950) shows stationary fluctuating residuals (Fig.
Decadal Oscillation (PDO), the zonal mode of the Northern
2B). The extrapolation period 1950 to 2010 shows an
Hemisphere Atmospheric Circulation Index, the Aleutian
upward linear pattern of residuals, which we interpret as a
Low Pressure Index, flood levels in the Neva River (Russia),
signal of anthropogenic warming. The slope of this
and precipitation in Oregon, among others. In addition, the
anthropogenic warming trend is 0.66C/century since 1942.
ups and downs of major ocean fish stocks are almost
Fig. (3A, B) show that about 50 +/- 10% of the 0.8oC global
completely explained by this ~60 year climate cycle (e.g.,
surface warming observed from 1850 to 2010 is likely the
Klyashtorin and Lyubushin [26]; Mantua et al., [36]). The
result of a natural warming trend recovery since the LIA plus
Atlantic Multidecadal Oscillation also closely tracks this
the combined effects of 20- and 60-year natural cycles. In
approximately 60-year cycle (figure 2 in Levitus et al. [46]).
particular, the sudden and alarming warming increase
Wiles et al. [42] found evidence for the persistent effect of
observed from 1970 to 2000 is mostly a consequence of the
the PDO on climate in Alaska over the past 1000 years. A
combination of the 20- and 60-year natural cycles. Both
quasi-60 year periodicity is found in secular monsoon
cycles were at their minima in about 1970 and at their
Climate Change Attribution Using Empirical Decomposition of Climatic Data The Open Atmospheric Science Journal, 2011, Volume 5 79

Fig. (3). A) Full reconstruction model, which uses Eq. 1 (= natural modulation) fit from 1850 to 1950 plus the additional residual upward
linear trend (=anthropogenic modulation) since 1942. B) Components of the model.

rainfall records from India (Agnihotri and Dutta [47]). (Patterson et al. [40]). Ogurtsov et al. [50] found strong
Yousef [48] showed a good anti-correlation between ENSO evidence for the cycle, such as a 60 to 64 year period in 10Be,
14
events and the Wolf-Gleissberg cycle over 300 years. A clear C and Wolf number over the past 1000 years, which may
quasi 60-year cycle is observed in the global sea level rise indicate a solar origin. The solar system oscillates with a 60-
record since 1700 (Jevrejeva et al. [49]). A 50 to 80 year year cycle due to the Jupiter/Saturn three-synodic cycle and
cycle over hundreds of years has also been found in to a Jupiter/Saturn beat tidal cycle (Scafetta [18]).
sedimentary records in the northeast Pacific and their cause
Fig. (4) depicts two multi-secular climate records
has been related to solar and cosmic ray activity cycles showing multiple quasi-60 year large oscillations since 1650.
80 The Open Atmospheric Science Journal, 2011, Volume 5 Loehle and Scafetta

Fig. (4). A) G. Bulloides abundance variation record found in the Cariaco Basis sediments in the Caribbean sea since 1650 [Black et al.,
1999]. B) tree-ring chronologies from Pinus Flexilis [MacDonald and Case, 2005] as an index to the PDO. Both records show five large
quasi 60-year cycles since 1650.

Fig. (4A) depicts the abundance of G. Bulloides (an indicator PDO based on tree-ring chronologies from Pinus flexilis in
of surface temperatures) found in the Cariaco Basin the United States West Coast (MacDonald and Case [52]):
sediments in the Caribbean sea since 1650 (Black et al. the best sinusoidal fit gives a period of about 59.5 +/- 4
[51]): the best sinusoidal fit gives a period of about 61.5 +/- years. Thus, both records show a clear 60-year cycle. These
4 years. This record, which correlates well with time series cases negatively related to temperature due to the type of
of solar output, is an indicator of the trade wind strength in proxy.
the tropical Atlantic Ocean and of the North Atlantic Ocean
atmosphere variability. Fig. (4B) depicts the reconstructed
Climate Change Attribution Using Empirical Decomposition of Climatic Data The Open Atmospheric Science Journal, 2011, Volume 5 81

A potential source of data for evaluating this cycle can be this phenomenon. It is not known how well the land surface
found in multiproxy temperature reconstructions (e.g., temperature records are corrected for the UHI effect.
Moberg et al., [29]). Unfortunately, most of these
The UHI effect arises not from the mere fact that a city is
reconstructions must be rejected as accurate data sources: in
warmer than a rural site (e.g., Klysik and Fortuniak [58]),
fact, paleoclimatic temperature reconstructions appear quite which is of no consequence for calculating trends, but from
different from each other (North et al. [53]). Many rely on
the effect at a given weather station location as the city
tree ring data, but it has been shown (e.g., Loehle [54]) that
grows around it, causing a gradual warming bias over time
tree ring data, as well as other proxy records have issues that
([e.g., Bhm [59]; Magee et al. [60]).
need to be resolved before they can be used to develop
reliable climate histories.. In other cases, multiple proxies While many studies have documented UHI effects at the
either are dated at wide intervals such as 100+ years, city level, few have done so at larger scales. The contribution
precluding detection of a multi-decadal signal, or suffer from of UHI effects at larger scales arises from the fact that a
large dating error, which in combined series will lead to a large proportion of weather stations have been established to
smearing of multi-decadal peaks (Loehle [55]). serve cities and airports, with few in remote locations. Thus,
UHI effects contaminate the instrumental temperature
Given that this ~60 year periodic climate pattern has such
record. In a landmark study, Jones et al. [61] evaluated the
a long empirical history with some evidence for links to solar UHI effect for China and found that 39.6% of the
activity and solar system oscillations, it cannot be attributed
instrumental warming from 1951 to 2004 was contributed by
to human activity and cannot be simply random noise.
UHI and related effects. In another set of studies, McKitrick
Therefore, we are justified in factoring it out of the historical
and Michaels [62, 63] and McKitrick [64] showed that local
record in order to discern the human influence on climate.
warming (deviations from global trends) was proportional to
We next evaluate additional components that can be
national economic activity, a surrogate for urbanization, and
potentially factored out and, then, develop a final model that that this effect is real and not caused by, for example,
is extrapolated to 2100.
atmospheric circulation patterns. They estimated that about
4. TROPOSPHERIC AEROSOLS half of the land warming claimed by the IPCC to be UHI and
land use related. In an analysis of trend differences between
A major factor often considered crucial for proper satellite and surface data, Klotzbach et al. [57] suggested that
climate modeling is the influence of tropospheric aerosols from 30 to 50% (based on UAH satellite data) of the
that result from industrial activity, which are often assumed instrumental land surface warming trend from 1979 to 2008
to have been a major cooling influence in the 1950s through was spurious. Kalnay and Cai [65] found 0.27C per century
the 1970s due to their reflective properties (although of the warming trend over the continental United States to be
Schiermeier [56] argues that the aerosol extent and forcing due to land use change effects. Other studies (e.g., Christy
data are shaky at best). However, because the climate et al. [66]; Fall et al. [67]) support these results.
appears to contain a ~60 year cycle for several centuries,
much of the cooling may have been induced by this cycle, The above studies suggest that the contribution of UHI
which was in its descending trend during this period. In fact, warming to land warming is difficult to determine accurately
the residuals from the pre-1950 model (Fig. 2B) over the and to filter from the global surface temperature record. The
decades of the 1950s through 1970s show not a decline but a warming since 1950 might be lower than what is currently
constant linear rise for the anthropogenic component, which believed and the climate effects of the anthropogenic
is inconsistent with a strong cooling effect over this period. forcings may be overestimated also for this reason. Thus, the
Our full model (including the linear anthropogenic effect modeled upward warming trend attributions depicted in
since 1940) shows a cooling in the 1950s through 1970s Fig. (3A) should be interpreted as an upper limit for an
without any forcing from tropospheric aerosols (Fig. 3A). anthropogenic (GHG plus aerosol) contribution to climate
Note that the amplitude of this 60-year cycle has been change.
determined fitting the data before 1950. Moreover, the 6. 21ST CENTURY FORECAST OF CLIMATE
periods 1880-1910 (slight cooling) and 1910-1940 CHANGE
(significant warming) are statistically equivalent to the
periods 1940-1970 (slight cooling) and 1970-2000 (strong Based on the analysis in this study, it is possible to make
warming), respectively. Thus, while tropospheric aerosols a preliminary forecast of future trends that may be reliable
might have contributed to the cooling between 1940 and for the coming few decades. We believe this is possible
1970, most of the observed cooling is likely associated with because the natural variability may be assumed in first
the 60-year cycle, which was in the declining phase during approximation to be a sum of a possible upward linear trend
that period. started since the Little Ice Age, which may still continue
during the next decades although it is likely part of a
5. LAND USE EFFECTS ON TRENDS millennial long natural cycle which includes the Medieval
The final factor considered is the urban heat island (UHI) Warm Period and the Little Ice Age (see Loehle and Singer
effect, including other land use/land cover (LULC) changes [32]), plus the periodic multidecadal component we have
(e.g., Klotzbach et al. [57]). The UHI effect has long been detected.
known (e.g., Klysik and Fortuniak [58]). It results because That a natural trend may not exceed the above estimated
urban structures have lower albedo, which raises upward linear trend from 1850 to 1950 may be indirectly
temperatures, and retain heat, which raises nighttime suggested by the fact that since 1930 the rate of sea-level is
minima. In addition, there is less evaporative cooling in a slowing slightly (Houston and Dean [68]), which may
city. Herein, we discuss the current uncertainty relative to suggest that a millenarian natural climate cycle is turning
82 The Open Atmospheric Science Journal, 2011, Volume 5 Loehle and Scafetta

Fig. (5). Forecast models. Thick line is the global surface temperature. Dashed thick line is the total (natural plus anthropogenic) model
reconstruction. Dashed light line is the continuation of the modeled natural component alone.

down. Interestingly, the measured deceleration of the sea- 7. CAN SOLAR ACTIVITY EXPLAIN THE TEMPE-
level rise from 1930 to 2010 occurred despite the strong RATURE MODULATION?
positive acceleration of anthropogenic GHG emissions
during the same period. Moreover, the anthropogenic If anthropogenic forcing has been overestimated by the
component of the warming (GHG and aerosol forcings models adopted by the IPCC, it is necessary that other
and/or LULC and UHI effects) seems to have resulted in a natural forcings such as the solar forcing have been
linear additional warming trend from 1942 through 2010, underestimated by the same models to balance the secular
which matches the long-term climate model outputs (IPCC warming trend observed in the temperature record.
[2]). The combined (20 and 60-year cycle plus two linear Herein we investigate whether solar activity can explain
trends) model depicted in Fig. (3A) matches the 160-year a quasi 60-year modulation of temperature and at least part
historical record exceptionally well. Thus, extrapolating the of the upward trend by expanding the discussion presented in
model another 90 years may be useful. Scafetta [8]. The task is not simple, because past total solar
The result (Fig. 5) is a continued warming with irradiance (TSI) variation is not known with certainty
oscillations to a high in 2100 of about 0.6 C above 2000 (Supplemental Information b). We develop an enhanced
values. However, if part of the linear warming trend is due to version of the Scafetta [8] model in Supplemental
UHI and LULC changes, the anthropogenic effect due to Information c, which uses the TSI record as a proxy for
GHG and aerosol would be smaller. Moreover, the full determining a phenomenological signature of the total solar
anthropogenic effect of 0.66 C/century would be masked by effect on climate. This model empirically evaluates the
the natural multidecadal 20 and 60-year cycles until 2040. climate sensitivity to solar changes and, therefore, it
Note that for the full model the downturn in temperature automatically takes into account natural amplification
evident in the satellite observations after 2000 (Loehle [69]) mechanisms such as those due to cloud and GHG feedbacks.
is present. This downturn in the temperature since 2000 was Figure Suppl. (2) in Supplemental Information c shows
not reproduced by any projection of the climate models that the proxy model (green) presents a slight increase
adopted by the IPCC [2] although it is predicted by the between 1850 and 1880/90, a decrease from 1880/1890 to
celestial empirical model of climate recently proposed by 1910, an increase from 1910 to 1940/50, a decrease from
Scafetta [18]. The above estimate of a possible warming of 1940/50 to 1970/80, and an increase from 1970/1980 to
only 0.6 C above 2000 should be compared against the 2000, and a decrease after 2000. Thus, it shows a quasi 60-
IPCC projections that claim a monotonic anthropogenic year cycle, which is clearly present in the gravitational
warming with an average global value in 2100 of about 3C oscillations of the solar system (Scafetta [18]). The latter 30-
above 2000 values for the likely scenarios and even more for year pattern since 1980 is perfectly compatible with the
the upper bound scenarios. ACRIM total solar irradiance satellite measurements that
show an increase from 1980 to 2000 and a decrease from
2000 to 2010 (Scafetta and Willson [70]). This is an
Climate Change Attribution Using Empirical Decomposition of Climatic Data The Open Atmospheric Science Journal, 2011, Volume 5 83

important result because some solar scientists currently timescales and is apparent in cosmogenic isotopes. For
believe that the ACRIM trend is instrumental and not of example, Ogurtsov et al. [50] found strong evidence for solar
solar origin (Frhlich [71]; Lockwood and Frohlich [72]). cycles with 60 to 64, 80 to 90, 128, 205, and 1020 year
periods in 10Be, 14C, and Wolf number over the past 1000
From the above, it appears that these TSI proxy models
can approximately reproduce the quasi 60-year modulation years. Mazzarella [73] also found a clear link between the
60-year solar cycle and climate oscillations during the last
of the temperature. In addition, they contain an upward trend
150 years.
that is part of a multisecular solar cycle responsible for much
of the cooling from the Medieval Warm Period to the Little 8. DISCUSSION AND CONCLUSION
Ice Age and the warming from the Little Ice Age to the
current warm period. To better show this we applied the Herein we have argued that the global surface
empirical temperature model of Scafetta [8] to the novel instrumental temperature record, as well as longer term
green TSI proxy curve and to the average of the three TSI records, contain a quasi 60-year natural cycle. This cycle
curves represented by the black curve in the Supplemental may be associated with a low frequency modulation of the
figure to obtain values for forcing. We plotted the two solar ocean oscillations such as the ENSO signal. It appears to be
phenomenological signatures in Fig. (6) against the global present in solar and astronomical records as well, although a
surface temperature record. perfect match is difficult to obtain because of the
uncertainties about the current total solar irradiance proxy
Fig. (6) confirms that there is good agreement, especially models. Better solar irradiance models or alternative solar
after 1900, between the empirical ~60 year modulation of system direct forcings (Scafetta, [18]) may be required to
temperature (Fig. 3A, thick line) and the climate signature find a perfect match.
that would be imprinted by solar multidecadal variation. If
the correspondence is not rigorous it is because the available Thus, this quasi 60-year cycle observed in the
TSI proxy models can only approximately reproduce the real temperature record likely has an astronomical/solar and,
multidecadal TSI variation and more research is needed on therefore, natural origin. The weaker quasi 20-year cycle is
this topic. It also appears that a modest upward trend over not as easily detected in geologic records but it is clearly
the period since 1850 of 0.11C/century is reproduced by the detected in the global surface instrumental records (Scafetta,
TSI reconstructions. This is a critical point because the rising [18]). The projected rate of temperature increase due to
temperatures from 1850 to 1950 (or 1942) could be due to human influences in our analysis suggests a value for climate
the secular increase of solar activity observed since the 18th sensitivity. Using an estimate of doubling time of
century (Scafetta [8]). Thus, the cause of both the 60-year greenhouse gases based on historical doubling times for
cycle modulation plus an upward trend since 1850 is likely carbon dioxide (Loehle [13]) as a proxy, the anthropogenic
to be linked to the multidecadal and multisecular oscillatory warming resulting from doubled forcing would be at least
nature of the solar variation. This is evident over long two to three times less than previously thought, that is on
average about 1-1.5C or less from our linear model, which

Fig. (6). Global surface temperature against the phenomenological solar signature obtained with the model by Scafetta [2009] applied to the
green and black TSI proxy model curves depicted in Supplemental Information c.
84 The Open Atmospheric Science Journal, 2011, Volume 5 Loehle and Scafetta

may also suggest the presence of a slight negative feedback recently claimed by Rockstrm et al. [77]. Moreover,
to CO2 in the climate system. because the 60-year natural cycle will be in its cooling phase
for the next 20 years, global temperatures will probably not
The models adopted by the IPCC [2] assume that more
increase for the next few decades in spite of the important
than 90% of the warming observed since 1850 to 2000 and
about 100% of the warming observed since 1970 has been role of human emissions (see Fig. 5A), as predicted by
multiple studies (reviewed in Loehle [69]; Scafetta [18]).
caused by anthropogenic forcing. This information can be
deduced from a direct analysis of the figures published there,
such as Fig. (9.5a, 9.5b), which claim that natural forcing
alone (solar plus volcano) would have induced a cooling CONFLICT OF INTEREST
since 1970. However, these climate models are not able to No outside funding was used to conduct this work. No
reproduce decadal and multi-decadal natural cycles (Scafetta, financial interests of conflicts of interest are involved.
[18]). We showed that the Hadley global anomaly data from
1850 to 1950 are matched closely by a periodic model with ACKNOWLEDGEMENTS
periods 60 and 20 years plus a linear trend. The residuals We thank anonymous reviewers for detailed helpful
from this model after 1950 are strictly linear, matching other comments.
attribution studies (e.g., Hegerl et al. [23]; Thompson et al.
[24]). Based on this result, our full model (natural plus SUPPLEMENTARY MATERIAL
anthropogenic effects after 1942) closely matched the full This article contains supplementary material and it can be
160-year record. It had far higher pattern accuracy than any viewed at publishers website.
traditional analytical computer general circulation climate
model output. We further found no clear evidence for a REFERENCES
significant cooling effect due to troposphere aerosols, since [1] Brohan P, Kennedy JJ, Harris I, Tett SFB, Jones PD. Uncertainty
our model shows cooling in the 1950s through 1970s without estimates in regional and global observed temperature changes: a
invoking this mechanism, and there are no unexplained new dataset from 1850. J Geophys Res 2006; 111: D12106.
residuals corresponding to the period of assumed influence [2] Intergovernmental Panel on Climate Change (IPCC). Climate
change 2007: the physical science basis. contribution of working
of this factor. Comparing the temperature rise from the group I to the fourth assessment report of the intergovernmental
1970s to 2010 (~0.5C) due to the 60 and 20-year cycles panel on climate change. Solomon S, Qin D, Manning M, et al.,
(~0.3C) with that evident in the model residuals (Fig. 2B, Eds, Cambridge University Press, Cambridge 2007.
~0.2 C), we can see that more than half of the warming [3] Andronova NG, Schlesinger, ME. Causes of global temperature
changes during the 19th and 20th centuries. Geophys Res Lett
since the 1970s is the result of the timing of the upturn of the 2000; 27: 2137-40.
20 and 60 year cycles. Our empirical model further predicts a [4] Crowley TJ. Causes of climate change over the past 1000 years.
lack of warming since about 2000, in contrast to IPCC model Science 2000; 289: 270-7.
projections of steady linear warming due to anthropogenic [5] Damon PE, Jirikowic JL. Solar forcing of global climate change? in
forcing alone of about +2.3C/century from 2000 to 2010 radiocarbon after four decades. Taylor RE, Long A, Kra RS, Eds.
Springer-Verlag, Berlin 1992; pp. 117-29,.
(see IPCC 2007 figure SPM.5), which is a 3.5 time larger [6] Hoyt DV, Schatten KH. A discussion of plausible solar irradiance
rate than the +0.66C/century warming that we have variations, 1700-1992. J Geophys Res 1993; 98: 18895-906.
estimated above. [7] Lean J, Beer J, Bradley R. Reconstruction of solar irradiance since
1610: implications for climate change. Geophys Res Lett 1995; 22:
In conclusion, we have shown that the effect of natural 3195-8.
oscillations is critical for proper assessment of anthropogenic [8] Scafetta N. Empirical analysis of the solar contribution of global
impacts on the climate system. Since the rapid increase in mean air surface temperature change. J Atmos Solar-Terrest Phys
2009; doi:10.1016/j.jastp.2009.07.007.
temperature in the 1980s and 1990s is due partially to natural [9] Serreze MC, Walsh JE, Chapin III FS, et al. Observational
cycles, then any model-based estimate of climate sensitivity evidence of recent change in the northern high-latitude
based on elevated greenhouse gases will be too high, a point environment. Clim Change 2000; 46: 159-207.
also made by Scafetta and West [74], Scafetta [8, 18], and [10] Stott PA, Tett SFB, Jones GS, Allen MR, Mitchell JFB, Jenkins
Klyashtorin and Lyubushin [44]. The same applies to GJ. External control of 20th Century temperature by natural and
anthropogenic forcings. Science 2000; 290: 2133-7.
attribution studies that focus on these decades. Proper [11] Tett SFB, Stott PA, Allen MR, Ingram WJ, Mitchell JFB. Causes
consideration of these cycles and of longer natural cycles is of twentieth-century temperature changes near the Earths surface.
also critical for detecting underlying trends. For example, a Nature 1999; 399: 569-72.
millenarian or longer period cycles might explain the pattern [12] Kiehl JT. Twentieth century climate model response and climate
sensitivity. Geophys Res Lett 2007; 34: L22710.
of the Medieval Warm Period and Little Ice Age (Loehle and [13] Visser H, Folkert RJM, Hoekstra J, de Wolff JJ. Identifying key
Singer [32]) as well as warming since about 1800. sources of uncertainty in climate change projections. Clim Change
2000; 45: 421-57.
It is also worth noting that by starting in a cycle trough in [14] Wyant MC, Khairoutdinov M, Bretherton CS. Climate sensitivity
1900 and ending on a cycle peak in 2000, graphical and cloud response of a GCM with a superparameterization.
depictions of warming in IPCC and other reports give an Geophys Res Lett 2006; 33: L06714.
exaggerated depiction of the rate of warming, a point noted [15] Rasmussen SO, Andersen KK, Svensson AM, et al. A new
by Karln [75] and Soon et al. [76]. By further failing to Greenland ice core chronology for the last glacial termination. J
Geophys Res 2006; 111: D06102.
mention solar effects and warm bias in the instrumental [16] Bendandi R. Un Principio Fondamentale dell'Universo. S.T.E.:
record, it is implied that almost all the warming of about Faenza, Italy 1931.
0.8C over this period is human caused, when really no more [17] Hung C-C. Apparent relations between solar activity and solar tides
than 0.4C is likely to be. Reporting such a lower number caused by the planets. Report NASA/TM-2007-214817, NASA
2007.
would probably cause much less alarm than what was
Climate Change Attribution Using Empirical Decomposition of Climatic Data The Open Atmospheric Science Journal, 2011, Volume 5 85

[18] Scafetta N. Empirical evidence for a celestial origin of the climate [45] Klyashtorin LB, Borisov V, Lyubushin A. Cyclic changes of
oscillations and its implications. J Atmos Solar-Terrest Phys 2010. climate and major commercial stocks of the Barents Sea. Mar Biol
doi:10.1016/j.jastp.2010.04.015. Res 2009; 5: 4-17.
[19] Wolff CL, Patrone, PN. A new way that planets can affect the sun. [46] Levitus S, Matishov G, Seidov D, Smolyar I. Barents Sea
Solar Phys 2010; 266: 227-46. multidecadal variability. Geophys Res Lett 2009; 36: L19604.
[20] Kirkby J. Cosmic rays and climate. Surv Geophys 2007; 28: 333- [47] Agnihotri R, Dutta K. Centennial scale variations in monsoonal
75. rainfall (Indian, east equatorial and Chinese monsoons):
[21] Tinsley BA. The global atmospheric electric circuit and its effects Manifestations of solar variability. Curr Sci 2003; 85: 459-63.
on cloud microphysics. Rep Prog Phys 2008; 71: 066801. [48] Yousef SM. 80-120 year long-term solar induced effects on the
earth, past and predictions. Phys Chem Earth 2006; 31: 113-22.
[22] Svensmark H, Bondo T, Svensmark J. Cosmic ray decreases affect [49] Jevrejeva S, Moore JC, Grinsted A, Woodworth PL. Recent global
atmospheric aerosols and clouds. Geophys Res Lett 2009; 36: sea level acceleration started over 200 years ago? Geophys Res Lett
L15101. 2008; 35: L08715.
[23] Hegerl GC, Jones PD, Barnett TP. Effect of observational sampling [50] Ogurtsov MG, Nagovitsyn YA, Kocharov GE, Jungner H. Long-
error on the detection of anthropogenic climate change. J Clim period cycles of the suns activity recorded in direct solar data and
2001; 14: 198-207. proxies. Solar Phys 2002; 211: 371-94.
[24] Thompson DWJ, Wallace JM, Jones PD, Kennedy JJ. Identifying [51] Black DE, Peterson LC, Overpeck JT, et al. Eight centuries of
signatures of natural climate variability in time series of global- North Atlantic Ocean atmosphere variability. Science 1999; 286:
mean surface temperature: methodology and insights. J Clim 2009; 1709-13.
22: 6120-41. [52] MacDonald GM, Case RA. Variations in the Pacific Decadal
[25] Hansen J, Sato M, Ruedy R, et al. Dangerous human-made Oscillation over the past millennium. Geophys Res Lett 2005; 32:
interference with climate: a GISS modelE study. Atmos Chem Phys L08703.
2007; 7: 2287-312. [53] North G, Biondi F, Bloomfield P, et al. Surface temperature
[26] Klyashtorin LB, Lyubushin AA. Cyclic climate changes and fish reconstructions for the last 2,000 years. Washington DC, USA:
productivity. Vniro Publishing: Moscow 2007. National Academy Press 2006.
[27] Eddy J A. The maunder minimum. Science 1976; 192: 1189-202. [54] Loehle C. A mathematical analysis of the divergence problem in
[28] Scafetta N, West BJ. Phenomenological reconstructions of the solar dendroclimatology. Clim Change 2009; 94: 233-45.
signature in the NH surface temperature records since 1600. J [55] Loehle C. Estimating climatic timeseries from multi-site data
Geophys Res 2007; 112: D24S03. afflicted with dating error. Math Geol 2005; 37: 127-40
[29] Moberg A, Sonechkin DM, Holmgren K, Datsenko NM, Karln W. [56] Schiermeier Q. The real holes in climate science. Nature 2010; 463:
Highly variable northern hemisphere temperatures reconstructed 284-7.
from low- and high-resolution proxy data. Nature 2005; 433: 613- [57] Klotzbach PJ, Pielke RA, Sr, Pielke RA, Jr, Christy JR, McNider
7. RT. An alternative explanation for differential temperature trends
[30] Mann ME, Bradley RS, Hughes MK. Northern hemisphere at the surface and in the lower troposphere. J Geophys Res 2009.
temperatures during the past millennium: inferences, uncertainties, doi:10.1029/2009JD011841.
and limitations. Geophys Res Lett 1999; 26: 759-62. [58] Klysik K, Fortuniak K. Temporal and spatial characteristics of the
[31] Loehle C. The estimation of historical CO2 trajectories is urban heat island of Lodz, Poland. Atmos Environ 1999; 33: 3885-
indeterminate: Comment on A new look at atmospheric carbon 95.
dioxide. Atmos Environ 2010; 44: 2257-9. [59] Bhm R. Urban bias in temperature time series: a case study for the
[32] Loehle C, Singer SF. Holocene temperature records show city of Vienna, Austria. Clim Change 1998; 38: 113-28.
millennial-scale periodicity. Can J Earth Sci 2010; 47: 1327-36. [60] Magee N, Curtis J, Wendler G. The urban heat island effect at
[33] Camuffo D, Bertollin C, Barriendos M, et al. 500 year temperature Fairbanks, Alaska. Theor Appl Climatol 1999; 64: 39-47.
reconstruction in the Mediterranean Basin by means of [61] Jones PD, Lister DH, Li Q. Urbanization effects in large-scale
documentary data and instrumental observations. Clim Change temperature records, with an emphasis on China. J Geophys Res
2010; 101: 169-99. 2008; 113: D16122.
[34] Biondi F, Gershunov A, Cayan DR. North Pacific decadal climate [62] McKitrick R, Michaels PJ. A test of corrections for extraneous
variability since 1661. J Clim 2001; 14: 5-10. signals in gridded surface temperature data. Clim Res 2005; 26:
[35] Mantua N, Hare S. The pacific decadal oscillation. J Oceanogr 159-73.
2002; 58: 35-44. [63] McKitrick, RR, Michaels PJ. Quantifying the influence of
[36] Mantua N, Hare S, Zhang Y, Wallace J, Francis R. A Pacific anthropogenic surface processes and inhomogeneities on gridded
interdecadal climate oscillation with impacts on salmon production. global climate data. J Geophys Res 2007; 112: D24S09.
Bull Am Meteorol Soc 1997; 78: 1069-79. [64] McKitrick R. Atmospheric circulations do not explain the
[37] Minobe S. A 50-70 year climatic oscillation over the North Pacific temperature-industrialization correlation. Statistics, Politics, and
and North America. Geophys Res Lett 1997; 24: 683-6. Policy 2010; 1: doi:10.2202/2151-7509.1004.
[38] Minobe S. Resonance in bi-decadal and penta-decadal climate [65] Kalnay E, Cai M. Impact of urbanization and land-use change on
oscillations over the North Pacific: role in climatic regime shifts. climate. Nature 2003; 423: 528-31.
Geophys Res Lett 1999; 26: 855-8. [66] Christy JR, Norris WB, Redmond K, Gallo KP. Methodology and
[39] Minobe S. Spatio-temporal structure of the pentadecadal climate results of calculating central California surface temperature trends:
oscillations over the North Pacific. Prog Oceanogr 2000; 47: 381- Evidence of human-induced change? J Clim 2006; 19: 548-63.
408. [67] Fall S, Niyogi D, Gluhovsky A, Pielke RA, Sr, Kalnay E, Rochon
[40] Patterson RT, Prokoph A, Chang AS. Late holocene sedimentary G. Impacts of land use land cover on temperature trends over the
response to solar and cosmic ray activity influenced climate continental United States: assessment using the North American
variability in the NE Pacific. Sediment Geol 2004; 172: 67-84. regional reanalysis. Int J Clim 2009; doi:10.1002/joc.1996.
[41] Shabalova M, Weber SIH. Patterns of temperature variability on [68] Houston JR, Dean RG. Sea-level acceleration based on U.S. tide
multidecadal to centennial timescales. J Geophys Res 1999; 105: gauges and extensions of previous global-gauge analyses. J Coast
31023-41. Res 2011; doi: 10.2112/JCOASTRES-D-10-00157.1
[42] Wiles GC, DArrigo RD, Villalba R, Calkin PE, Barclay DJ. [69] Loehle C. Trend analysis of global satellite temperature data.
Century-scale solar variability and Alaskan temperature change Energy Environ 2009; 20: 1087-98.
over the past millennium. Geophys Res Lett 2004; 31: L15203. [70] Scafetta N, Willson RC. ACRIM-gap and TSI trend issue resolved
[43] Gergis JL, Fowler AM. A history of ENSO events since A.D. 1525: using a surface magnetic flux TSI proxy model. Geophys Res Lett
Implications for future climate change. Clim Change 2009; 92: 2009; 36: L05701.
343-87. [71] Frhlich C. Solar irradiance variability since 1978: revision of the
[44] Klyashtorin LB, Lyubushin AA. On the coherence between PMOD composite during solar cycle 21. Space Sci Rev 2006; 125:
dynamics of the world fuel consumption and global temperature 53-65.
anomaly. Energy Environ 2003; 14: 773-82. [72] Lockwood M, Frhlich C. Recent oppositely directed trends in
solar climate forcings and the global mean surface air temperature:
86 The Open Atmospheric Science Journal, 2011, Volume 5 Loehle and Scafetta

Proceedings of the Royal Society of London, Series A 2007; 463: [75] Karln W. Recent global warming: an artifact of a too-short
2447-60. temperature record? Ambio 2005; 34: 263-4.
[73] Mazzarella A. The 60-year solar modulation of global air [76] Soon WW-H, Legates DR, Baliunas SL. Estimation and
temperature: the Earth's rotation and atmospheric circulation representation of long-term (>40 year) trends of Northern-
connection. Theor Appl Climatol 2007; 88(3-4): 193-9. Hemisphere-gridded surface temperature: a note of caution.
[74] Scafetta N, West BJ. Is climate sensitive to solar variability? Phys Geophys Res Lett 2004; 31: L03209.
Today 2008; 3: 50-51. [77] Rockstrm J, Steffen W, Noone K, et al. A safe operating space for
humanity. Nature 2009; 461: 472-5.

Received: December 27, 2010 Revised: May 16, 2011 Accepted: June 4, 2011

Loehle and Scafetta; Licensee Bentham Open.


This is an open access article licensed under the terms of the Creative Commons Attribution Non-Commercial License (http: //creativecommons.org/licenses/by-
nc/3.0/) which permits unrestricted, non-commercial use, distribution and reproduction in any medium, provided the work is properly cited.
Criteria for assessing climate change impacts on ecosystems
Craig Loehle
National Council for Air and Stream Improvement Inc., Washington Street, Naperville, Illinois, USA

Keywords Abstract
Biodiversity, climate envelope model,
extinction risk, General Circulation Model, There is concern about the potential impacts of climate change on species and
impact assessment. ecosystems. To address this concern, a large body of literature has developed in
which these impacts are assessed. In this study, criteria for conducting reliable
Correspondence and useful assessments of impacts of future climate are suggested. The major de-
Craig Loehle, National Council for Air and cisions involve: clearly defining an emissions scenario; selecting a climate model;
Stream Improvement, Inc., 552 S Washington
evaluating climate model skill and bias; quantifying General Circulation Model
Street, Suite 224, Naperville, IL 60540
Tel: 630-579-1190; Fax: 630-579-1195;
(GCM) between-model variability; selecting an ecosystem model and assessing un-
E-mail: CLoehle@ncasi.org certainty; properly considering transient versus equilibrium responses; including
effects of CO2 on plant response; evaluating implications of simplifying assump-
Received: 05 May 2011; Revised: 09 June tions; and considering animal linkage with vegetation. A sample of the literature
2011; Accepted: 11 June 2011. was surveyed in light of these criteria. Many of the studies used climate simu-
lations that were >10 years old and not representative of best current models.
doi: 10.1002/ece3.7 Future effects of elevated CO2 on plant drought resistance and productivity were
generally included in growth model studies but not in niche (habitat suitability)
studies, causing the latter to forecast greater future adverse impacts. Overly simpli-
fied spatial representation was frequent and caused the existence of refugia to be
underestimated. Few studies compared multiple climate simulations and ecosys-
tem models (including parametric uncertainty), leading to a false impression of
precision and potentially arbitrary results due to high between-model variance. No
study assessed climate model retrodictive skill or bias. Overall, most current studies
fail to meet all of the proposed criteria. Suggestions for improving assessments are
provided.

type is currently found is characterized by some sort of sta-


Introduction tistical model, and this model is applied to a future climate
The potential future impacts of climate change are of increas- scenario (e.g., Aitken et al. 2008). In simulations, a tree, stand,
ing concern. Impacts on ecosystems could potentially affect population, or ecosystem growth model is applied to current
the capacity of natural systems to produce wood products, and scenario cases and the results are compared (e.g., Coops
crops, livestock, and game. There is also concern about im- and Waring 2011).
pacts on biodiversity in general and species extinctions in While the literature using these approaches is vast, it is
particular. not systematic or standardized. Many choices must be made
To assess these risks, a voluminous literature has appeared during the analysis that can lead to large and arbitrary differ-
in which ecosystem or species responses to future possible ences in study outcomes. For example, Cramer et al. (2001),
climates are evaluated. These studies often utilize the exper- using six dynamic global vegetation models, gave a range of
imental literature as a basis for longer term extrapolation. outcomes for the global carbon sink in 2100 under rising CO2
Two basic approaches have been used to evaluate long-term and climate change of 0.3 to 6.6 Pg C y1 , an immense range
natural system response (whereas crops can be evaluated ex- of 22. Verburg et al. (2011) showed that spatial data errors
perimentally and are not considered further here): habitat in mapped attributes can produce larger uncertainties than
suitability (niche) models and simulations. In habitat suit- the process under investigation. Further examples are cited
ability models, the climate space (with or without consider- in later sections herein. Furthermore, it will be shown herein
ation of soils, topography, etc.) where a species or vegetation that certain key factors have been left out of many analyses,

c 2011 The Author. Published by Blackwell Publishing Ltd. This is an open access article under the terms of the Creative 1
Commons Attribution Non Commercial License, which permits use, distribution and reproduction in any medium, provided
the original work is properly cited and is not used for commercial purposes.
Assessing Climate Change Impacts C. Loehle

potentially affecting their interpretation. Such issues can lead the A2 2001 scenario is projected to produce 3.8 C warm-
to the observed widespread failures of model forecasting (e.g., ing by 2100, the 2007 A2 scenario only projects 3.4 C global
PilkeyJarvis and Pilkey 2008). mean increase by 2100. Thus, studies using the same sce-
The purpose of this study, then, is to propose criteria nario published at earlier or later dates will not be strictly
for developing ecosystem assessments of response to climate comparable. In addition, model runs from 2001 show wider
change. These criteria involve the choices that must be made spread than those from the 2007 report. In particular, earlier
at each step of the assessment process. In some cases, it is studies are likely to show more impact and should be consid-
sufficient to make explicit the limits that a particular choice ered to be superseded by more recent ones. Model outputs
(e.g., emissions scenario) places on what can be inferred from from earlier scenarios and GCMs may still be available but
the study. In others, including or excluding a particular factor should no longer be used and will exaggerate impacts if used.
or model can create a particular bias that must be taken into A second criterion for utilizing scenarios is to distinguish
account. For example, if a particular climate model predicts between worst-case scenario and most likely cases. If the
too much or too little rainfall in the 20th century base runs worst-case scenario is used in the assessment, it should not
compared to actual rainfall, this will bias or distort results be stated that the simulated impact is what will happen,
of an ecosystem simulation. The presentation of criteria is but rather is a worst case. If evaluation of a likely scenario is
followed by an assessment of the extent to which the criteria desired, the appropriate case should be used.
have been addressed in recent literature.
Selecting a climate model
Ecosystem Assessment Criteria
Early (before 1995 or even up to 2000+) GCMs either did
The assessment of any ecosystem impact involves a series of not simulate precipitation or were known to do so poorly
steps, especially when long-term effects are being evaluated. (CorteReal et al. 1995). This includes most of the mod-
Failure to exercise care in this process can lead to underesti- els used in the 2001 IPCC report. For evaluating ecosystem
mation of uncertainty and even to arbitrary results. In order impacts, this limitation cannot be addressed by using a con-
to avoid such negative outcomes, a minimal set of criteria stant precipitation regime. If precipitation is held constant
is proposed for some critical steps in this process. More de- and temperature is increased, one obviously would predict
tailed criteria could be developed (e.g., for statistical testing), negative effects on plants. Theory predicts, in contrast, that
but this minimal set is sufficient for the present analysis. The a warmer climate will be accompanied by increased precipi-
criteria for decisions that must be made to put together a tation (Wentz et al. 2007; Schliep et al. 2010; Stephens et al.
complete and valid assessment involve the following: 2010). Whether increased evaporative demand will be bal-
(1) Clearly define emissions scenario. anced by increased precipitation is an open question and one
(2) Select climate model(s). on which different models do not necessarily agree. The fu-
(3) Evaluate climate model skill and bias. ture predicted water balance will also vary regionally, from
(4) Quantify General Circulation Model (GCM) between- wetter to drier to neutral. Thus, it is critical that older GCM
model variability. models/model runs no longer be used for any ecosystem
(5) Select an ecosystem model and assess uncertainty. studies that involve water balance. Further, in literature sum-
(6) Properly consider transient versus equilibrium re- maries, older studies should not be given equal weight to
sponses. those using more recent GCM results.
(7) Include effects of CO2 on plant response.
(8) Evaluate implications of simplifying assumptions. Evaluating climate model skill and bias
(9) Consider animal linkage with vegetation.
Output from a climate model is often used to evaluate ecosys-
Each of these criteria is discussed more fully in what fol- tem response, perhaps with a regional downscaling first.
lows. These outputs are virtually always taken at face value for
conducting impact studies. Any particular model, however,
Clearly defining an emissions scenario
may have known skill and bias issues with respect to regional
It is standard practice to evaluate ecosystem impacts in terms climates. Skill can be evaluated by matching the pattern in
of the Intergovernmental Panel on climate change standard time or distribution of temperatures or rainfall between the
scenarios for future emissions of greenhouse gases (Intergov- model and historical data. For example, a GCM might pro-
ernmental Panel on Climate Change [IPCC] 2007). The first duce skillful annual precipitation amounts, but the seasonal
factor to consider is that both the scenarios themselves and distribution could be very different from actual. The sea-
the model ensemble warming due to the scenarios differ be- sonal amounts could be critical to properly simulating plant
tween the IPCC 2001 (Intergovernmental Panel on Climate growth. Similarly, a GCM could consistently predict temper-
Change [IPCC] 2001) and 2007 reports. For example, while atures too warm for a region, even though matching historical

2 
c 2011 The Author. Published by Blackwell Publishing Ltd.
C. Loehle Assessing Climate Change Impacts

trends. This could lead to modeled impacts that are not re- predicted habitat between two climate models for an emis-
alistic. In IPCC assessments, in which global trends over 100 sions scenario were greater than the differences between sce-
years are being evaluated, these skill and bias issues may not narios using the same climate model. Given the large num-
affect the calculation of trends, but at the regional/local scale ber of climate models and their difficulties modeling regional
for simulating ecosystem response, they cannot be ignored. climate (noted above), it is apparent that vastly different eco-
Anagnostopoulos et al. (2010) compared six climate mod- logical impacts could be predicted depending on the model
els at multiple scales to weather data for the contiguous used and the geographic region of interest, thereby violat-
United States. Model results were all warmer than the ob- ing criteria of rigor and reproducibility. One way around the
served mean annual temperature and minimum monthly problem is to use multiple models to sample the range of
temperature by up to 4 C. This bias creates an obvious prob- possible projections. This is especially important because at
lem for any bioclimatic (niche) or simulation model used to the regional scale some models may project wetter and some
forecast future distributions because absolute temperatures, drier conditions. While this solution requires more work to
not just trends, are critical to biological processes. This is perform an analysis, a valid result depends on some estimate
particularly true when models are developed using actual of uncertainty.
(not model) climate data, which do not share the model
bias. Likewise, at the continental scale, they found modeled Selecting an ecosystem model and assessing
precipitation to be 36% above the true value, which has im- uncertainty
plications for any biological model. Schliep et al. (2010) and
Just as with climate models, ecosystem models can differ in
Stephens et al. (2010) noted multiple failings in the ability of
their skill and bias as well as in their suitability for a particular
GCMs to model precipitation, including both bias and lack
task, and may produce widely varying results between models
of skill.
(Cramer et al. 2001). The reasons for choosing a particular
Consider a climate model that produces output at the dry
model for an assessment need to be clearly stated. Process
end of what occurs in a particular region. A model of forest
models can have considerable parametric uncertainty (e.g.,
growth is run under constant climate, but because of the cli-
Larocque et al. 2011) that can carry through the analysis to
mate models temperature bias, the simulated forest is near
reduce outcome certainty. Ignoring these issues can make it
the point at which drought stress would cause dieback. Under
look like model output is without error, whereas it is stan-
almost any additional warming, this simulated forest using
dard practice to report confidence intervals on experimental
this climate model would show dieback. This bias can be
results.
addressed by comparing the GCM output to local weather
Niche (suitability) models likewise are not guaranteed to
data and adjusting the model bias (as Ines and Hansen 2006)
be accurate (Segurado and Araujo 2004; Araujo and Guisan
before simulating forest growth (for both control and impact
2006). Classification accuracy of niche models does not tell
scenarios). If it is determined that skill is lacking (e.g., rainfall
the full story. Beale et al. (2008) and Pearson and Dawson
frequency distribution is wildly wrong or seasonal temper-
(2003) provide a more complete discussion of the inherent
atures are not proportional) then another model should be
uncertainty in these models including issues of variable co-
used. In any case, climate model adequacy for the intended
variance and spurious correlation. There is unfortunately no
purpose should be evaluated rather than treating model out-
modeling method that has unambiguously been shown to be
put as if it were data.
universally superior. It is critical to evaluate uncertainty due
to these issues.
Quantifying GCM between-model variability
Properly considering transient versus
GCMs exhibit considerable variation in outputs (Furrer et al.
equilibrium responses
2007), especially at regional scales (e.g., Rowell 2006). For
example, Woollings (2010) found that simulations for Europe A large part of the ecosystem response literature consists of
differed between models (and from actual weather) for the the analysis of habitat suitability (or geographic range) maps.
jet stream location, zonal air flow, blocking highs, and other Data on climate and other variables are used to predict pres-
key weather patterns. It is important to note that models do ence of a species, forest type, or ecosystem using regression
not differ only in trends or stochastically, but also in terms of analysis or other tools. This model is then applied to future-
seasonality and spatial patterns of weather. There may not be modeled climates. A large shift in the range is usually con-
any sense in which these differences average out between sidered to be an indicator of extinction risk because species,
models. especially plants, are not able to migrate rapidly (e.g., Aitken
Few studies have compared ecosystem impacts using more et al. 2008; Morin et al. 2008).
than one climate model. In a study of animal habitat suitabil- Implicit in these conclusions is the equating of suitable
ity in Spain, Real et al. (2010) found that the differences in habitat with the habitat necessary for survival (e.g., Araujo


c 2011 The Author. Published by Blackwell Publishing Ltd. 3
Assessing Climate Change Impacts C. Loehle

and Guisan 2006). However, the presence of a species means shown a positive response of plant growth to CO2 enrich-
that it is competitive in the area where it occurs, not just that it ment (Curtis and Wang 1998; de Graaf et al. 2006). For plant
can survive there. Small differences in competitive ability will growth or habitat suitability models, failure to include CO2
eventually work themselves out and show up in distribution effects and WUE increases with time over the next 100 years
data, but this process may take a very long time, especially will lead to significant overestimates of negative impacts of
for long-lived organisms such as trees. Conditions outside elevated temperature, reduced moisture, or both together
the realized niche (i.e., where the species currently occurs) (Cramer et al. 2001; Pan et al. 2009; Friend 2010). Mechanis-
cannot be assumed to be lethal without proof. Projected cli- tic models that incorporate CO2 effects are generally based
mate changes are small compared to even daily temperature on experimental results such as FACE experiments, in some
variation and, for most species, are not near lethal limits cases with medium-sized trees over many years. Keenan et al.
(e.g., Wertin et al. 2010). This general point was discussed (2011) simulated forest growth with and without CO2 effects.
in Loehle and LeBlanc (1996) but needs repeating because Without including CO2 effects in the simulation, all three
the confusion between transient and equilibrium response species showed decreased net primary productivity (NPP)
continues. over time under the warming scenario. However, when CO2
Multiple studies suggest that actual vegetation responses effects were included, the species exhibited increased growth
to even large climate shifts are slow. Cole (1985, 2009) doc- until about 2070 followed by a slight decline, ending at 2100
umented a several thousand year transition in the Grand with slightly higher NPP than in 2000. How CO2 effects are
Canyon following past warming episodes. Cwynar and Spear addressed can make the difference between positive and neg-
(1991) showed that boreal forest advance in response to past ative growth responses under many scenarios. It is also the
warming took up to 1000 years, with a faster retreat due to case, of course, that no CO2 experiments have gone as far as
cooling (which damages trees and prevents regeneration), as 100 years, covered large spatial extents, or allowed for species
also shown by Tinner et al. (2007) for boreal dieback during shifts, so models incorporating CO2 effects may give unre-
the Little Ice Age. Williams et al. (2011) documented a slow alistic or exaggerated responses at some time/space scales
late Quaternary boreal forest expansion across the North- (Leuzinger et al. 2011).
ern Hemisphere. In the boreal forest, where 20th century
warming would suggest rapid geographic displacement, the Evaluating implications of simplifying
response has been very slow (Masek 2001; Payette 2007). In assumptions
no case has any study documented a dead zone in response
In order to conduct a geographic analysis of climate effects,
to a past climate shift, as is implied by static habitat suitability
it is often necessary to make simplifying assumptions due
models. Simulation models also suggest that such transitions
to data resolution. We may, for example, have GCM output
should be gradual (Noble 1993; Loehle 2000, 2003).
at 1 1 resolution. In the real world, however, fine-scale
Application of these criteria involves the careful delineation
topographic effects moderate regional climate in complex
of implications from a study. It is not necessary that dynamic
terrain. An analysis based on coarse-scale data might suggest
simulations be used. Rather, it is critical that information on
that no suitable habitat remains under some scenario, when
transient response (relaxation times) can be used to qualify
in fact topography might provide for refugia (Scherrer and
conclusions. If the study concerns mobile species, equilib-
Korner 2010; Dobrowski 2011; Godfree et al. 2011). Other
rium might be quickly achieved, but if it concerns trees it
simplifications might be made, such as using a single very dry
could take hundreds to thousands of years. Nonoverlap of
soil type for the simulation (e.g., Coops and Waring 2011).
current and hypothetical geographic ranges does not mean
The criterion here is that such simplifications, while perhaps
extinction unless climate lethality can be proven by other
necessary because of data or computational limitations, have
means.
implications that qualify the results and should be discussed.
In addition, geographic data can have errors resulting from
Including effects of CO2 on plant response vegetation classification, measurement error and bias, ag-
gregation error, and other sources, leading to effects possibly
CO2 enrichment directly increases growth and water-use ef-
larger than those resulting from the scenarios being evaluated
ficiency (WUE) in plants (Medlyn et al. 2001; Leuzinger and
(Verburg et al. 2011).
Korner 2007; Loehle 2007; Kohler et al. 2010), with effects
differing by taxa and CO2 level. Increased WUE is particu-
Considering animal linkage with vegetation
larly important when water is limiting, and has been shown to
mitigate considerable drought stress (e.g., de Graff et al. 2006; When suitability/range map-type models are developed for
Wertin et al. 2010). WUE is not considered here in terms of plants, there are good reasons to suppose that climate directly
hydrology but only with respect to drought response. Free- determines plant distribution (as modified by competition
air CO2 exchange (FACE) and open top chambers have both and fire, or course). It is equally easy to develop models for

4 
c 2011 The Author. Published by Blackwell Publishing Ltd.
C. Loehle Assessing Climate Change Impacts

animals range, but these models might actually be proxies for met most of the criteria proposed for impact assessments.
the vegetation on which the animals depend as much as for Most used the most recent IPCC AR4 climate simulations.
climate per se. That is, spurious correlation is a real danger Keenan et al. (2011) did underestimate outcome variability
in the absence of experimental data. For example, ruminants by only using one climate simulation and one forest growth
depend on grasslands and waterfowl depend on marshes and model, but is the only study that directly compared niche
lakes. If an animal depends on certain vegetation that exhibits and growth model results (using multiple niche models but
a long lag in responding to climate, the animal may likewise a single vegetation model). The inclusion of CO2 effects led
persist in its current range unless its physiological tolerance to an opposite conclusion on impacts (positive) compared to
is actually exceeded. If the niche model uses climate vari- leaving them out (negative). Doherty et al. (2010) evaluated
ables, the projection of animal response could be unrealistic. effects of future climate on NPP of East Africa. They used out-
Thus, it should be recognized that suitability models are cor- put from nine GCM models to capture uncertainty but only
relational rather than fundamental, and investigators should a single vegetation model that did simulate CO2 enrichment
carefully consider model output in this light. effects. All cases projected increased NPP in the study area.
Poulter et al. (2010) simulated response of the Amazon forest
Literature Survey to CO2 and eight different GCMs as well as vegetation model
parameter uncertainty. They found that between-climate-
A literature survey was conducted to evaluate compliance model differences dominated outcome spread compared to
with the assessment criteria suggested here. All issues of CO2 or vegetation model parameter uncertainty, with future
Global Change Biology in 2009 and 2010 and the first four precipitation uncertainty the greatest unknown determinant
issues in 2011 were assessed. All terrestrial studies discussing of outcomes.
future impacts on species or ecosystems were considered (not Chen et al. (2010) evaluated response of Douglas-fir to
carbon or fire studies). Experimental and field studies were projected climate change using a tree ring index model of
outside the scope of the review. A total of 20 papers were growth response. Variability was captured by using output
evaluated (Table 1), with one study evaluated under two cat- from five climate models. The inability of this approach to
egories. In the authors experience, these 20 papers are similar include effects of rising CO2 is problematic because the main
to others in the field and serve to illustrate the points made. impacts of climate change in their analysis were manifested
via drought stress, which rising CO2 and thus increasing
Growth models WUE would help ameliorate. Xu et al. (2009) simulated forest
Nine papers (seven of vegetation) with some sort of dynamic growth in northern Minnesota using a single process model.
or growth model utilized a variety of approaches and often CO2 was included in the model. Future climate data (27

Table 1. Percent compliance with criteria for literature survey cases (percentages calculated only for criteria relevant to the studies in question). NA
reflects that criterion not relevant to that model type.

Criteriaa

n 1 2 3 4 5 6 7 8 9

Dynamic models
Plant 7 1.0 0.86 0 0.57 0.14 1.0 0.71 1.0 NA
Animal 2 0 0.5 0 0.5 1.0 1.0 NA 0 0
Niche models
Plant 7 0.71 0.43 0 0.43 0.43 0 0 0.57 NA
Animal 5 0.8 0.6 0 0.4 0.4 0 0 0 0.6

NA = not applicable
a
criteria areas follows:
(1) Clearly define emissions scenario.
(2) Select climate model(s).
(3) Evaluate climate model skill and bias.
(4) Quantify GCM between-model variability.
(5) Select an ecosystem model and assess uncertainty.
(6) Properly consider transient versus equilibrium responses.
(7) Include effects of CO2 on plant response.
(8) Evaluate implications of simplifying assumptions.
(9) Consider animal linkage with vegetation.


c 2011 The Author. Published by Blackwell Publishing Ltd. 5
Assessing Climate Change Impacts C. Loehle

profiles) were mostly from older GCMs used in the third IPCC between-climate model differences by Dlamini (2011). The
assessment in 2001. The large number of climate runs allowed equilibrium assumption (equating niche model changes with
the large uncertainty due to forecasting to be quantified. immediate population changes) was universal in these stud-
Morin et al. (2009) used a process-based statistical model ies, with only a mention of possible lags by Dirnbock and
for leaf unfolding for 22 North American tree species as a Rabitsch (2011), although possible lags were considered to
function of climate. A single GCM run from 2001 was used for be in terms of mere decades. Two studies of invasive plants
the A2 and B2 scenarios only. The effect of CO2 on phenology (Bradley 2009; Bradley et al. 2009) using niche models used 10
was not assessed. Spatial resolution was probably not an issue. AR4 GCM runs and were thus able to quantify model-based
With a single older GCM run, there is no way to evaluate the uncertainty, but did not assess model skill or bias and did
representativeness of their forecasts. Scheiter and Higgens not clearly cite the model runs used. Because they only eval-
(2009) used a mechanistic model of grassland and forest uated potential habitat for invasion, there was less problem
to evaluate response of the vegetation of Africa to climate with the equilibrium assumption, but they did assume that
change. The model included CO2 effects that were believed future unsuitable habitat for the invasive species would im-
to account for increased forest area and biomass as well as mediately become a conservation opportunity. Randin et al.
the increase in total vegetated land (at the expense of desert) (2009) compared scenarios at 50 km 50 km versus 25 m
by 2100. Spatial resolution was good at the scale of 1 ha. The 25 m plot scales. Because they interpolated climate to ac-
GCM used was a 2007 model run, but the lack of intermodel count for topography at the smaller plot size, the local-scale
comparison makes it difficult to evaluate results. models showed persistence of up to 100% of species that the
Two animal process models were found. Gilg et al. (2009) European-scale model predicted would lose their entire habi-
explored effects of future climate on Arctic predatorprey tat. The study was limited by the use of (apparently) an older
systems. The climate scenarios were qualitative and hence single (approximately 2000) GCM run. The net effect in all
difficult to evaluate. Snall et al. (2009) used seven up-to-date these models of excluding the beneficial effects of elevated
GCM models and a process model for plague in prairie dogs. CO2 and assuming rapid vegetation changes due to changes
By sampling from the plague model parameter uncertainty in available suitable habitat is to greatly amplify likely neg-
space as well as the multiple GCMs they were able to do an ative effects or even to convert positive effects into negative
exemplary job quantifying uncertainty, although they did not ones, as in Keenan et al. (2011). Likewise, niche models at too
assess GCM bias or skill. coarse a resolution will miss refugia from climate change.
Overall, process-based studies did a good job assessing un- Five niche models were used to assess animal distributions
certainty due to between climate model variation and within (Jarema et al. 2009; Carroll et al. 2010; Carvalho et al. 2010;
ecosystem model parametric uncertainty, although only Snall Rebelo et al. 2010; Habel et al. 2011). These studies used
et al. (2009) and Poulter et al. (2010) evaluated both. Most climate projections from 2004 or 2005 except for Carvalho
used current climate models but none evaluated GCM skill et al. (2010) for which the date of the climate model could
and bias issues. not be determined and Jarema et al. (2009), which used a
2001 GCM run. Carvalho et al. (2010) used more than one
niche model (nine) and GCM (three) and Jarema et al. (2009)
Niche models
used two GCM models and multiple niche models. Carroll
All seven studies using niche models for plant distributions et al. (2010) developed animal niche models that included
(Bradley 2009; Bradley et al. 2009; Randin et al. 2009; Fee- vegetation and held vegetation constant based on considera-
ley and Silman 2010; Dirnbock and Rabitsch 2011; Dlamini tion of lags that were likely to occur in forests of the Pacific
2011; Keenan et al. 2011) failed to consider CO2 effects on Northwest United States. Jarema et al. (2009) found niche
future growth. Because much of the impact of future climate models based on habitat to be almost as good as those using
on plants results from net water deficiency, the lack of consid- climate variables alone. All other studies used niche models
eration of CO2 effects calls results into question, especially in dominated by climate variables and thus ignored the possibil-
light of Keenan et al.s (2011) results discussed above. Use of ity that animals may be found in their current habitat based
future climate scenarios was mixed. Keenan et al. (2011) used on the vegetation, which would probably respond to climate
IPCC AR4 climate model results, but Dlamini (2011) used a change in a lagged fashion. They all further assumed equi-
result from 2000, which is rather old. Dirnbock and Rabitsch librium response, whereas animal distributions are likely to
(2011) and Feeley and Silman (2010) used simple increases be heavily influenced by competition and predation effects
in temperature, including an extreme 8 C rise in the latter that would take time to come to equilibrium in a new cli-
case, with no increase in precipitation. Failure to consider fu- mate. Finally, a fairly extreme scenario of 6 C warming was
ture precipitation changes generally will make impacts more used by Rebelo et al. (2010). Simplifying assumptions that
negative for plants. Uncertainty due to niche model differ- may have affected results include use of 35 km2 grid cells in
ences was considered by Keenan et al. (2011), and that due to Dirnbock and Rabitsch (2011), only a simple temperature

6 
c 2011 The Author. Published by Blackwell Publishing Ltd.
C. Loehle Assessing Climate Change Impacts

gradient (no heterogeneity or precipitation) in Feeley and was immediately uninhabitable by the species or vegetation
Silman (2010), the coarse spatial scale used in Carvalho et al. without presenting evidence that the altered climate would
(2010) that would not allow for refugia, and the assumption be actually lethal. Instead, competitive displacement, an in-
in Rebelo et al. (2010) that bats are dispersal-limited. Over- herently slow process that may take centuries (Noble 1993;
all, uncertainty was underestimated and effects were probably Loehle 2000, 2003), could eventually result in the changes
exaggerated in this set of studies. suggested by niche models. While niche models are inherently
equilibrium based, it is possible to perform some assessment
of nonequilibrium response (e.g., Araujo and Pearson 2005).
Discussion In addition, data do exist on drought and heat tolerance of
The criteria suggested in this study are intended to guide de- many species, and such data should be consulted before mak-
cisions in assessments of future climate change impacts and ing claims that species or ecosystems will perish/vanish over
to help prevent results from being arbitrary or showing false the next few decades due to a few degrees warming. If the
precision due to lack of uncertainty information. The litera- temperature rise is within the thermal tolerance of a species,
ture survey revealed that recent studies have not considered then range shifts, not extinctions, are going to result and
some of the proposed criteria (Table 1). Violation of any of may take centuries. Consulting the literature to evaluate the
the criteria can be serious, but some are more easily reme- likely time dynamics of the output of niche models is not a
died than others. The criteria are next considered in turn, burdensome requirement.
with suggestions for improved assessments. Most studies using growth models included beneficial ef-
The examined papers generally referred back to IPCC AR4 fects of elevated CO2 on plant growth. It is noteworthy that
emissions scenarios (IPCC 2007). Usually, either the A2 sce- these models forecast much less impact or even positive ef-
nario was used or a range of scenarios was evaluated. This fects compared to niche models, which are inherently unable
standard set of scenarios helps with interstudy comparisons. to account for future elevated CO2 effects. While the ex-
Some studies used arbitrary temperature increases rather act effects of elevated CO2 over the long term are not known
than model outputs, which has the added problem that pre- (e.g., Leuzinger et al. 2011), the experimental literature seems
cipitation was held constant or ignored. If simple temperature to support a growth-enhancement conclusion and models
increases are evaluated, they should at least be more clearly have widely adopted this result. It would seem that dynamic
related to model scenarios. It is difficult to justify leaving (mechanistic) models of plant response to climate change
precipitation changes out of an assessment and not diffi- should be preferred, though niche models are much more
cult to add this factor. Recent generation climate simulations widely published, perhaps due to simplicity and cost. Cost
were not necessarily used, although the growth model stud- alone does not justify a method that leads to the wrong an-
ies mostly used up-to-date results. Four of seven plant niche swer. At the very least, the result of any niche analysis needs
models used out-of-date climate models. There is really no to be qualified by discussion of how CO2 effects would alter
good reason to use old climate simulations. the result.
No study took into account the skill or bias of the climate The primary simplifying assumption that impacted results
models. This is problematic, especially because many used in a number of studies was the reduction of spatial hetero-
models for current ecosystems calibrated with actual weather geneity considered, either due to grid scale or to elevational
data and compared them to future simulated climates in gradient representation (e.g., representing it as a smooth
which a bias (offset) could exist. At least a qualitative assess- curve). Spatial heterogeneity can provide refugia, and fail-
ment of climate model outputs compared to the study area ure to consider this effect will tend to exaggerate impacts, as
seems necessary before doing an assessment. Of the sources clearly shown by Randin et al. (2009). The limiting factors
of uncertainty (climate model, and niche model or growth for performing more spatially resolved analyses are com-
model), only three studies (Jarema et al. 2009; Carvalho et al. puter time and input data. The former can be overcome by
2010; Poulter et al. 2010) considered both and many stud- distributed computing or weekend runs. The latter is at least
ies included only a single model and GCM dataset. It is not mitigated by using the highest resolution data available rather
impossible to avoid this problem by explicitly using the mul- than aggregated data. The conceptual and analysis steps are
tiple climate model outputs, which are becoming increasingly identical in either case.
available for regional scales. Likewise, ensemble means of cli- Finally, in animal impact studies, a pervasive trend was
mate data might be available and ensemble means or overlap to model animal response solely as a function of climate. If
for niche models (Araujo and New 2007) are not difficult to animals select habitat based on vegetation and the response
develop, since the statistical tools are widely available. of vegetation to changing climate is lagged, the response of
Studies using niche models mostly failed to properly con- animals will probably also be lagged. The assumption that
sider lags in vegetation response to climate change. They as- animals are limited by climate as defined by a niche model is
sumed that habitat defined as unsuitable by the niche model virtually unverified.


c 2011 The Author. Published by Blackwell Publishing Ltd. 7
Assessing Climate Change Impacts C. Loehle

When study results can be completely altered by including climate variability and projected climate change. Glob. Change
or excluding a factor such as CO2 or spatial heterogeneity and Biol. 16:33743385.
when between-model forecasts can be extremely variable, it Cole, K. 1985. Past rates of change, species richness and a model
is essential that these and other known factors be recognized of vegetational inertia in the Grand Canyon, Arizona. Am. Nat.
and addressed in any analysis of climate change effects. The 125:289303.
majority of surveyed studies might have yielded very different Cole, K. L. 2009. Vegetation response to early Holocene warming
results if the criteria suggested here had been applied. More as an analog for current and future changes. Conserv. Biol.
attention to criteria such as those proposed would lead to 24:2937.
much more reliable and useful impact assessments. It has Coops, N. C., and R. H. Waring. 2011. A process-based approach
been shown above that most of these problems are easily fixed, to estimate lodgepole pine (Pinus contorta Dougl.) distribution
require a modest effort (e.g., using niche model ensembles), in the Pacific Northwest under climate change. Climatic
or can be dealt with in the discussion of results. Change 105:313328.
Corte-Real, J., X. Zhang, and X. Wang. 1995. Downscaling
Acknowledgments GCM information to regional scales: a non-parametric
multivariate regression approach. Climate Dynam.
No outside funding was received for this study. I thank A. 11:413424.
Lucier, D. Sleep, J. Verschuyl, and B. Wigley for helpful com- Cramer, W., A. Bondeau, F. I. Woodward, I. C. Prentice, R. A.
ments on the manuscript. Betts, V. Brovkin, P. M. Cox, V. Fisher, J. A. Foley, A. D. Friend,
C. Kucharik, M. R. Lomas, N. Rmankutty, S. Sitch, B. Smith,
References A. White, and C. Young-Molling. 2001. Global response of
Aitken, S. N., S. Yeaman, J. A. Holliday, T. Wang, and S. terrestrial ecosystem structure and function to CO2 and
Curtis-McLane. 2008. Adaptation, migration or extirpation: climate change: results from six dynamic global vegetation
climate change outcomes for tree populations. Evol. Appl. models. Glob. Change Biol. 7:357373.
1:95111. Curtis, P. S., and X. Wang. 1998. A meta-analysis of elevated CO2
M. B., and A. Guisan. 2006. Five (or so) challenges for
Araujo, effects on woody plant mass, form, and physiology. Oecologia
species distribution modelling. J. Biogeogr. 33:16771688. 113:299313.
M. B., and M. New. 2007. Ensemble forecasting of species
Araujo, Cwynar, L. C., and R. W. Spear. 1991. Reversion of forest to
distributions. Trends Ecol. Evol. 22:4247. tundra in the central Yukon. Ecology 72:202212.
M. B., and R. G. Pearson. 2005. Equilibrium of species
Araujo, De Graaff, M-A., K-J. van Groenigen, J. Six, B. Hungate, and C.
distributions with climate. Ecography 28:693695. van Kessel. 2006. Interactions between plant growth and soil
Anagnostopoulos, G. G., D. Koutsoyiannis, A. Christofides, A. nutrient cycling under elevated CO2: a meta-analysis. Global
Efstradiadis, and N. Mamassis. 2010. A comparison of local Change Biology 12:2077-2091.
and aggregated climate model outputs with observed data. Dirnbock, T., F. Essl, and W. Rabitsch. 2011. Disproportional risk
Hydrolog. Sci. J. 55:10941110. for habitat loss of high-altitude endemic species under climate
Beale, C. M., J. J. Lennon, and A. Gimona. 2008. Opening the change. Glob. Change Biol. 17:990996.
climate envelope reveals no macroscale associations with Dlamini, W. 2011. Probabilistic spatio-temporal assessment of
climate in European birds. Proc. Nat. Acad. Sci. vegetation vulnerability to climate change in Swaziland. Glob.
105:1490814912. Change Biol. 17:14251441.
Bradley, B. A. 2009. Regional analysis of the impacts of climate Dobrowski, S. Z. 2011. A climatic basis for microrefugia: the
change on cheatgrass invasion shows potential risk and influence of terrain on climate. Glob. Change Biol.
opportunity. Glob. Change Biol. 15:196208. 17:10221035.
Bradley, B. A., M. Oppenheimer, and D. S. Wilcove. 2009. Doherty, R. M., S. Sitch, B. Smith, S. L. Lewis, and P. K.
Climate change and plant invasions: restoration opportunities Thornton. 2010. Implications of future climate and
ahead? Glob. Change Biol. 15:15111521. atmospheric CO2 content for regional biogeochemistry,
Carroll, C., J. R. Dunk, and A. Moilanen. 2010. Optimizing biogeography and ecosystem services across East Africa. Glob.
resiliency of reserve networks to climate change: multispecies Change Biol. 16:617640.
conservation planning in the Pacific Northwest, USA. Glob. Feeley, K. J., and M. R. Silman. 2010. Land-use and climate
Change Biol. 16:891904. change effects on population size and extinction risk of
Carvalho, S. B., J. C. Brito, E. J. Crespo, and H. P. Possingham. Andean plants. Glob. Change Biol. 16:32153222.
2010. From climate change predictions to actions conserving Friend, A. D. 2010. Terrestrial plant production and climate
vulnerable animal groups in hotspots at a regional scale. Glob. change. J. Exp. Bot. 61:12931309.
Change Biol. 16:32573270. Furrer, R., S. R. Sain, D. Nychka, and G. A. Meehl. 2007.
Chen, P.-Y., C. Welsh, and A. Hamann. 2010. Geographic Multivariate Bayesian analysis of atmosphere-ocean general
variation in growth response of Douglas-fir to interannual circulation models. Environ. Ecol. Stat. 14:249266.

8 
c 2011 The Author. Published by Blackwell Publishing Ltd.
C. Loehle Assessing Climate Change Impacts

Gilg, O., B. Sittler, and I. Hanski. 2009. Climate change and cyclic Loehle, C. 2000. Forest ecotone response to climate change:
predator-prey population dynamics in the high Arctic. Glob. sensitivity to temperature response functional forms. Can. J.
Change Biol. 15:26342652. For. Res. 30:16321645.
Godfree, R., B. Lepschi, A. Reside, T. Bolger, B. Robertson, D. Loehle, C. 2003. Competitive displacement of trees in response to
Marshall, and M. Carnegie. 2011. Multiscale topoedaphic climate change or introduction of exotics. Environ. Manage.
heterogeneity increases resilience and resistance of a dominant 32:106115.
grassland species to extreme drought and climate change. Loehle, C. 2007. Predicting Pleistocene climate from vegetation
Glob. Change Biol. 17:943958. in North America. Climate of the Past 3:109118.
Habel, J. C., D. Rodder, T. Schmitt, and G. N`eve. 2011. Global Loehle, C., and D. C. LeBlanc. 1996. Model-based assessments of
warming will affect the genetic diversity and uniqueness of climate change effects on forests: a critical review. Ecol.
Lycaena helle populations. Glob. Change Biol. 17:194205. Modell. 90:131.
Ines, A. V. M., and J. W. Hansen. 2006. Bias correction of daily Masek, J. G. 2001. Stability of boreal forest stands during recent
GCM rainfall for crop simulation studies. Agric. For. Meteorol. climate change: Evidence from Landsat satellite imagery. J.
138:4453. Biogeogr. 28:967976.
Intergovernmental Panel on Climate Change (IPCC) 2001. In Medlyn, B. E., C. V. M. Barton, M. S. J. Broadmeadow, R.
J.T. Houghton, Y. Ding, D.J. Griggs, M. Noguer, P.J. van der Ceulemans, P. De Angelis, M. Forstreuter, M. Freeman, S. B.
Linden, X. Dai, K. Maskell, and C.A. Johnson, eds. Climate Jackson, S. Kellomaki, E. Laitat, A. Rey, P. Roberntz, B. D.
Change 2001: The Physical Science Basis. Contribution of Sigurdsson, J. Strassemeyer, K. Wang, P. S. Curtis, and P. G.
working group I to the Third Assessment Report of the Jarvis. 2001. Stomatal conductance of forest species after
Intergovernmental Panel on Climate Change. Cambridge long-term exposure to elevated CO2 concentration: a
Univ. Press, Cambridge and New York. synthesis. New Phytol. 149:247264.
Intergovernmental Panel on Climate Change (IPCC) 2007. In Morin, X., M. J. Lechowicz, C. Augspurger, J. OKeefe, D. Viner,
S. Soloman, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. and I. Chuine. 2009. Leaf phenology in 22 North American
Avery, M. Tignor, and H. L. Miller, eds. Climate change 2007: tree species during the 21st century. Glob. Change Biol.
The Physical Science Basis. Contribution of working group I to 15:961975.
the Fourth Assessment Report of the Intergovernmental Panel Morin, X., D. Viner, and I. Chuine. 2008. Tree species range shifts
on Climate Change. Cambridge Univ. Press, Cambridge and at a continental scale: new predictive insights from a
New York. process-based model. J. Ecol. 96:784794.
Jarema, S. I., J. Samson, B. J. McGill, and M. M. Humphries. Noble, I. R. 1993. A model of the responses of ecotones to climate
2009. Variation in abundance across a species range predicts change. Ecol. Appl. 3:396403.
climate change responses in the range interior will exceed Pan, Y., R. Birdsey, J. Hom, and K. McCullough. 2009. Separating
those at the edge: a case study with North American beaver. effects of changes in atmospheric composition, climate and
Glob. Change Biol. 15:508522. land-use on carbon sequestration of U.S. Mid-Atlantic
Keenan, T., J. M. Serra, F. Lloret, M. Ninyerola, and S. Sabate. temperate forests. For. Ecol. Manage. 259:151164.
2011. Predicting the future of forests in the Mediterranean Payette, S. 2007. Contrasted dynamics of northern Labrador tree
under climate change, with niche- and process-based models: lines caused by climate change and migration lag. Ecology
CO2 matters! Glob. Change Biol. 17:565579. 88:770780.
Kohler, I. H., P. R. Poulton, K. Auerswald, and H. Schnyder. 2010. Pearson, R. G., and T. P. Dawson. 2003. Predicting the impacts of
Intrinsic water-use efficiency of temperate seminatural climate change on the distribution of species: are bioclimate
grassland has increased since 1857: an analysis of carbon envelope models useful? Glob. Ecol. Biogeogr. 12:361371.
isotope discrimination of herbage from the Park Grass Pilkey-Jarvis, L., and O. H. Pilkey. 2008. Useless arithmetic: ten
Experiment. Glob. Change Biol. 16:15311541. points to ponder when using mathematical models in
Larocque, G. R., J. S. Bhatti, J. C. Ascough II, J. Liu, N. Luckai, D. environmental decision making. Public Admin. Rev.
Mailly, L. Archambault, and A. M. Gordon. 2011. An analytical 68:470479.
framework to assist decision makers in the use of forest Poulter, B., F. Hattermann, E. Hawkins, S. Zaehle, S. Sitch, N.
ecosystem model predictions. Environ. Model. Software Restrepo-Coupe, U. Heyder, and W. Cramer. 2010. Robust
26:280288. dynamics of Amazon dieback to climate change with
Leuzinger, S., and C. Korner. 2007. Water savings in mature perturbed ecosystem model parameters. Glob. Change Biol.
deciduous forest trees under elevated CO2 . Glob. Change Biol. 16:24762495.
13:24982508. Randin, C. F., R. Engler, S. Normand, M. Zappa, K. E.
Leuzinger, S., Y. Luo, C. Beier, W. Dieleman, S. Vicca, and C. Zimmermann, P. B. Pearman, P. Vittoz, W. Thuiller, and A.
Korner. 2011. Do global change experiments overestimate Guisan. 2009. Climate change and plant distribution: local
impacts on terrestrial ecosystems? Trends Ecol. Evol. models predict high-elevation persistence. Glob. Change Biol.
26:236241. 15:15571569.


c 2011 The Author. Published by Blackwell Publishing Ltd. 9
Assessing Climate Change Impacts C. Loehle

Real, R., A. L. Marquez, J. Olivero, and A. Estrada. 2010. Species A. Bodas-Salcedo, K. Suzuki, P. Gabriel, and J. Haynes. 2010.
distribution models in climate change scenarios are still not Dreary state of precipitation in global models. J. Geophys. Res.
useful for informing policy planning: n uncertainty assessment 115:D24211.
using fuzzy logic. Ecography 33:304314. Tinner, W., C. Bigler, S. Gedye, I. Gregory-Eaves, R. T. Jones, P.
Rebelo, H., P. Tarroso, and G. Jones. 2010. Predicted impact of Kaltenrieder, U. Krahenbuhl, and F. S. Hu. 2007. A 700-year
climate change on European bats in relation to their paleoecological record of boreal ecosystem responses to
biogeographic patterns. Glob. Change Biol. 16:561576. climatic variation from Alaska. Ecology 89:729743.
Rowell, D. P. 2006. A demonstration of the uncertainty in Verburg, P. H., K. Neumann, and L. Nol. 2011. Challenges in
projections of UK climate change resulting from regional using land use and land cover data for global change studies.
model formulation. Climatic Change 79:243257. Glob. Change Biol. 17:974989.
2004. An evaluation of methods
Segurado, P., and M. B. Araujo. Wentz, F. J., L. Ricciardulli, K. Hilburn, and C. Mears. 2007. How
for modeling species distributions. J. Biogeogr. 31:15551568. much more rain will global warming bring? Science
Scheiter, S., and S. I. Higgins. 2009. Impacts of climate change on 317:233235.
the vegetation of Africa: an adaptive dynamic vegetation Wertin, T. M., M. A. McGuire, and R. O. Teskey. 2010. The
modelling approach. Glob. Change Biol. 15:22242246. influence of elevated temperature, elevated atmospheric CO2
Scherrer, D., and C. Korner. 2010. Infra-red thermometry of concentration and water stress on net photosynthesis of
alpine landscapes challenges climatic warming projections. loblolly pine (Pinus taeda L.) at northern, central and southern
Glob. Change Biol. 16:26022613. sites in its native range. Glob. Change Biol. 16:20892103.
Schliep, E. M., D. Cooley, S. R. Sain, and J. A. Hoeting. 2010. A Williams, J. W., P. Tarasov, S. Brewer, and M. Notaro. 2011. Late
comparison study of extreme precipitation from six different Quaternary variations in tree cover at the northern
regional climate models via spatial hierarchical modeling. forest-tundra ecotone. J. Geophys. Res. 116:G01017.
Extremes 13:219239. Woollings, T. 2010. Dynamical influences on European climate:
Snall, T., R. E. Benestad, and N. C. Stenseth. 2009. Expected an uncertain future. Philos. Trans. R. Soc. A 368:37333756.
future plague levels in a wildlife host under different scenarios Xu, C., G. Z. Gertner, and R. M. Scheller. 2009. Uncertainties in
of climate change. Glob. Change Biol. 15:500507. the response of a forest landscape to global climatic change.
Stephens, G. L., T. L. Ecuyer, R. Forbes, A. Gettlemen, J-C. Golaz, Glob. Change Biol. 15:116131.

10 
c 2011 The Author. Published by Blackwell Publishing Ltd.
TREND ANALYSIS OF SATELLITE GLOBAL TEMPERATURE
DATA

by

Craig Loehle (USA)

Reprinted from

ENERGY &
ENVIRONMENT
VOLUME 20 No. 7 2009

MULTI-SCIENCE PUBLISHING CO. LTD.


5 Wates Way, Brentwood, Essex CM15 9TB, United Kingdom
1087

TREND ANALYSIS OF SATELLITE


GLOBAL TEMPERATURE DATA
Craig Loehle
National Council for Air and Stream Improvement, Inc.
552 S Washington Street, Suite 224
Naperville, Illinois 60540
phone: 630-579-1190
fax: 630-579-1195
CLoehle@ncasi.org

ABSTRACT
Global satellite data is analyzed for temperature trends for the period January 1979
through June 2009. Beginning and ending segments show a cooling trend, while
the middle segment evinces a warming trend. The past 12 to 13 years show
cooling using both satellite data sets, with lower confidence limits that do not
exclude a negative trend until 16 years. It is shown that several published studies
have predicted cooling in this time frame. One of these models is extrapolated
from its 2000 calibration end date and shows a good match to the satellite data,
with a projection of continued cooling for several more decades.

INTRODUCTION
Temperature trends provide critical evidence for evaluating claims regarding
anthropogenic climate change. On the one hand, models project continued warming
(e.g., Hansen et al., 2006). On the other, it has been argued that Earths weather
should be expected to exhibit long-term persistence (LTP) at some scale (Cohn and
Lins, 2005; Easterling and Wehner, 2009; Wood, 2008). LTP could at any given time
give a false impression of the strength of anthropogenic effects by adding a warming
trend to an existing anthropogenic signal, or it could act to counter such signal for some
period of years or decades. Thus it is critical to examine climate history data at
multiple scales to evaluate these effects. Typical trend studies, however, have not
evaluated the most recent decades per se. For example, periods evaluated include sea
surface temperatures over 1960-1990 (Casey and Cornillon, 2001), lower troposphere
RSS data over 1979-2001 (Fu et al., 2004), the global surface record for 1950-2004,
1979-2004, 1950-1980 (Vose et al., 2005), 1988-2005 (Hansen et al., 2006), and
1977-2001 (Jones and Moberg, 2003), the NCEP reanalysis Northern Hemisphere
surface record for 1960-2000 (Lucarini and Russell, 2002), 1969-2000 for Southern
Hemisphere data (Thompson and Solomon, 2002), and the satellite and balloon data
for 1979-2004 (Christy et al., 2007). The most recent Intergovernmental Panel on
Climate Change report (IPCC, 2007) shows 100 and 140 year trends. In all cases, the
1088 Energy & Environment Vol. 20, No. 7, 2009

period of rapid warming in the late 1970s through 1990s was included in these
analyses. All of these studies looked only at long-term trends.
Given that satellite data are now available for more than 30 years and that recent
years do not show a visible upward trend, it seems appropriate to re-examine
temperature trends. The purpose here is not to obtain the 30 year trend, which has
been done previously, but rather to parse the data to evaluate evidence for LTP and
recent trends. In spite of the importance of the satellite data for assessing trends, no
analysis of trends for the most recent decades has been published using these data.

METHODS
The purpose of the analysis is to examine the data for linear trends. Looking backward
from the most recent dates, the question is what is the temperature trend over recent
years? If it is negative (cooling), how long can a cooling trend be said to have existed?
Starting at the beginning of the record in 1979 and looking forward, how long a record
is needed before a warming trend is detectable? Starting with intervals in the middle
of the record, how does the trend in this middle interval compare to the starting and
ending periods? Linear trends are the simplest way to assess climate change, and are
used in the IPCC reports and most of the trend studies cited in the introduction, among
many others. Linear trends also have the advantage that confidence intervals are well
defined, which aids in interpretation. Calculating such linear trends overcomes issues
due to subjective interpretation of noisy data and the arbitrariness of various methods
of smoothing the data, especially at the end points (see Soon et al., 2004). By showing
trends of data segments of multiple lengths, issues of the studied interval being unduly
influential are avoided. Longer-term trends (or red noise) are graphically uncovered
by the analysis, if they exist, and are not being estimated per se. Instead the questions
are strictly empirical: how much has it warmed/cooled over various lengths of time,
according to satellite data?
University of Alabama at Huntsville (UAH) and Remote Sensing Systems (RSS)
Microwave Sounding Unit (MSU) data were obtained on July 20, 2009. UAH data were
downloaded from http://vortex.nsstc.uah.edu/data/msu/t2lt/uahncdc.lt (Fig. 1a). RSS
(http://www.ssmi.com/msu/msu_data_description.html) data were downloaded from
ftp://ftp.ssmi.com/msu/monthly_time_series/rss_monthly_msu_amsu_channel_tlt_ano
malies_land_and_ocean_v03_2.txt (Fig. 1b). Data from January 1979 through June
2009 (366 months) were available from both series.
The data were not smoothed. Series of different length were considered, with 60
months being the shortest. Data were considered for intervals that began in January
1979, for intervals that ended in March 2009, and for intervals starting at the record
midpoint. Linear regression was performed using the Regress function in Mathematica
(www.Wolfram.com). Slopes were computed on a per decade basis. Adjusted slope
confidence intervals were calculated to account for temporal autocorrelation. The
single month-to-month correlation of regression residuals, r, was calculated as
n 1

( )
r1 = ( et et +1 ) / ( n k 1) / s 2 (1)
t =1

where e are the regression residuals at t, n is the sample size, k is the number of
Trend Analysis of RSS and UAH MSU Global Temperature Data 1089

parameters (1 here), and s2 is the residuals variance. The effective sample size ne is
(Bartlett 1935; Quenouille 1952; Santer et al. 2008)

ne = n
(1 r1 ) (2)
(1 + r1 )

where ne <n after adjustment. This value for ne is used to recompute the standard error
of the slope to get new confidence intervals. Longer-period autocorrelation is
accounted for by computing trends at multiple time scales.

a)
C

Months from January 1 9 79


b)
C

Months from January 1 9 79


Figure 1. Satellite data. a) UAH MSU global data from January 1979 through June 2009
(367 months). b) RSS global data from January 1979 through June 2009 (366 months).
1090 Energy & Environment Vol. 20, No. 7, 2009

RESULTS
As expected, as series length increases the confidence intervals narrow. For series ending
in 2009, the shortest (60 months) show a strongly negative trend of -0.33C per decade
for UAH and -0.38C per decade for RSS (Fig. 2). The trend does not stay positive until
the period length is longer than 146 months (12.2 yr) for UAH or 149 months (12.4 yr)
for RSS. There is a slightly more negative slope (faster cooling) at 130 months as the
1998 el Nio is included in the record. The lower confidence limit includes 0 until 196
months (16.3 yr) for UAH and 195 months (16.3 yr) for RSS. Thus for the past 16
years it is not possible to detect a warming trend with this data (lower confidence
intervals include zero) and for the past 12+ years the data actually show a cooling trend.
As the data length increases beyond month 200, the slope curve becomes stable
(constant) and confidence intervals become narrow because a longer sequence is being
analyzed and the addition of more months has little effect on the trend.

a)
Trend (C/decade)

Length of series from March 2009

b)
Trend (C/decade)

Length of series from March 2009

Figure 2. Trend analysis for records of different lengths (months) ending in March 2009
with autocorrelation corrected 95% confidence intervals. The first point on the left side
is the slope for data from June 2004 to June 2009, with subsequent points being series
beginning at earlier dates. a) UAH data. b) RSS data.
Trend Analysis of RSS and UAH MSU Global Temperature Data 1091

The same analysis was conducted for time series of different length that all originate
in January 1979 (Fig. 3). Shorter series up to 193 months (16.1 yr) long for UAH
and 129 months (10.8 yr) for RSS surprisingly show a negative slope. In fact, not
until the record is 273 months (22.8 yr) long for UAH or 229 months (19.1 yr) long
for RSS does the lower 95% confidence interval not include 0.
Next, series were constructed in the middle of the 366 month record. Beginning
with a 60 month segment centered on the middle of the record, slopes were again
calculated, but this time at each increment a month was added at the start and end of
the record until the full series was captured (Fig. 4). With this approach, the records
of various lengths all show positive slope (warming). However, for the short records
(up to 200 months) the slope fluctuates considerably, and the confidence intervals are
quite wide and include negative values. It is noticeable that at about 120 months the
1998 el Nio comes into play and creates a big apparent warming spike, which is
quickly damped out for longer records.

a)
Trend (C/decade)

Length of series from January 1979

b)
Trend (C/decade)

Length of series from January 1979

Figure 3. Trend analysis for records of different lengths (months) beginning in February
1979 with autocorrelation corrected 95% confidence intervals. The left-most point is the
five year data from February 1979 to February 1984. a) UAH data. b) RSS data.
1092 Energy & Environment Vol. 20, No. 7, 2009

Trend (C/decade)

Length of Series from Midpoint

b)
Trend (C/decade)

Length of Series from Midpoint

Figure 4. Trend analysis for records of different lengths (months) beginning at record
midpoint with autocorrelation corrected confidence intervals. The left-most point is
the five year interval centered on the series midpoint. a) UAH data. b) RSS data.

DISCUSSION
The data clearly show a cooling phase over the past 12 to 13 years, with lower
confidence intervals including negative trend over the past 16 to nearly 23 years,
depending on dataset. The recent cooling trend is also evident in the Hadley and the
Goddard Institute for Space Studies (GISS) data, though with some lag, and in ocean
heat content data (Loehle, 2009).
Satellite data since 1978 are clearly nonstationary. The early and late periods show
modest cooling trends, while the middle portion shows a strong warming trend. This
could be an indication of the combination of a linear or other warming trend with one
or more periodic climate cycles, as suggested by Chylek et al. (2009), Loehle (2004),
Klyashtorin and Lyubushin (2003), Schlesinger and Ramankutty (1994), Soon (2005),
Trend Analysis of RSS and UAH MSU Global Temperature Data 1093

and Zhen-Shan and Xian (2007), among others. Zhen-Shan and Xian (2007) and
Klyashtorin and Lyubushin (2003) noted that the result of this combination of forcings
could be to exaggerate the apparent warming (and therefore the apparent greenhouse
forcing effect) of the last third of the 20th Century. Empirical identification of periodic
or semi-periodic climate signals has a precedent in the identification of sunspot cycles
and is thus worthy of consideration for longer periods, which is considered next in the
context of forecasting future climates.
Several papers have forecast a cooling episode for the coming decades, each using
different methods. Keenlyside et al. (2008), for example, suggested that a decade of
cooling could be in store. They based this on climate models forced with more detailed
historical ocean data. Based on a dynamical analysis of coupling between regional
modes of climate variability (Tsonis et al., 2007; Wang et al., 2009), it has been
argued that the Earths climate system in 2001/02 underwent a dynamic shift to a new
state that is predicted to show a flat to cooling trend for several decades (Swanson and
Tsonis, 2009). The timing of this prediction closely matches the trends in satellite data
analyzed herein. De Jager and Duhau (2009) argue that the solar dynamo is currently
undergoing a transition, which began in 2000, from the recent Grand Maximum to a
different regime that will be marked by lower solar activity. Based on analysis of solar
cycles, Landscheidt (2003) shows that the peak in the geomagnetic aa index in 1990
would lead to a prediction, due to an eight year lag, of a peak temperature in 1998,
which is clearly visible in Fig. 7. He further posits that this peak will be followed by
a cooling period expected to last until the next Gleissberg minimum around 2030.
Schlesinger and Ramankutty (1994) used singular spectrum analysis on 1858-1992
surface temperature data to show the effect of a 65-70 year periodic signal, which they
interpreted to be due to oscillations in the North Atlantic currents and thereby on the
weather of nearby land masses. The last peak in their analysis is in 1941, giving a
following expected peak between 2006 and 2011. This expectation nicely coincides
with the downturn observed in this study. Bratcher and Giese (2002) showed that
tropical Pacific temperatures lead global air temperatures by 4 years for sea surface
temperature and 11 years for subsurface temperatures. On the basis of a tropical
Pacific cooling for 8 years from the early 1990s, they forecast a shift in the PDO
similar to the 1976 shift, but with opposite sign, leading to cooling over coming
decades. This forecast seems particularly prescient since NASA identified a shift in
the PDO to its cool phase over the past two years. Klyashtorin and Lyubushin (2003)
demonstrated that a 50-60 year period temperature signal is dominant from about 1650
(the end of the Little Ice Age) in Greenland ice core records, in several very long tree
ring records, and in sardine and anchovy records in marine sediment cores (Klyashtorin
et al. 2009), a result also reported by Biondi et al. (2001). They then modeled 1861-
2000 global temperature data with a linear trend plus a 64.13 year cycle (estimated
from global temperature records). They extrapolated this model, obtaining a peak in
2005 followed by cooling for the next 32 years, since the periodic component is
dominant over the linear warming trend over the scale of several decades. Loehle
(2004, but written in 2001) used two 3000 year paleo-timeseries to estimate periodic
models of varying complexity. Most of the models predicted cooling in the early 21st
Century, with cooling initiation generally near the year 2000. Zhen-Shan and Xian
1094 Energy & Environment Vol. 20, No. 7, 2009

(2007) used Empirical Mode Decomposition on data over 1881-2002. They found a
strong 60 year cycle and a 20 year cycle plus linear trend. Their model peak was in
2001 for China (which slightly leads world values in their analysis). After this time
both cycles are in negative mode. Thus they forecast about 20 more years of cooling.
Except for Keenlyside et al. (2008) the studies agree on the influence of the
multidecadal quasicycle generally associated with the Gleissberg solar cycle (e.g.,
Yousef, 2006), although they disagree on the precise cycle length. Furthermore, the
studies agree on the timing of the transition or peak as being between 1998 and 2005.
This is in spite of being based on very different models and data.
It is worth examining the model of Klyashtorin and Lyubushin (2003) in more
detail. The equation of their fitted curve is:

2 (t t 0
Z (t ) = a + b (t t 0 ) + A cos
T

where: t0 = 1861; t is calendar year; a = 0.332632; b = 0.0038827; A = 0.131125; T


= 64.1453; and j = 1.43346. Their model for the period 1861-2000 AD has a linear
component slope of 0.0388C/decade, a cycle amplitude (peak-to-trough) of 0.262C,
and a cycle period of 64.15 years (Fig. 5a). The IPCC Third Assessment gives a
linear trend over this same period of 0.044C/decade (Folland et al., 2002), but this
higher value results from the series ending on an upturn in the 64 year cycle.
Residuals (Fig. 5b) indicate that the model fits equally well over the entire 140 year
span. This model suggests that a constant rate of warming (the linear term in the
model) has been ongoing since before human influence was significant (i.e., prior to
1950), with no acceleration in recent decades. This is evident from the residuals plot
(Fig. 5b). The oscillations in the model are persistent features of the climate over the
past 350 years, possibly related to solar activity and/or ocean oscillations such as the
Pacific Decadal Oscillation. We can extend this model from 2000 through 2020 and
compare it to the UAH data on a common anomaly basis (Fig. 6). The extrapolated
model almost perfectly captures the temperature turning point in the UAH data.
Continued cooling is forecast by this model until 2037.
Easterling and Wehner (2009) argued that periods of no warming or even cooling of
a decade or two are possible even in the presence of a greenhouse warming trend. They
assume that variability in trends is due to internal oscillations rather than forcings (i.e.,
to LTP). This means that detection of a cooling trend, as in the present study, is not an
automatic refutation of greenhouse theory (although the same argument can be used to
suggest that the most rapid warming interval in the 1990s was itself merely LTP).
However, even in their study such periods were unusual and here we have shown that
the present cooling is correlated with solar activity indices (such as the aa index) and
was predicted by multiple studies dating to 1994. Furthermore, simple LTP can not
produce a periodic signal over 350 years (since the Little Ice Age) nor would it be so
consistently predicted by the various methods discussed here, including those based on
patterns of solar activity and ocean mode indices. Thus suggests that the cooling is not
merely a random fluctuation, but definitive evidence will take more time.
Trend Analysis of RSS and UAH MSU Global Temperature Data 1095

a)
C

Calendar Year

b)
C

Calendar Year

Figure 5. Best fit linear plus periodic model from Klyashtorin and Lyubushin (2003).
a) Model overlaid over global temperature anomaly. Linear trend is 0.0388 C/decade
for the entire period. b) Residuals, showing constant good fit over the record and no
evidence of nonlinearity or recent acceleration of warming.

CONCLUSIONS
Analysis of the satellite data shows a statistically significant cooling trend for the past
12 to 13 years, with it not being possible to reject a flat trend (0 slope) for 16 years.
This is a length of time at which disagreement with climate models can no longer be
attributed to simple LTP. On the other hand, studies cited herein have documented a
50-70 year cycle of climate oscillations overlaid on a simple linear warming trend since
the mid-1800s and have used this model to forecast cooling beginning between 2001
and 2010, a prediction that seems to be upheld by the satellite and ocean heat content
data. Other studies made this same prediction of transition to cooling based on solar
activity indices or from ocean circulation regime changes. In contrast, the climate
1096 Energy & Environment Vol. 20, No. 7, 2009

models predict the recent flat to cooling trend only as a rare stochastic event. The
linear warming trend in these models that is obtained by subtracting the 60-70 yr cycle,
while unexplained at present, is clearly inconsistent with climate model predictions
because it begins too soon (before greenhouse gases were elevated) and does not
accelerate as greenhouse gases continue to accumulate. This model and the empirical
evidence for recent cooling thus provide a challenge to climate model accuracy.

a)
C

Calendar Year

b)
C

Calendar Year

Figure 6. Linear plus period model from Klyashtorin and Lyubushin (2003) overlaid
on satellite data after intercept shift. Dotted line is model extrapolation post-2000
calibration period end. a) UAH. b) RSS.

ACKNOWLEDGEMENTS
Thanks to L. Klyashtorin for prompt provision of his modeling results.
Trend Analysis of RSS and UAH MSU Global Temperature Data 1097

REFERENCES
Bartlett, M.S. 1935. Some aspects of the time-correlation problem in regard to tests of
significance. Journal of the Royal Statistical Society 98:536-543.
Biondi, F., A. Gershunov, and D.R. Cayan. 2001. North Pacific decadal climate variability since
1661. Journal of Climate 14:5-10.
Bratcher, A.J. and B.S. Giese. 2002. Tropical Pacific decadal variability and global warming.
Geophysical Research Letters 29:19,1918,doi:10.1029/2002GL015191.
Casey, K.S. and P. Cornillon. 2001. Global and regional sea surface temperature trends.
Journal of Climate 14:38013818.
Christy, J.R., W.B. Norris, R.W. Spencer, and J.J., Hnilo. 2007. Tropospheric temperature
change since 1979 from tropical radiosonde and satellite measurements. Journal of
Geophysical Research 112, D06102, doi:10.1029/2005JD006881.
Chylek, P., C.K. Folland, G. Lesins, M.K. Dubey, and M. Wang. 2009. Arctic air temperature
change amplification and the Atlantic Multidecadal Oscillation. Geophysical Research
Letters 36, L14801. doi:10.1029/2009GL038777.
Cohn, T.A. and H.F. Lins. 2005. Natures style: Naturally trendy. Geophysical Research Letters
32, L23402, doi:10.1029/2005GL024476.
de Jager, C. and S. Duhau. 2009. Forecasting the parameters of sunspot cycle 24 and beyond.
Journal of Atmospheric and Solar-Terrestrial Physics 71:239-245.
Easterling, D.R. and M.F. Wehner. 2009. Is the climate warming or cooling? Geophysical
Research Letters 36, L08706, doi:10.1029/2009GL037810.
Folland, C.K., T.R. Karl, and M.J. Salinger. 2002. Observed climate variability and change.
Weather 57:269-278.
Fu, Q., C.M. Johanson, S.G. Warren, and D.J. Seidel. 2004. Contribution of stratospheric
cooling to satellite-inferred tropospheric temperature trends. Nature 429:55-58.
Hansen, J., M. Sato, R. Ruedy, K. Lo., D.W. Lea, and M. Medina-Elizade. 2006. Global
temperature change. Proceedings of the National Academy of Sciences 103:14288-14293.
Intergovernmental Panel on Climate Change (IPCC). 2007. Climate change 2007: The physical
science basis. Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. Soloman, S, D.Qin, M. Manning, Z. Chen,
M. Marquis, K.B. Avery, M. Tignor, and H.L Miller (eds.). Cambridge University Press,
Cambridge UK, New York USA. 996 pp.
Jones, P.D. and A. Moberg. 2003. Hemispheric and large-scale surface air temperature
variations: An extensive revision and an update to 2001. Journal of Climate 16:206-223.
Keenlyside, N.S., M. Latif, J. Jungclaus, L. Kornblueh, and E. Roeckner. 2008. Advancing
decadal-scale climate prediction in the North Atlantic sector. Nature 453:84-88.
doi:10.1038/nature06921.
Klyashtorin, L.B., V. Borisov, and A. Lyubushin. 2009. Cyclic changes of climate and major
commercial stocks of the Barents Sea. Marine Biology Research 5:4-17.
Klyashtorin, L.B. and A.A. Lyubushin. 2003. On the coherence between dynamics of the world
fuel consumption and global temperature anomaly. Energy & Environment 14:773-782.
1098 Energy & Environment Vol. 20, No. 7, 2009

Landscheidt, T. 2003. New Little Ice Age instead of global warming? Energy and Environment
14:327-350.
Loehle, C. 2004. Climate change: Detection and attribution of trends from long term geologic
data. Ecological Modelling 171:433-450.
Loehle, C. 2009. Cooling of the global ocean since 2003. Energy & Environment 20:99-102.
Lucarini, V. and G.L. Russell. 2002. Comparison of mean climate trends in the Northern
Hemisphere between National Centers for Environmental Prediction and two atmosphere-
ocean model forced runs. Journal of Geophysical Research 107(D15):ACL 7-1, 4269,
doi:10.1029/2001JD001247.
Quenouille, M.H. 1952. Associated Measurements. Butterworth Scientific Publications,
London. 242 pp.
Santer, B.D., P.W. Thorne, L. Haimberger, K.E. Taylor, T.M.L. Wigley, J.R. Lanzante, S.
Solomon, M. Free, P.J. Gleckler, P.D. Jones, T.R. Karl, S.A. Klein, C. Mears, D. Nychka,
G.A. Schmidt, S.C. Sherwood, and F.J. Wentz. 2008. Consistency of modelled and
observed temperature trends in the tropical troposphere. International Journal of
Climatology 28:1703-1722.
Schlesinger, M.E. and N. Ramankutty. 1994. An oscillation in the global climate system of
period 65-70 years. Nature 367(6465):723-726.
Soon, W.W.-H. 2005. Variable solar irradiance as a plausible agent for multidecadal variations
in the Arctic-wide surface air temperature record of the past 130 years. Geophysical
Research Letters 32, L16712, doi:10.1029/2005GL023429.
Soon, W.W.-H., D.R. Legates, and S.L. Baliunas. 2004. Estimation and representation of long-
term (>40 year) trends of Northern-Hemisphere-gridded surface temperature: A note of
caution. Geophysical Research Letters 31, L03209, doi:10.1029/2003GL019141.
Swanson, K.L. and A.A. Tsonis. 2009. Has the climate recently shifted? Geophysical Research
Letters 36, L06711, doi:10.1029/2008GL037022.
Thompson, D.W.J. and S. Solomon. Interpretation of recent Southern Hemisphere climate
change. Science 296:895-899.
Tsonis, A.A., K. Swanson, and S. Kravtsov. 2007. A new dynamical mechanism for major
climate shifts. Geophysical Research Letters 34, L13705, doi:10.1029/2007GL030288.
Vose, R.S., D.R. Easterling, and B. Gleason. 2005. Maximum and minimum temperature trends
for the globe: An update through 2004. Geophysical Research Letters 32, L23822,
doi:10.1029/2005GL024379.
Wang, G., K.L. Swanson, and A.A. Tsonis. 2009. The pacemaker of major climate shifts.
Geophysical Research Letters 36, L07708, doi:10.1029/2008GL036874.
Wood, R. 2008. Natural ups and downs. Nature 453:43-45.
Yousef, S.M. 2006. 80-120 yr Long-term solar induced effects on the earth, past and
predictions. Physics and Chemistry of the Earth 31:113-122;
doi:10.1016/j.poe.2005.04.004.
Zhen-Shan, L. and S. Xian. 2007. Multi-scale analysis of global temperature changes and trend
of a drop in temperature in the next 20 years. Meteorology and Atmospheric Physics
95:115-121.
99

COOLING OF THE GLOBAL OCEAN SINCE 2003

Craig Loehle, Ph.D.


National Council for Air and Stream Improvement, Inc. (NCASI)
552 S Washington Street, Suite 224
Naperville IL 60540
tel: 630-579-1190; fax: 630-579-1195
cloehle@ncasi.org

ABSTRACT
Ocean heat content data from 2003 to 2008 (4.5 years) were evaluated for trend.
A trend plus periodic (annual cycle) model fit with R2 = 0.85. The linear
component of the model showed a trend of -0.35 (0.2) x 1022 Joules per year. The
result is consistent with other data showing a lack of warming over the past few
years.

Key Words: climate change, ocean heat content, trend analysis

1.0 INTRODUCTION
There is great interest in detecting rates of temperature change in the earth system. It
has been suggested (e.g., Pielke 2003) that changes in ocean heat content should be
particularly informative. A recent study (Lyman et al. 2006) claimed to find rapid
cooling of the ocean between 2003 and 2005, but it was later determined that data from
certain instruments caused a substantial cool bias in the result (Willis et al. 2007,
2008a; Wijffels et al. 2008). A corrected and longer dataset has now become available
to redo this analysis.

2.0 METHODS
The study is based on ocean heat content anomaly (OHCA) data compiled by Willis et
al. (2008b). This monthly dataset (Fig. 1) uses only data from the Argo array of
profiling floats. Heat content is evaluated down to 900 m depth.
The objective is to estimate the linear trend in heat content. However, there is an
obvious one year periodicity in the data (Fig. 1a) as noted by Willis et al. (2008b).
Proper assessment of trend needs to take this into account, especially when the data
are over a 4.5-year interval. Therefore, a model was fit with slope, intercept, and
sinusoidal (1-year fixed period) terms using nonlinear least-squares estimation. The
model allowed the cycle amplitude to change linearly with time. Before fitting, the
data were minimally smoothed with a 1-2-1 filter (Fig. 1b).
100 Energy & Environment Vol. 20, No. 1&2, 2009

Figure 1.
a) Ocean heat content smoothed with a 1-2-1 filter
b) Heat content smoothed with 1-2-1 filter and overlaid with best-fit linear plus
sinusoidal (seasonal) model (R2 = 0.85)
c) Heat content smoothed with 1-2-1 filter and overlaid with linear trend portion of
best-fit model (slope = -0.35 x 1022 J/yr)

3.0 RESULTS
The model, fit to the smoothed data, gave an excellent fit (r = 0.922, R2 = 0.85) and
showed clearly that there is an annual periodicity in the data (Fig. 1b), probably due
to the north-south asymmetry in ocean area and the effect of orbital variations over the
Cooling of the Global Ocean Since 2003 101

year. The peak-to-trough amplitude of the model is 6.30 x 1022 Joules (J) at the
beginning of the period and declines to 3.88 x 1022 J at the end, showing a damping of
the cycle over the 4.5-year period. The slope of the linear component of the model
(Fig. 1c) is -0.35 x 1022 J/yr, which is 6.9% of the average annual cycle amplitude. The
95% confidence intervals on the trend are from -0.148 x 1022 to -0.550 x 1022 J/yr.
This result clearly excludes warming as a possible interpretation of this data.
Examination of residuals from the model fit shows no evidence of nonlinearities,
indicating a constant linear cooling trend. Over the 4.5-year period of the data, this
equates to a total non-seasonal loss of 1.572 x 1022 (0.668 to 2.48 x 1022) J of heat,
which is 31% of the average annual cycle amplitude over the interval.

4.0 DISCUSSION
It has previously been estimated by Willis et al. (2004) that from 1993 to 2003 the
upper ocean gained 8.1 (1.4) x 1022 J of heat. This study estimates a loss since then
of from 0.668 to 2.48 x 1022 J, or 19.4% (up to 31%) of the gain of the prior decade.
Ishii and Kimoto (In Press) also show a bias-corrected cooling from 2003 to 2006. On
an annual basis, this is a cooling of 0.35 x 1022 J compared to 0.81 x 1022 J warming
for 1993 to 2003 (Willis et al. 2004) and slightly less for the same period to 700 m in
Ishii and Kimoto (in press). Dominguez et al. (2008) show a 700 m depth annual
warming from 1961 to 2003 of 0.38 x 1022 J. Thus the estimate of cooling in the
present study is not out of line with past results. It is also consistent with satellite and
surface instrumental records that do not show a warming trend over recent years.
Another bias-corrected estimate (Gouretski and Koltermann 2007) is based on
depth profiles too different to make a comparison. By comparison, Willis et al.
(2008a) do not find any significant trend (slight negative trend) for 2003 to 2006, but
had a shorter record and performed their trend analysis using simple annual means.
Heat loss from the ocean has been estimated to also have occurred in the 1980s (Ishii
and Kimoto, In Press; Gouretski and Koltermann 2007; Levitus et al. 2001). The data
also indicate an interesting damping with time of the annual fluctuations in heat gain
and loss (Fig. 1b). While the current study takes advantage of a globally consistent
data source, a 4.5-year period of ocean cooling is not unexpected in terms of natural
fluctuations. The problem of instrumental drift and bias is quite complicated,
however, (Domingues et al. 2008; Gouretski and Koltermann 2007; Wijffels et al.
2008; Willis et al. 2004, 2008a) and it remains possible that the result of the present
analysis is an artifact.

ACKNOWLEDGMENTS:
Thanks to J. Willis for providing ARGOS data.
102 Energy & Environment Vol. 20, No. 1&2, 2009

REFERENCES
Domingues, C.M., Church, J.A., White, N.J., Gleckler, P.J., Wijffels, S.E., Barker, P.M. and
Dunn, J.R. 2008, Improved Estimates of Upper-Ocean Warming and Multi-Decadal Sea-
Level Rise, Nature, 453, 1090-1093, doi:10.1038/nature07080.
Gouretski, V. and Koltermann, K.P. 2007, How Much is the Ocean Really Warming?,
Geophysical Research Letters, 34, L01610, doi:10.1029/2006GL027834.
Ishii, M. and Kimoto, M. in press, Reevaluation of Historical Ocean Heat Content Variations
with an XBT Depth Bias Correction, Journal of Oceanography.
Levitus, S., Antonov, J.I., Wang, J., Delworth, T.L., Dixon, K.W. and Broccoli, A.J. 2001,
Anthropogenic Warming of Earths Climate System, Science, 292(5515), 267-270
Lyman, J.M., Willis, J.K. and Johnson, G.C. 2006, Recent Cooling of the Upper Ocean.
Geophysical Research Letters, 33(18), L18604, doi:10.1029/2006GL027033.
Pielke, R.A., Sr. 2003, Heat Storage Within the Earth System, Bulletin of the American
Meteorological Society, 84(3), 331-335, doi:10.1175/BAMS-84-3-331.
Wijffels, S.E., Willis, J., Domingues, C.M., Barker, P., White, N.J., Gronell, A., Ridgway, K. and
Church, J.A. 2008, Changing Expendable Bathythermograph Fall Rates and Their Impact
on Estimates of Thermosteric Sea Level Rise, Journal of Climate, 21, 56575672,
doi:10.1175/2008JCLI2290.1.
Willis, J.K., Roemmich, D. and Cornuelle, B. 2004, Interannual Variability in Upper Ocean Heat
Content, Temperature, and Thermosteric Expansion on Global Scales. Journal of
Geophysical Research, 109(12), C12036, doi:10.1029/2003JC002260c.
Willis, J.K., Lyman, J.M., Johnson, G.C. and Gilson, J. 2007, Correction to "Recent Cooling of
the Upper Ocean," Geophysical Research Letters, 34(16), L16601,
doi:10.1029/2007GL030323.
Willis, J.K., Lyman, J.M., Johnson, G.C. and Gilson, J. 2008a, In Situ Data Biases and Recent
Ocean Heat Content Variability, Journal of Atmospheric and Oceanic Technology,
doi:10.1175/2008JTECHO608.1.
Willis, J.K., Chambers, D.P. and Nerem, R.S. 2008b, Assessing the Globally Averaged Sea
Level Budget on Seasonal to Interannual Timescales, Journal of Geophysical Research, 113,
C06015, doi:10.1029/2007JC004517.

You might also like