Content uploaded by Pablo Ortega
Author content
All content in this area was uploaded by Pablo Ortega on Jan 27, 2017
Content may be subject to copyright.
Estimating changes in global temperature since the pre-industrial period1
Ed Hawkins⇤, Pablo Ortega, Emma Suckling2
NCAS-Climate, Department of Meteorology, University of Reading, Reading, UK3
Andrew Schurer, Gabi Hegerl4
School of GeoSciences, Grant Institute, University of Edinburgh, Edinburgh, UK5
Phil Jones6
School of Environmental Sciences, University of East Anglia, Norwich, UK and Center of
Excellence for Climate Change Research, Department of Meteorology, King Abdulaziz
University, Jeddah, Saudi Arabia
7
8
9
Manoj Joshi, Timothy J. Osborn10
School of Environmental Sciences and Climatic Research Unit, University of East Anglia,
Norwich, UK
11
12
Val´
erie Masson-Delmotte13
Institut Pierre Simon Laplace, Laboratoire des Sciences du Climat et de l’Environnement
(CEA-CNRS-UVSQ), Gif-sur-Yvette, France
14
15
Juliette Mignot16
Climate and Environmental Physics, Physics Institute & Oeschger Center for Climate Change
Research, University of Bern, Switzerland and LOCEAN/IPSL (Sorbonne Universit´
es,
UPMC-CNRS-IRD-MNHN), France
17
18
19
Peter Thorne20
Department of Geography, Maynooth University, Maynooth, County Kildare, Ireland21
Geert Jan van Oldenborgh22
Koninklijk Nederlands Meteorologisch Instituut, De Bilt, Netherlands23
Generated using v4.3.2 of the AMS L
A
T
E
X template 1
LaTeX File (.tex, .sty, .cls, .bst, .bib)
⇤Corresponding author address: Ed Hawkins, Department of Meteorology, University of Reading,
Reading, UK. RG6 6BB.
24
25
E-mail: e.hawkins@reading.ac.uk26
2
ABSTRACT
The United Nations Framework Convention on Climate Change (UNFCCC)
process agreed in Paris to limit global surface temperature rise to ‘well below
2C above pre-industrial levels’. But what period is ‘pre-industrial’? Some-
what remarkably, this is not defined within the UNFCCC’s many agreements
and protocols. Nor is it defined in the IPCC’s Fifth Assessment Report (AR5)
in the evaluation of when particular temperature levels might be reached be-
cause no robust definition of the period exists. Here we discuss the important
factors to consider when defining a pre-industrial period, based on estimates
of historical radiative forcings and the availability of climate observations.
There is no perfect period, but we suggest that 1720-1800 is the most suit-
able choice when discussing global temperature limits. We then estimate the
change in global average temperature since pre-industrial using a range of
approaches based on observations, radiative forcings, global climate model
simulations and proxy evidence. Our assessment is that this pre-industrial
period was likely 0.55 0.80C cooler than 1986-2005 and that 2015 was
likely the first year in which global average temperature was more than 1C
above pre-industrial levels. We provide some recommendations for how this
assessment might be improved in future and suggest that reframing temper-
ature limits with a modern baseline would be inherently less uncertain and
more policy-relevant.
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
3
Better defining (or altogether avoiding) the term ‘pre-industrial’
would aid interpretation of internationally agreed global temperature
limits and estimation of the required constraints to avoid reaching
those limits.
47
The basis for international negotiations on climate change has been to ‘prevent dangerous an-48
thropogenic interference with the climate system’, using the words of the United Nations Frame-49
work Convention on Climate Change (UNFCCC). The 2015 Paris COP21 Agreement1, aims to50
maintain global average temperature ‘well below 2C above pre-industrial levels and to pursue51
efforts to limit the temperature increase to 1.5C above pre-industrial levels’. However, there is52
no formal definition of what is meant by ‘pre-industrial’ in the UNFCCC or the Paris Agreement.53
Neither did the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change54
(IPCC) use the term when discussing when global average temperature might cross various levels,55
due to the lack of a robust definition (Kirtman et al., 2013).56
Ideally, a pre-industrial period should represent the mean climate state just before human ac-57
tivities started to demonstrably change the climate through combustion of fossil fuels. Here we58
discuss which time period might be most suitable, considering various factors such as radiative59
forcings, availability of observations and uncertainties in our knowledge.60
We will focus on global temperatures, specifically for informing discussions on future tempera-61
ture limits, and make an assessment of how much global average temperature has already warmed62
since our defined pre-industrial period using a range of approaches. We will also provide rec-63
ommendations for: (i) how future international climate reports and agreements might use this64
assessment; and (ii) how the assessment itself may be improved in future, particularly regarding65
the use of instrumental data, proxy evidence and simulations of past climate.66
1http://unfccc.int/resource/docs/2015/cop21/eng/l09r01.pdf
4
Relevance of the pre-industrial period for crossing global temperature thresholds67
In the absence of a formal definition for pre-industrial, the IPCC AR5 made a pragmatic choice68
to reference global temperature to the mean of 1850-1900 when assessing the time at which par-69
ticular temperature levels would be crossed (Kirtman et al., 2013). In the final draft, 1850-190070
was referred to as ‘pre-industrial’, but at the IPCC AR5 plenary approval session, ‘a contact group71
developed a proposal, in which reference to “pre-industrial” is deleted, and this was adopted [by72
the governments]’ (IISD, 2013). However, the term ‘pre-industrial’ was used in AR5, often incon-73
sistently, in other contexts, e.g., when discussing atmospheric composition, radiative forcing (the74
year 1750 is used as a zero-forcing baseline), sea level rise and paleoclimate information. These75
discussions highlight the importance of defining pre-industrial consistently and more precisely.76
In AR5, the observed increase in global temperature was calculated as the mean of 1986-200577
minus the mean of 1850-1900 in the HadCRUT4 dataset (0.61C, Morice et al., 2012), which was78
the only combined global land and ocean temperature dataset available back to 1850 at the time.79
The 1986-2005 modern period was chosen2because the design of the CMIP5 simulations required80
a recent reference baseline for the projections of future climate (discussed further in Hawkins and81
Sutton, 2016). Note that the warming between 1850-1900 and the most recent decade covered82
(2003-2012) was given by AR5 as 0.78 ±0.03C (IPCC, 2013).83
The choice of 1850-1900 as the historical reference period benefits from relatively widespread,84
but still sparse, temperature observations, and quantified uncertainties in the estimates of global85
temperature. Since the AR5, two further datasets have been produced that allow a comparison86
for the 1850-1900 period. In the Cowtan and Way (2014) dataset (hereafter CW14), which is87
based on interpolating the spatial gaps in HadCRUT4, the difference from 1850-1900 to 1986-88
2The World Meteorological Organisation uses 1981-2010 for ‘operational normals’, which is very similar to the 1986-2005 period in terms of
global mean temperature.
5
2005 is 0.65C and in the Berkeley Earth global land & sea data (BEST-GL, berkeleyearth.org),89
it is 0.71C, suggesting that the AR5 value may be slightly too low3. Also, Cowtan et al. (2015)90
presented GCM-based evidence that sparse observation-based datasets may have significantly un-91
derestimated the changes in global surface air temperature due to slower warming regions being92
preferentially sampled in the past. However, infilling the gaps in the early period is especially93
problematic due to the sparse observations and may accentuate the dominant observed anomaly.94
However, some anthropogenic warming is estimated to have already occurred by 1850 (Hegerl95
et al., 2007; Schurer et al., 2013; Abrams et al., 2016) as greenhouse gas concentrations had started96
increasing around a century earlier (Fig. 1). On the other hand, the 1880s and 1890s were cooler97
than the preceding decades because of the radiative impact of aerosols from several volcanic erup-98
tions (Fig. 1) which may have compensated for the earlier anthropogenic influence. It is therefore99
plausible that a ‘true’ pre-industrial temperature could be warmer or cooler than 1850-1900, de-100
pending on the balance of these two factors. A key question which we will consider is how101
representative the 1850-1900 period is for pre-industrial global average temperature.102
Defining a suitable pre-industrial period using radiative forcing estimates103
Anthropogenic climate change is occurring on top of: (i) internal climate variability, such as104
ENSO, the Pacific Decadal Oscillation (PDO), Atlantic Multi-decadal Variability (AMV) and pos-105
sibly longer timescales (see Deser et al. (2010) for a review) and (ii) multi-decadal scale variations106
in natural radiative forcings, such as solar activity, changes in Earth’s orbit and the frequency of107
large volcanic eruptions.108
3These three datasets all use the Hadley Centre estimates for the sea surface temperatures since 1850 (HadSST3, Kennedy et al., 2011), and are
based on similar land-based observations, so are not independent.
6
A pre-industrial climate should therefore be defined as a period close to present but which is109
before the ‘industrial age’, with small anthropogenic forcings. Ideally, levels of natural forcings110
would also be similar to present and widespread direct or indirect observations would be available.111
The better part of a century would appear to be required to average over the longer-timescale112
internal variations.113
Unfortunately, such a perfect time period does not exist so compromises have to be made. In114
particular, there are very few instrumental temperature records before 1850 which limits our abil-115
ity to determine pre-1850 global temperatures. Changes in land-use and other human activities116
(e.g., biomass burning, deforestation) may have altered the composition of the atmosphere several117
millennia ago (Ruddiman, 2003; Ruddiman et al., 2016). There are also variations in greenhouse118
gas concentrations (of a few ppm) before 1700 (Bauska et al., 2015). However, we assume that119
these early influences are not relevant for defining a pre-industrial period for use by policymakers.120
Bradley et al. (2016) identified the period 725-1025 as a ‘medieval quiet period’, without major121
tropical eruptions or solar variations, and which might represent a reference climate state. How-122
ever, proxy evidence suggests a slow decline of global temperatures, surface ocean temperatures123
and reductions in sea level over the last two millennia, which has been attributed to orbital forcing124
(Kaufman et al., 2009) or to increasing volcanic activity (McGregor et al., 2015; Stoffel et al.,125
2015; Kopp et al., 2016). Given this multi-millennial trend, whatever its cause, it makes sense to126
chose a reference period as close to the present as possible.127
An important moment at the start of the industrial age was when James Watt patented the steam128
engine condenser in 1769, dramatically improving Thomas Newcomen’s 1712 steam engine de-129
sign. Various agricultural revolutions also began around the same time. However, there was130
probably only a small climate effect of these developments for several decades at least. For these131
7
reasons, historical anthropogenic radiative forcings are often considered relative to 1750 levels132
(Solomon et al., 2007; Meinshausen et al., 2011).133
It is also important to ensure that the natural forcings in any chosen period are not unusual,134
compared to the present (Fig. 1). The period before 1720, often called the Little Ice Age (Mann135
et al., 2009), was influenced by several large tropical volcanic eruptions in the 1600s (Briffa et al.,136
1998; Crowley et al., 2008; Gao et al., 2008; Sigl et al., 2013) and the Maunder Minimum in solar137
activity which finished in the early 1700s (Steinhilber et al., 2009; Lockwood et al., 2014; Usoskin138
et al., 2015). The period after 1800 is influenced by the Dalton Minimum in solar activity and139
the large eruptions of an unlocated volcano in 1808/9, Tambora (1815, Raible et al., 2016), and140
several others in the 1820s and 1830s. In addition, greenhouse gas concentrations had already141
increased slightly by this time (Fig. 1).142
In contrast, between 1720 and 1800 the evidence suggests that natural radiative forcings are143
closer to modern levels, with only very weak anthropogenic forcings. It could be argued that144
this period has slightly anomalously low volcanic activity, including one relatively small tropical145
eruption (Makian, Indonesia in 1761) and one long-lasting northern extra-tropical eruption (Laki,146
Iceland in 1783). This issue is returned to later.147
There is also no evidence for unusual AMV/PDO variability during the 1720-1800 period (e.g.,148
Gray et al., 2004; MacDonald and Case, 2005), suggesting that these modes of variability are not149
expected to significantly affect the multi-decadal temperature average.150
We therefore suggest that 1720-1800 is the most suitable period to be called pre-industrial for151
assessing global temperature levels in terms of the radiative forcings and we concentrate on this152
period in the analysis which follows. Different choices may be made if considering changes in153
other variables (Knutti et al., 2015), such as regional temperatures, rainfall, sea level, carbon154
storage or glacier extents, but assessing those is beyond the scope of this study.155
8
Using three different approaches, we now address two related questions, based on the reference156
periods used in IPCC AR5: (i) what is the global temperature change from our pre-industrial157
choice to a recent baseline (1986-2005), and (ii) is 1850-1900 a reasonable pragmatic surrogate158
for the pre-industrial period? We also consider the precision to which such questions can be159
answered.160
Approach 1: using radiative forcings161
Our first approach uses radiative forcings to estimate changes in global temperature before the162
available observations. The Coupled Model Intercomparison project, phase 5 (CMIP5) provides163
estimated historical radiative forcings for 1765-2005, referenced to 1750, and for a range of repre-164
sentative concentration pathways (RCPs) after 2005 (Meinshausen et al., 2011). We use RCP4.5165
for the period 2006-2015 but this makes little difference.166
We adopt a weighted least-squares multiple linear regression approach, using the radiative forc-167
ings (provided in Wm2), multiplied by individual scaling factors, to best fit the observed global168
mean surface temperature (GMST):169
GMST(t)= 4
Â
f=1
afFf(t)!+gE(tt)b(1)
We consider four radiative forcings (Ff, with scalings af): greenhouse gases, other anthropogenic170
effects (mainly aerosols, land use and ozone), solar, and volcanic activity. Annual means are used171
everywhere. We also use an ENSO index (E, scaled by g) as a ‘forcing’ to remove the effects of172
the leading mode of interannual variability from the observations. This Eindex is defined as the173
linearly detrended Nino3.4 anomaly from 1857-2015 (Kaplan et al., 1998) and zero before 1857,174
with a lag (t) of 4 months to maximise the variance explained (i.e. the annual mean is a September175
to August average). A similar approach to fitting global temperatures was taken by Lean and Rind176
9
(2009) and Suckling et al. (2016). All global temperature data are referenced to 1986-2005 to177
match the analysis in IPCC AR5 (Kirtman et al., 2013) and bis a constant offset to account for178
this reference period.179
We perform the analysis separately for five global temperature datasets to represent the uncer-180
tainty in temperature reconstructions, although this is an underestimate of the true uncertainty181
because they are all based on similar observations. For HadCRUT4, BEST-GL and CW14, the182
multiple linear regression is performed over the period 1850-2015. The NOAA GlobalTemp (Karl183
et al., 2015) and NASA GISTEMP (Hansen et al., 2010) datasets are fitted over the full extent of184
their available data (1880-2015). We use the HadCRUT4 uncertainties in the weighted regression185
(except for BEST-GL and NOAA GlobalTemp which have their own uncertainty estimates), so186
that the older (and more uncertain) data has less weight.187
Fig. 2a shows one estimate of GMST (HadCRUT4) and the scaled forcings for the full 1765-188
2015 period, using the regression parameters derived over 1850-2015. The correlation between189
the scaled forcings (including ENSO) and observed temperatures is 0.94 for each of the global190
datasets.191
There are two ways to estimate a change in temperature using this approach4. Firstly, we can192
average the scaled forcings over 1765-1800 to produce an estimate of the pre-industrial global193
temperature for each dataset with associated uncertainties, accounting for the covariance in derived194
af’s. Note that this is the longest period available using the CMIP5 forcings in the 1720-1800195
period. The Paleoclimate Modelling Intercomparison Project (PMIP) protocol does not currently196
provide consistent forcing estimates in this way for the 850-1850 period (Schmidt et al., 2012).197
For the five temperature datasets, the best estimates are found to range from 0.64 0.76C with198
4These estimates are largely insensitive to whether a lag is introduced in the greenhouse gas forcing (as done in Lean and Rind, 2009), or if only
the 1900-2015 period is used for fitting or if the anthropogenic forcings are combined before fitting.
10
uncertainties of around ±0.05C. Alternatively, the value of the regression constant (b) is an199
estimate of the temperature change from a state of zero forcing (in this case 1750) to 1986-2005.200
For the five temperature datasets, branges from 0.69 0.82C (with uncertainties of ±0.02C),201
which is around 0.06C larger than using the 1765-1800 average. This difference is consistent with202
the small increase in greenhouse gas forcing and the relatively weak volcanic forcing after 1765.203
Overall, these results suggest that pre-industrial was slightly cooler than the 1850-1900 period.204
Also, the derived estimates for the warming are all larger than the value used in IPCC AR5205
(0.61C), with the HadCRUT4-based estimates being the smallest and GISTEMP the largest. The206
differences between estimates from the various datasets are larger than the stated uncertainties, and207
are dominated by the uncertainty in global change since 1850, partly related to the way missing208
data is treated. The CW14 dataset, which interpolates between the gaps in HadCRUT4, finds209
slightly larger warming, consistent with Cowtan et al. (2015) who show a similar effect when210
examining simulated data to determine the effects of incomplete spatial coverage. The NOAA211
and GISTEMP datasets also use slightly different interpolation techniques. These various infilling212
approaches may reduce the bias from poor spatial sampling, especially for fast warming regions213
such as the Arctic, but may simply accentuate the dominant anomaly and add uncertainty. These214
inconsistencies merit further investigation elsewhere.215
This approach does not account for non-linearities in the temperature response to forcings, or216
uncertainties in the assumed CMIP5 forcing history itself, which are likely to be particularly large217
for aerosols (e.g. Carslaw et al., 2013; Stevens, 2013) and ozone (Marenco et al., 1994). However,218
this approach does allow for varying sensitivities (af) to the different assumed forcings (or ‘effi-219
cacies’) (Hansen et al., 2005; Shindell, 2014). Another approach would be to use a simple energy220
balance model, tuned to the observational record (e.g. Osborn et al., 2006; Aldrin et al., 2012) and221
this could be examined in future work.222
11
Approach 2: using last millennium simulations223
An alternative approach to considering the forcings alone is to use ‘last millennium’ ensembles224
(LMEs) which use global climate models (GCMs) to simulate global climate from 850 to 2005225
using the PMIP3 estimates of greenhouse gas concentrations, solar variations and volcanic erup-226
tions detailed by Schmidt et al. (2012). Here we consider three ensembles with different GCMs:227
GISS E2-R (3 members, Schmidt et al., 2014), CESM1 (10 members, Otto-Bliesner et al., 2016)228
and MPI-ESM (3 members, Jungclaus et al., 2014). These are the only models to have made con-229
tinuous simulations available for the whole time period using all radiative forcings5and multiple230
ensemble members (Fig. 2b).231
In the GCM simulations, 1720-1800 is 0.00 0.06C cooler than 1850-1900 (using ensemble232
means), which is slightly smaller than the result using Approach 1. However, the three GCMs233
produce very different estimates for the warming from 1720-1800 until 1986-2005 (0.51 ±0.08C234
for CESM1, 1.04 ±0.07C for GISS E2-R and 0.91 ±0.04C for MPI-ESM)6. These differences235
are not what would be expected due to climate sensitivity alone as CESM1 has the largest tran-236
sient climate response (TCR, 2.2K) and GISS E2-R the smallest (1.5K). It is more likely that the237
differences are due to a combination of several factors, including climate sensitivity, different am-238
plitude responses to anthropogenic aerosols and volcanic eruptions (Stoffel et al., 2015), different239
assumed forcings (e.g., the size of the 1761 eruption), and different implementations of the forc-240
ings. In addition, the global temperature response to volcanic eruptions appears to be larger in the241
GCMs than the real world (e.g. Schurer et al., 2013), although Stoffel et al. (2015) suggest this242
effect is much reduced with an improved representation of the aerosol microphysics.243
5Note that the GISS E2-R simulations used a different aerosol forcing over the historical period than the CMIP5 historical simulations performed
with the same GCM. The PMIP3 simulations warm by about 0.3K more than the CMIP5 simulations (not shown).
6We also tested Approach 1 using the global temperatures from the PMIP simulations. This produced compatible values for the warming
(0.45 ±0.09, 1.09 ±0.04and 0.90 ±0.06C respectively), building confidence in that approach.
12
Given the diversity in global temperature response, a robust estimate of change in global tem-244
perature since pre-industrial using these simulations should consider scaling the responses to the245
observations or using detection and attribution techniques on the range of simulations available246
(Schurer et al., 2013; Otto-Bliesner et al., 2016). In addition, the comparison with observations is247
not necessarily like-with-like given sparse observations and different use of air or sea temperatures248
(Cowtan et al., 2015; Richardson et al., 2016).249
However, an additional use for the LMEs is to examine uncertainty in the estimate of pre-250
industrial temperatures due to internal variability alone. This can be done by considering the251
spread of estimated change using the ten CESM1 ensemble members (s=0.05K), which sug-252
gests an uncertainty of around ±0.1C. Note that this range is similar to the uncertainty ranges253
from long instrumental records discussed below. The other ensembles are too small to reliably254
estimate this range. We also use the CESM1 simulations to consider issues of differential seasonal255
warming in the Appendix.256
Approach 3: using long instrumental records257
The above two approaches have considered the response to estimated radiative forcings. An258
alternative approach to estimate GMST further back in time is to use direct observations from259
long instrumental records and calibrate them against each of the five global mean temperature260
datasets.261
For example, Central England Temperature (HadCET, Manley, 1974; Parker et al., 1992, here-262
after referred to as CET) is available for 1659-present. CET covers just 0.005% of the Earth’s263
surface but is highly correlated with GMST on multi-decadal timescales (Sutton et al., 2015).264
Here, we utilise this correlation and scale GMST to CET:265
CET =dGMST +e(2)
13
using the overlapping periods (1850-2015), and adopt the same parameters to scale CET back to266
1659 as an estimate of GMST (Fig. 3a). When using HadCRUT4 as GMST, d=1.20 ±0.23,267
although other global temperature datasets give lower values (e.g., for BEST-GL, d=1.06 ±268
0.21). The major caveats to this approach are that we assume the historical temperature estimates269
are unbiased, and that the relationship between GMST and CET is the same whatever forcing is270
dominant, neither of which may be true (Zanchettin et al., 2013; Haarsma et al., 2013, and see271
Appendix).272
We take the mean of the scaled CET over two periods: (i) 1765-1800 (for consistency with273
Approach 1) and (ii) 1720-1800 (the full period identified from the radiative forcing history).274
An additional issue that arises from scaling a local record to global temperatures is the possible275
regional effect of external forcing. In particular, the eruption of Laki (located in Iceland) in 1783276
likely only had a small global effect, but it certainly influenced western Europe (Thordarson and277
Self, 2003). Therefore the years 1783 and 1784 are removed from the averages due to the eruption278
of Laki to avoid biasing the estimated temperature change. However, this does not change the279
results significantly.280
These two periods produce consistent estimates for the warming to 1986-2005: 0.75 ±0.10 C281
(for 1765-1800) and 0.64±0.08C (for 1720-1800) when using HadCRUT4 for GMST. The other282
global temperature datasets give larger values for the warming to 1986-2005, by up to 0.09C283
(Fig. 3a). The quoted uncertainty ranges account for the uncertainties in the regression parameters284
and assume the uncertainty in each CET annual mean from 1720-1800 is independent and equal285
to 0.2C (based on Parker, 2010).286
The difference between the two averaging periods is mainly because the 1720s and 1730s were287
unusually warm in the CET record. Internal climate variability and a recovery from the nega-288
14
tive forcings of the previous decades are possible explanations, although this warmth was less289
pronounced in some other European instrumental records (e.g. Berlin) (Jones and Briffa, 2006).290
Figs. 3b repeats this analysis with the Berkeley global land temperature (BEST-Land, Rohde291
et al., 2013), which starts in 1753. A similar approach was adopted by Mann (2014). Using BEST-292
Land produces a consistent but slightly lower warming than derived with CET. Using the scaled293
temperatures over the 1753-1800 period, the estimates of the warming to 1986-2005 range from294
0.62 ±0.10C for HadCRUT4 to 0.71 ±0.12C for GISTEMP.295
It may seem surprising that the error bars are not smaller for the BEST-Land dataset than for296
CET. The regression uncertainty is indeed much larger for the local example, however the error297
in representing the whole global land area with sparse data is larger than in representing central298
England with a small number of stations. These two sources of uncertainty combine to give similar299
overall ranges. Note that BEST-Land looks very similar to the long European records and the300
variability increases further back in time (also for CET), highlighting that fewer and fewer (mostly301
European) stations are used in the reconstruction.302
We also consider a long temperature series from the Netherlands, referenced to De Bilt, which303
starts in 1706 (Van Engelen and Nellestijn, 1990) and a Central Europe instrumental series from304
Dobrovoln`
y et al. (2010) which starts in 1760, which are also both well correlated with GMST305
in the overlapping period. These results are summarised in Fig. 4 which shows that the Central306
Europe series consistently produces slightly lower estimates of the warming than CET or BEST-307
Land.308
Overall assessment309
We consider that approaches based on the radiative forcings and scaled instrumental obser-310
vations currently produce more reliable estimates of the global temperature change since pre-311
15
industrial than the last millennium GCM simulations. This weighting of methods could change312
in future with additional evidence, analysis and model development (see implications discussed313
below). Furthermore, the estimates using radiative forcings are generally larger than when using314
the observational datasets, as summarised in Fig. 4. Much of the uncertainty in the assessment315
derives from the range of global temperature change estimates available since 1850. For example,316
the uninterpolated HadCRUT4 dataset produces lower values than the other infilled records.317
Our overall assessment is that the change in global average temperature from pre-industrial to318
1986-2005 is ‘likely’ between 0.55 0.80C.319
This range reflects the authors’ aggregated assessment of the three approaches and contains vir-320
tually all of the best estimates using the various combinations of regional and global temperature321
datasets and scaled radiative forcing estimates. Note that there are potentially important uncertain-322
ties in each approach which we cannot quantify. As in IPCC AR5 we consider that ‘likely’ refers323
to greater than 66% probability, although this is not a formal uncertainty quantification.324
It is also helpful to assess a lower bound and we suggest that the warming since pre-industrial325
is ‘likely’ greater than 0.60C, implying that the value used by IPCC AR5 for the warming since326
1850-1900 (0.61C) was probably smaller than the true change since pre-industrial. Such dif-327
ferences matter more when considering the chance of crossing lower temperature levels such as328
1.5C than when considering higher values.329
Using this lower bound, 2015 was the first year to be more than 1C above pre-industrial levels330
in each global temperature dataset (Fig. 5). 2016 is currently on track to be warmer than 2015, but331
future years could still be cooler than 2015 due to internal variability, such as a La Ni ˜
na event.332
The available proxy-based evidence is consistent with our assessment, but currently too un-333
certain to make more precise estimates, partly due to different seasonal signals (see Appendix).334
16
However, defining a pre-industrial period offers a target for proxy reconstructions to aid future335
assessments.336
Conclusions & implications337
We have examined estimates of historical radiative forcings to determine which period might338
be most suitable to be termed pre-industrial and used several approaches to estimate a change in339
global temperature since this pre-industrial reference period. The main conclusions are:340
1. The 1720-1800 period is most suitable to be defined as pre-industrial in physical terms, al-341
though we have incomplete information about the radiative forcings and very few direct ob-342
servations during this time. However, this definition offers a target period for future analysis343
and data collection to inform this issue.344
2. The 1850-1900 period is a reasonable pragmatic surrogate for pre-industrial global mean tem-345
perature. The available evidence suggests it was slightly warmer than 1720-1800 by around346
0.05C, but this is not statistically significant.347
3. We assess a ‘likely’ range of 0.55 0.80C for the change in global average temperature348
from pre-industrial to 1986-2005.349
4. We also consider a likely lower bound on warming from pre-industrial to 1986-2005 of350
0.60C, implying that the AR5 estimate of warming was probably too small and that 2015351
was the first year to be more than 1C above pre-industrial levels.352
We have assumed in the motivation for this discussion and choice of reference periods that the353
UNFCCC agreements on temperature limits refer to anthropogenic increases only, but this is not354
explicitly stated. We have not attempted to attribute the observed increase in global temperatures355
(but see Schurer et al., 2013; Otto et al., 2015); non-anthropogenic factors (including internal356
17
variability) may have either offset or contributed to the warming. We have attempted to minimise357
issues of varying natural forcing and internal variability, but this effect cannot be removed entirely.358
Our chosen pre-industrial period likely has slightly weaker volcanic activity than a typical period359
and the modern reference period (1986-2005) includes the large Pinatubo eruption. These effects360
would bias our estimated change in temperature to be slightly too low, highlighting the value of361
assessing a lower bound in the warming since pre-industrial. We also note that future climate pro-362
jections do not usually include volcanic eruptions so choosing a relatively weak volcanic baseline363
is perhaps appropriate. The recent period has a slightly positive PDO index which would act as364
a small positive bias for some of our estimates, but this modern reference period will likely be365
updated for the next IPCC assessment.366
There are a number of ways that this assessment could be improved. Better understanding of367
historical radiative forcings, particularly of volcanic eruptions, solar activity and anthropogenic368
aerosols, would help narrow the uncertainties in past global and regional temperature change. We369
did not include the estimates for pre-industrial temperature from the last millennium simulations in370
this assessment due to the diverse derived values, which is due to differences in both the forcings371
used and climate sensitivity (Fern´
andez-Donado et al., 2013). Future work might consider scaling372
the simulations (Schurer et al., 2013) or use of simple Energy Balance Models (EBMs).373
However, we may not necessarily expect simulated and observed values to agree, even in the374
case of perfect knowledge of radiative forcings and climate sensitivity. This is because the global375
observations are a sparse blend of sea surface temperatures over the ocean and air temperatures376
over the land, whereas virtually all analyses of GCM simulations use air temperatures with com-377
plete global coverage. Cowtan et al. (2015) and Richardson et al. (2016) used GCM simulations378
to suggest that if we had complete coverage of air temperature, the observed change from 1850379
18
to present would be 24 ±15% larger than currently estimated in HadCRUT4. The use of infilled380
temperature datasets only partly overcomes this issue.381
This creates a dilemma - are the temperature limits adopted by the UNFCCC designed to use382
observationally-based estimates of global temperature change (as generally used here) or on what383
those observations mean for a ‘true’ global mean air temperature change (as used in most climate384
impact assessments)? The available evidence suggests that the latter is larger. If such findings are385
borne out by further research, and if the ‘true’ change is what is desired by UNFCCC, then our386
assessed temperature change since pre-industrial is too small and should probably be increased by387
0.05 0.10C.388
It is possible to obtain additional data for the historical period. Recovery of additional instru-389
mental observations of temperature and sea level pressure from undigitised hand-written logbooks390
from ships and in currently data sparse regions could significantly aid similar future assessments.391
Some such efforts are ongoing (e.g. the ACRE and OldWeather.org initiatives, Allan et al., 2011)392
but these could be expanded. The available observations can also be combined with data assim-393
ilation techniques to allow longer atmospheric reanalyses to be produced (Widmann et al., 2010;394
Compo et al., 2011; Matsikaris et al., 2015; Brohan et al., 2016). Additional seasonal proxy infor-395
mation would be of great value for informing this discussion, especially for winter (see Appendix)396
and for the tropics and Southern Hemisphere (e.g. Jones et al., 2016), although the temporal res-397
olution and continuity of proxies into the modern period is also a potential issue. Also note that398
a suitable pre-industrial period may be different for other climate variables, e.g. sea level, or for399
carbon cycle considerations.400
Two specific recommendations for future GCM-based analyses and simulations are: (i) to use401
blended observation-like estimates of global mean temperature when comparing observations and402
simulations, and (ii) use 1750 forcings to perform pre-industrial control simulations and to start403
19
historical transient simulations, rather than 1850. Adopting these recommendations would allow404
an ensemble of transient historical simulations to better quantify the role of natural variability and405
the impacts of the total radiative forcing changes since the pre-industrial period, especially the po-406
tentially long-term impact of the large volcanic eruptions in the early 1800s (Raible et al., 2016).407
We recognise, however, that this increases the computational demand in producing historical sim-408
ulations. In addition, increased usage of tracers (e.g. water stable isotopes) and proxy models409
within GCMs would allow more direct comparisons between simulations and proxy observations,410
including GCM simulations nudged to atmospheric reanalyses (e.g. Jouzel et al., 2000; LeGrande411
and Schmidt, 2009; Evans et al., 2013).412
Finally, these findings have a number of implications for policy-relevant issues. For example, the413
date at which future temperature thresholds are expected to be crossed may be shifted slightly ear-414
lier than estimated in IPCC AR5 (see Joshi et al., 2011; Kirtman et al., 2013; Hawkins and Sutton,415
2016). In addition, the cumulative emissions allowed to avoid reaching a particular temperature416
threshold (Meinshausen et al., 2009; Allen et al., 2009) may need to be reassessed, although any417
difference would likely be well within the current uncertainty ranges. Moving the baseline may418
also affect how historical responsibility for emissions needs to be accounted for (Knutti et al.,419
2015).420
More specifically, given the uncertainty in the global mean temperature change since pre-421
industrial, the UNFCCC might consider alternative equivalent baselines and limits to global tem-422
perature change. For example, “well below 2Cabove pre-industrial” might be translated to “well423
below XCabove 1986-2005”. Using a recent baseline is possibly more relevant for defining some424
impacts of climatic changes, with the value of X(and choice of baseline period) being decided by425
the UNFCCC. Given the uncertainty in defining the temperature change since pre-industrial, such426
a framing would allow a more precise assessment of when such levels might be reached in future,427
20
given our much improved recent observational coverage and availability of atmospheric reanalyses428
for the modern period (e.g. Dee et al., 2011; Simmons et al., 2016). It would also remove the need429
to precisely assess inherently uncertain changes since the pre-industrial period.430
APPENDIX431
Comparison with proxy reconstructions432
There are numerous efforts to reconstruct past climate using different proxies and archives which433
could be used to aid an assessment of change since the pre-industrial period. For temperature, these434
include ice cores, glaciers, tree rings, pollen, corals and sediment cores.435
For example, Leclercq and Oerlemans (2012) suggest a global land warming of 0.94 ±0.31C436
between 1830 and 2000 using glacier reconstructions, although the mid-1700s is around 0.25C437
warmer than 1830 in their estimates. Pollack and Smerdon (2004) suggest that global land temper-438
atures in the mid-1700s were around 0.65 0.90C below the year 2000 using borehole proxies.439
Mann et al. (2008) perform a multi-proxy analysis and report that global average temperature440
was around 0.3C below 1961-90 in the mid-1700s, with large uncertainties. This is equivalent441
to around 0.6C below 1986-2005, consistent with the recent PAGES2k global reconstruction442
(PAGES 2k Consortium et al., 2013) and this study.443
Overall, these proxy reconstruction estimates for pre-industrial temperature are consistent with444
the approaches adopted above, but the uncertainties are currently too large to make more precise445
statements. Defining a pre-industrial period (1720-1800) will hopefully provide a target for future446
reconstructions using the proxy data available. Certain long proxy series could also be used in447
Approach 3. However, it is important that such efforts focus on all seasons, as we next discuss.448
21
Seasonal effects in proxies, observations and simulations449
There are likely some seasonal differences in the rates of temperature change which are impor-450
tant to consider (e.g. Hegerl et al., 2011; Jones et al., 2014). For example, different proxies are451
sensitive to climate in certain seasons. In general, summer is more widely represented because452
many proxies rely on biological activity which tends to occur in the extended summer season.453
This is a potential issue for using proxies to reconstruct past temperatures if winter and summer454
change at different rates (Jones et al., 2003). In that case, the different seasonal proxies may not455
agree and/or produce biased estimates of an annual average. Some reconstructions (e.g. Van Enge-456
len et al., 2001; Luterbacher et al., 2004; Vinther et al., 2010) for Holland, Europe and Greenland457
respectively do show seasonal warming differences. However, the restricted availability of winter458
proxies limits the scope of such a comparison.459
To investigate how representative of annual mean changes the seasonal data is, we repeated the460
instrumental analysis (Approach 3) using extended seasons (April to September and October to461
March) for the regional data, whilst retaining the annual global data as the reference. Fig. 6a462
shows how the derived warming since the 1753-1800 period depends on the choice of season for463
the instrumental series - the extended winter season warms much faster than the extended summer464
season.465
However, if this seasonal difference in the rate of change over Europe was constant with time it466
should be scaled out. This suggests that there is: (i) a seasonal bias in the observed temperatures in467
certain periods (e.g. before standardised measurements) and/or (ii) a different seasonal response468
to different radiative forcings.469
For example, there is evidence that some historical observations may be biased, especially in470
summer, where warm biases due to non-optimal observation techniques in the past have been471
22
identified (Parker, 1994; B¨
ohm et al., 2010; Jones, 2016), which fits the pattern seen in Fig. 6a.472
Dobrovoln`
y et al. (2010) note that their documentary temperature data agrees best with their in-473
strumental data during winter, adding credence to this hypothesis. In addition, the cooling due to474
tropospheric aerosols in the 20th century may be seasonally dependent (Hunter et al., 1993; Krish-475
nan and Ramanathan, 2002), there is a trend in westerly wind characteristics in winter (Haarsma476
et al., 2013) and many of the observations are located in the northern extra-tropics and therefore477
influenced by Arctic amplification, which is observed and simulated to be larger in winter than in478
summer (Serreze et al., 2009; Pithan and Mauritsen, 2014).479
We can also examine whether this seasonal warming difference is present in the last millen-480
nium model simulations. Fig. 6b highlights that the CESM1 LME simulations do not show a481
strong global mean warming seasonal difference since the pre-industrial period, and only a very482
small seasonal effect when considering the central England location. The complex nature of these483
different seasonal features merits further analysis in a range of observations and simulations.484
Acknowledgments. We thank John Fasullo and Johann Jungclaus for providing the CESM1485
LME and MPI-ESM data respectively. EH is funded by the UK National Centre for At-486
mospheric Science (NCAS) and a Natural Environment Research Council (NERC) fellow-487
ship (Grant NE/I020792/1). TJO and GH were supported by the NERC (grant number488
NE/N006348/1, SMURPHS). ES was supported by the European Union Seventh Framework Pro-489
gramme (FP7/2007-2013) under the SPECS project (grant Agreement No. 308378) and by the490
UK-China Research and Innovation Partnership Fund through the Met Office Climate Science491
for Service Partnership (CSSP) China as part of the Newton Fund. AS and GH are supported492
by the ERC funded project TITAN (EC-320691). GH was further supported by NCAS and the493
23
Wolfson Foundation and the Royal Society as a Royal Society Wolfson Research Merit Award494
(WM130060) holder.495
References496
Abrams, N., H. McGregor, J. Tierney, M. Evans, N. McKay, D. Kaufman, and the PAGES con-497
sortium, 2016: Early onset of industrial-era warming across the oceans and continents. Nature,498
536, 411–418.499
Aldrin, M., M. Holden, P. Guttorp, R. B. Skeie, G. Myhre, and T. K. Berntsen, 2012: Bayesian500
estimation of climate sensitivity based on a simple climate model fitted to observations of501
hemispheric temperatures and global ocean heat content. Environmetrics,23 (3), 253–271, doi:502
10.1002/env.2140.503
Allan, R., P. Brohan, G. P. Compo, R. Stone, J. Luterbacher, and S. Bronnimann, 2011: The in-504
ternational atmospheric circulation reconstructions over the earth (ACRE) initiative. Bull. Amer.505
Meteor. Soc.,92, 1421, doi:10.1175/2011BAMS3218.1.506
Allen, M. R., D. J. Frame, C. Huntingford, C. D. Jones, J. A. Lowe, M. Meinshausen, and N. Mein-507
shausen, 2009: Warming caused by cumulative carbon emissions towards the trillionth tonne.508
Nature,458, 1163–1166.509
Bauska, T. K., F. Joos, A. C. Mix, R. Roth, J. Ahn, and E. J. Brook, 2015: Links between atmo-510
spheric carbon dioxide, the land carbon reservoir and climate over the past millennium. Nature511
Geoscience,8, 383–387, doi:10.1038/ngeo2422.512
B¨
ohm, R., P. D. Jones, J. Hiebl, D. Frank, M. Brunetti, and M. Maugeri, 2010: The early instru-513
mental warm-bias: a solution for long central european temperature series 1760–2007. Climatic514
Change,101, 41–67.515
24
Bradley, R. S., H. Wanner, and H. F. Diaz, 2016: The medieval quiet period. The Holocene, in516
press, doi:10.1177/0959683615622552.517
Briffa, K. R., P. D. Jones, F. H. Schweingruber, and T. J. Osborn, 1998: Influence of volcanic518
eruptions on northern hemisphere summer temperature over the past 600 years. Nature,393,519
450–455, doi:10.1038/30943.520
Brohan, P., G. P. Compo, S. Br¨
onnimann, R. J. Allan, R. Auchmann, Y. Brugnara, P. D. Sardesh-521
mukh, and J. S. Whitaker, 2016: The 1816 year without a summer in an atmospheric reanalysis.522
Climate of the Past Discussions,2016, 1–11, doi:10.5194/cp-2016-78.523
Carslaw, K. S., and Coauthors, 2013: Large contribution of natural aerosols to uncertainty in524
indirect forcing. Nature,503, 67–71, doi:10.1038/nature12674.525
Compo, G. P., and Coauthors, 2011: The twentieth century reanalysis project. Quarterly Journal526
of the Royal Meteorological Society,137, 1–28.527
Cowtan, K., and R. G. Way, 2014: Coverage bias in the hadcrut4 temperature series and its impact528
on recent temperature trends. QJRMS,140, 1935–1944, doi:10.1002/qj.2297.529
Cowtan, K., and Coauthors, 2015: Robust comparison of climate models with observations using530
blended land air and ocean sea surface temperatures. Geophysical Research Letters,42 (15),531
6526–6534.532
Crowley, T., and M. Unterman, 2013: Technical details concerning development of a 1200 yr533
proxy index for global volcanism. Earth System Science Data,5 (1), 187–197.534
Crowley, T. J., G. Zielinski, B. Vinther, R. Udisti, K. Kreutz, J. Cole-Dai, and E. Castellano, 2008:535
Volcanism and the little ice age. PAGES news,16 (2), 22–23.536
25
Dee, D. P., S. M. Uppala, A. J. Simmons, and et al., 2011: The ERA-Interim reanalysis:537
configuration and performance of the data assimilation system. QJRMS,137, 553–597, doi:538
10.1002/qj.828.539
Deser, C., M. A. Alexander, S.-P. Xie, and A. S. Phillips, 2010: Sea surface temperature vari-540
ability: Patterns and mechanisms. Annual Review of Marine Science,2 (1), 115–143, doi:541
10.1146/annurev-marine-120408-151453.542
Dobrovoln`
y, P., and Coauthors, 2010: Monthly, seasonal and annual temperature reconstructions543
for central europe derived from documentary evidence and instrumental records since ad 1500.544
Climatic Change,101 (1-2), 69–107.545
Evans, M., S. Tolwinski-Ward, D. Thompson, and K. Anchukaitis, 2013: Applications of proxy546
system modeling in high resolution paleoclimatology. Quaternary Science Reviews,76, 16 –547
28, doi:http://dx.doi.org/10.1016/j.quascirev.2013.05.024.548
Fern´
andez-Donado, L., and Coauthors, 2013: Large-scale temperature response to external forcing549
in simulations and reconstructions of the last millennium. Climate of the Past,9 (1), 393–421,550
doi:10.5194/cp-9-393-2013.551
Gao, C., A. Robock, and C. Ammann, 2008: Volcanic forcing of climate over the past 1500552
years: An improved ice core-based index for climate models. Journal of Geophysical Research:553
Atmospheres,113, doi:10.1029/2008JD010239, d23111.554
Gray, S. T., L. J. Graumlich, J. L. Betancourt, and G. T. Pederson, 2004: A tree-ring based recon-555
struction of the atlantic multidecadal oscillation since 1567 ad. Geophysical Research Letters,556
31 (12).557
26
Haarsma, R. J., F. Selten, and G. J. van Oldenborgh, 2013: Anthropogenic changes of the thermal558
and zonal flow structure over Western Europe and Eastern North Atlantic in CMIP3 and CMIP5559
models. Clim. Dyn.,41 (9-10), 2577–2588, doi:10.1007/s00382-013-1734-8.560
Hansen, J., R. Ruedy, M. Sato, and K. Lo, 2010: Global surface temperature change. Reviews of561
Geophysics,48 (4).562
Hansen, J., and Coauthors, 2005: Efficacy of climate forcings. Journal of Geophysical Research:563
Atmospheres,110, D18104, doi:10.1029/2005JD005776.564
Hawkins, E., and R. Sutton, 2016: Connecting climate model projections of global temperature565
change with the real world. Bulletin of the American Meteorological Society,97 (6), 963–980,566
doi:10.1175/BAMS-D-14-00154.1.567
Hegerl, G., J. Luterbacher, F. Gonzlez-Rouco, S. F. B. Tett, T. Crowley, and E. Xoplaki, 2011:568
Influence of human and natural forcing on european seasonal temperatures. Nature Geoscience,569
4, 99103, doi:10.1038/ngeo1057.570
Hegerl, G. C., T. J. Crowley, M. Allen, W. T. Hyde, H. N. Pollack, J. Smerdon, and E. Zorita,571
2007: Detection of human influence on a new, validated 1500-year temperature reconstruction.572
Journal of Climate,20 (4), 650–666, doi:10.1175/JCLI4011.1.573
Hunter, D. E., S. E. Schwartz, R. Wagener, and C. M. Benkovitz, 1993: Seasonal, latitudinal,574
and secular variations in temperature trend: Evidence for influence of anthropogenic sulfate.575
Geophysical Research Letters,20 (22), 2455–2458, doi:10.1029/93GL02808.576
IISD, 2013: Earth negotiations bulletin: Summary of the 12th session of working group i of the577
intergovernmental panel on climate change (ipcc) and thirty-sixth session of the ipcc, URL578
http://www.iisd.ca/vol12/enb12581e.html.579
27
IPCC, 2013: Summary for policymakers. Climate Change 2013: The Physical Science Basis. Con-580
tribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on581
Climate Change, T. Stocker, D. Qin, G.-K. Plattner, M. Tignor, S. Allen, J. Boschung, A. Nauels,582
Y. Xia, V. Bex, and P. Midgley, Eds., Cambridge University Press, Cambridge, United Kingdom583
and New York, NY, USA, book section SPM, 1˘
201330, doi:10.1017/CBO9781107415324.004,584
URL www.climatechange2013.org.585
Jones, J., S. Gille, H. Goosse, N. Abram, P. Canziani, D. Charman, and co-authors, 2016: Assess-586
ing recent trends in high-latitude southern hemisphere surface climate. Nature Climate Change,587
in press.588
Jones, P., 2016: The reliability of global and hemispheric surface temperature records. Advances589
in Atmospheric Sciences,33 (3), 269–282.590
Jones, P., and K. Briffa, 2006: Unusual climate in northwest europe during the period 1730 to591
1745 based on instrumental and documentary data. Climatic Change,79 (3-4), 361–379.592
Jones, P. D., K. R. Briffa, and T. J. Osborn, 2003: Changes in the northern hemisphere annual593
cycle: Implications for paleoclimatology? Journal of Geophysical Research: Atmospheres,594
108, 4588, doi:10.1029/2003JD003695.595
Jones, P. D., C. Harpham, and B. M. Vinther, 2014: Winter-responding proxy temperature recon-596
structions and the north atlantic oscillation. Journal of Geophysical Research: Atmospheres,597
119 (11), 6497–6505, doi:10.1002/2014JD021561.598
Joshi, M., E. Hawkins, R. Sutton, J. Lowe, and D. Frame, 2011: Projections of when temperature599
change will exceed 2C above pre-industrial levels. Nature Climate Change,1, 407–412, doi:600
10.1038/nclimate1261.601
28
Jouzel, J., G. Hoffmann, R. Koster, and V. Masson, 2000: Water isotopes in precipitation:602
data/model comparison for present-day and past climates. Quaternary Science Reviews,19 (1),603
363–379.604
Jungclaus, J. H., K. Lohmann, and D. Zanchettin, 2014: Enhanced 20th-century heat transfer to605
the arctic simulated in the context of climate variations over the last millennium. Climate of the606
Past,10 (6), 2201–2213, doi:10.5194/cp-10-2201-2014.607
Kaplan, A., M. Cane, Y. Kushnir, A. Clement, M. Blumenthal, and B. Rajagopalan, 1998: Anal-608
yses of global sea surface temperature 1856-1991. Journal of Geophysical Research,103,609
18 567–18 589.610
Karl, T. R., and Coauthors, 2015: Possible artifacts of data biases in the recent global surface611
warming hiatus. Science,348 (6242), 1469–1472.612
Kaufman, D. S., and Coauthors, 2009: Recent warming reverses long-term arctic cooling. Science,613
325 (5945), 1236–1239, doi:10.1126/science.1173983.614
Keeling, C. D., S. C. Piper, R. B. Bacastow, M. Wahlen, T. P. Whorf, M. Heimann, and H. A. Mei-615
jer, 2001: Exchanges of atmospheric CO2 and 13CO2 with the terrestrial biosphere and oceans616
from 1978 to 2000. I. Global aspects. SIO Reference Series, No. 01-06, Scripps Institution of617
Oceanography, San Diego.618
Kennedy, J. J., N. A. Rayner, R. O. Smith, D. E. Parker, and M. Saunby, 2011: Reassessing biases619
and other uncertainties in sea surface temperature observations measured in situ since 1850: 1.620
measurement and sampling uncertainties. Journal of Geophysical Research: Atmospheres,116,621
doi:10.1029/2010JD015218, d14103.622
29
Kirtman, B., and Coauthors, 2013: Near-term climate change: Projections and predictability. Cli-623
mate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth624
Assessment Report of the Intergovernmental Panel on Climate Change, T. Stocker, D. Qin, G.-625
K. Plattner, M. Tignor, S. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, and P. Midgley, Eds.,626
Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, chapter 11,627
953–1028, doi:10.1017/CBO9781107415324.023.628
Knutti, R., J. Rogelj, J. Sedlacek, and E. M. Fischer, 2015: A scientific critique of the two-degree629
climate change target. Nature Geoscience,9, 13–18, doi:10.1038/ngeo2595.630
Kopp, R. E., and Coauthors, 2016: Temperature-driven global sea-level variability in the common631
era. Proceedings of the National Academy of Sciences, in press, doi:10.1073/pnas.1517056113.632
Krishnan, R., and V. Ramanathan, 2002: Evidence of surface cooling from absorbing aerosols.633
Geophysical Research Letters,29 (9), 541–544, doi:10.1029/2002GL014687.634
Lean, J. L., and D. H. Rind, 2009: How will Earth’s surface temperature change in future decades?635
Geophys. Res. Lett.,36, L15 708, doi:10.1029/2009GL038932.636
Leclercq, P. W., and J. Oerlemans, 2012: Global and hemispheric temperature reconstruction from637
glacier length fluctuations. Climate Dynamics,38 (5-6), 1065–1079.638
LeGrande, A., and G. Schmidt, 2009: Sources of holocene variability of oxygen isotopes in pale-639
oclimate archives. Climate of the Past,5 (3), 441–455.640
Lockwood, M., M. J. Owens, and L. Barnard, 2014: Centennial variations in sunspot number,641
open solar flux, and streamer belt width: 2. comparison with the geomagnetic data. Journal642
of Geophysical Research: Space Physics,119 (7), 5183–5192, doi:10.1002/2014JA019972,643
2014JA019972.644
30
Luterbacher, J., D. Dietrich, E. Xoplaki, M. Grosjean, and H. Wanner, 2004: European seasonal645
and annual temperature variability, trends, and extremes since 1500. Science,303 (5663), 1499–646
1503.647
MacDonald, G. M., and R. A. Case, 2005: Variations in the pacific decadal oscillation over the648
past millennium. Geophysical Research Letters,32 (8).649
MacFarling Meure, C., D. Etheridge, C. Trudinger, P. Steele, R. Langenfelds, T. Van Ommen,650
A. Smith, and J. Elkins, 2006: Law dome co2, ch4 and n2o ice core records extended to 2000651
years bp. Geophysical Research Letters,33 (14).652
Manley, G., 1974: Central england temperatures: monthly means 1659 to 1973. Quarterly Journal653
of the Royal Meteorological Society,100 (425), 389–405.654
Mann, M., 2014: Earth will cross the climate danger threshold by 2036. Scientific Amer-655
ican, http://www.scientificamerican.com/article/earth-will-cross-the-climate-danger-threshold-656
by-2036/.657
Mann, M. E., Z. Zhang, M. K. Hughes, R. S. Bradley, S. K. Miller, S. Rutherford, and F. Ni, 2008:658
Proxy-based reconstructions of hemispheric and global surface temperature variations over the659
past two millennia. Proceedings of the National Academy of Sciences,105 (36), 13 252–13 257.660
Mann, M. E., and Coauthors, 2009: Global signatures and dynamical origins of the little ice age661
and medieval climate anomaly. Science,326, 1256–1260, doi:10.1126/science.1177303.662
Marenco, A., H. Gouget, P. Ndlec, J.-P. Pags, and F. Karcher, 1994: Evidence of a long-663
term increase in tropospheric ozone from pic du midi data series: Consequences: Positive664
radiative forcing. Journal of Geophysical Research: Atmospheres,99, 16 617–16 632, doi:665
10.1029/94JD00021.666
31
Matsikaris, A., M. Widmann, and J. Jungclaus, 2015: Assimilating continental mean temperatures667
to reconstruct the climate of the late pre-industrial period. Climate Dynamics, in press, doi:668
10.1007/s00382-015-2785-9.669
McGregor, H. V., and Coauthors, 2015: Robust global ocean cooling trend for the pre-industrial670
common era. Nature Geoscience,8, 671 – 677, doi:10.1038/ngeo2510.671
Meinshausen, M., N. Meinshausen, W. Hare, S. C. Raper, K. Frieler, R. Knutti, D. J. Frame, and672
M. R. Allen, 2009: Greenhouse-gas emission targets for limiting global warming to 2 c. Nature,673
458 (7242), 1158–1162.674
Meinshausen, M., and Coauthors, 2011: The rcp greenhouse gas concentrations and their exten-675
sions from 1765 to 2300. Climatic Change,109 (1-2), 213–241.676
Morice, C. P., J. J. Kennedy, N. A. Rayner, and P. D. Jones, 2012: Quantifying uncertainties677
in global and regional temperature change using an ensemble of observational estimates: The678
HadCRUT4 data set. J. Geophy. Res. Atmos.,117, D08101, doi:10.1029/2011JD017187.679
Osborn, T. J., S. C. B. Raper, and K. R. Briffa, 2006: Simulated climate change during the last680
1,000years: comparing the echo-g general circulation model with the magicc simple climate681
model. Climate Dynamics,27 (2), 185–197, doi:10.1007/s00382-006-0129-5.682
Otto, F. E., D. J. Frame, A. Otto, and M. R. Allen, 2015: Embracing uncertainty in climate change683
policy. Nature Climate Change,5, 917920, doi:10.1038/nclimate2716.684
Otto-Bliesner, B. L., and Coauthors, 2016: Climate variability and change since 850 ce: An en-685
semble approach with the community earth system model (cesm). Bulletin of the American686
Meteorological Society, in press.687
32
PAGES 2k Consortium, and Coauthors, 2013: Continental-scale temperature variability during the688
past two millennia. Nature Geoscience,6 (5), 339, doi:10.1038/ngeo1797.689
Parker, D., 1994: Effects of changing exposure of thermometers at land stations. International690
Journal of Climatology,14 (1), 1–31.691
Parker, D. E., 2010: Uncertainties in early central england temperatures. International Journal of692
Climatology,30 (8), 1105–1113, doi:10.1002/joc.1967.693
Parker, D. E., T. P. Legg, and C. K. Folland, 1992: A new daily central england temperature series,694
1772–1991. International Journal of Climatology,12 (4), 317–342.695
Pithan, F., and T. Mauritsen, 2014: Arctic amplification dominated by temperature feedbacks in696
contemporary climate models. Nature Geoscience,7, 181–184, doi:10.1038/ngeo2071.697
Pollack, H. N., and J. E. Smerdon, 2004: Borehole climate reconstructions: Spatial structure and698
hemispheric averages. Journal of Geophysical Research: Atmospheres (1984–2012),109 (D11).699
Raible, C. C., and Coauthors, 2016: Tambora 1815 as a test case for high impact volcanic erup-700
tions: Earth system effects. Wiley Interdisciplinary Reviews: Climate Change, in press, doi:701
10.1002/wcc.407.702
Richardson, M., K. Cowtan, E. Hawkins, and M. Stolpe, 2016: Reconciled climate response es-703
timates from climate models and the energy budget of earth. Nature Climate Change, in press,704
doi:10.1038/nclimate3066.705
Rohde, R., and Coauthors, 2013: A new estimate of the average earth surface land temperature706
spanning 1753 to 2011. Geoinfor. Geostat.,1, doi:10.4172/2327-4581.1000101.707
Ruddiman, W. F., 2003: The anthropogenic greenhouse era began thousands of years ago. Climatic708
Change,61 (3), 261–293, doi:10.1023/B:CLIM.0000004577.17928.fa.709
33
Ruddiman, W. F., and Coauthors, 2016: Late holocene climate: Natural or anthropogenic? Re-710
views of Geophysics,54 (1), 93–118, doi:10.1002/2015RG000503.711
Schmidt, G., and Coauthors, 2012: Climate forcing reconstructions for use in pmip simulations of712
the last millennium (v1. 1). Geoscientific Model Development,5, 185–191.713
Schmidt, G. A., and Coauthors, 2014: Configuration and assessment of the giss modele2 con-714
tributions to the cmip5 archive. Journal of Advances in Modeling Earth Systems,6, 141–184,715
doi:10.1002/2013MS000265.716
Schurer, A. P., G. C. Hegerl, M. E. Mann, S. F. B. Tett, and S. J. Phipps, 2013: Separating forced717
from chaotic climate variability over the past millennium. Journal of Climate,26, 6954–6973,718
doi:10.1175/JCLI-D-12-00826.1.719
Serreze, M., A. Barrett, J. Stroeve, D. Kindig, and M. Holland, 2009: The emergence of surface-720
based arctic amplification. The Cryosphere,3 (1), 11–19.721
Shindell, D. T., 2014: Inhomogeneous forcing and transient climate sensitivity. Nature Climate722
Change,4 (4), 274–277.723
Sigl, M., and Coauthors, 2013: A new bipolar ice core record of volcanism from wais divide724
and neem and implications for climate forcing of the last 2000years. Journal of Geophysical725
Research: Atmospheres,118 (3), 1151–1169, doi:10.1029/2012JD018603.726
Simmons, A. J., P. Berrisford, D. P. Dee, H. Hersbach, S. Hirahara, and J.-N. Thpaut, 2016: A727
reassessment of temperature variations and trends from global reanalyses and monthly surface728
climatological datasets. Quarterly Journal of the Royal Meteorological Society,in press, doi:729
10.1002/qj.2949.730
34
Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K. Averyt, M. M. B. Tignor, and H. L.731
Miller, Eds., 2007: Climate Change 2007: The Physical Science Basis. Contribution of Working732
Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.733
Cambridge University Press, Cambridge, UK.734
Steinhilber, F., J. Beer, and C. Fr¨
ohlich, 2009: Total solar irradiance during the holocene. Geo-735
physical Research Letters,36 (19), doi:10.1029/2009GL040142.736
Stevens, B., 2013: Aerosols: Uncertain then, irrelevant now. Nature,503, 47–48, doi:10.1038/737
503047a.738
Stoffel, M., and Coauthors, 2015: Estimates of volcanic-induced cooling in the northern hemi-739
sphere over the past 1,500 years. Nature Geoscience,8, 784 – 788, doi:10.1038/ngeo2526.740
Suckling, E., E. Hawkins, J. Eden, and G. J. van Oldenborgh, 2016: An empirical model for741
probabilistic decadal prediction: global attribution and regional hindcasts. Climate Dynamics,742
in press, doi:10.1007/s00382-016-3255-8.743
Sutton, R., E. Suckling, and E. Hawkins, 2015: What does global temperature tell us about local744
climate? Phil. Trans. A,373, 20140 426, doi:10.1098/rsta.2014.0426.745
Thordarson, T., and S. Self, 2003: Atmospheric and environmental effects of the 1783–1784 laki746
eruption: a review and reassessment. Journal of Geophysical Research: Atmospheres,108 (D1),747
1–29.748
Usoskin, I. G., and Coauthors, 2015: The maunder minimum (1645–1715) was indeed a grand749
minimum: A reassessment of multiple datasets. Astronomy & Astrophysics,581, A95.750
Van Engelen, A., J. Buisman, and F. Ijnsen, 2001: A millennium of weather, winds and water in751
the low countries. Springer.752
35
Van Engelen, A., and J. Nellestijn, 1990: Monthly, seasonal, and annual means of the air temper-753
ature in tenths of centigrade in de bilt, netherlands, 1706–1990. KNMI, De Bilt.754
Vinther, B. M., P. Jones, K. Briffa, H. Clausen, K. Andersen, D. Dahl-Jensen, and S. Johnsen,755
2010: Climatic signals in multiple highly resolved stable isotope records from greenland. Qua-756
ternary Science Reviews,29 (3), 522–538.757
Widmann, M., H. Goosse, G. van der Schrier, R. Schnur, and J. Barkmeijer, 2010: Using data as-758
similation to study extratropical northern hemisphere climate over the last millennium. Climate759
of the Past,6 (5), 627–644, doi:10.5194/cp-6-627-2010.760
Zanchettin, D., C. Timmreck, O. Bothe, S. J. Lorenz, G. Hegerl, H.-F. Graf, J. Luterbacher, and761
J. H. Jungclaus, 2013: Delayed winter warming: A robust decadal response to strong tropical762
volcanic eruptions? Geophys. Res. Lett.,40 (1), 204–209, doi:10.1029/2012GL054403.763
36
LIST OF FIGURES764
Fig. 1. Historical natural forcings and greenhouse gas variations. Top left: annual sunspot num-765
ber since 1612, with the Maunder Minimum and Dalton Minimum indicated (Lockwood766
et al., 2014). Top right: estimated global volcanic aerosol optical depth (Crowley and Un-767
terman, 2013). Bottom row: the Law Dome greenhouse gas data (MacFarling Meure et al.768
(2006), black) for carbon dioxide (left) and methane (right), along with the annual means769
from Mauna Loa (Keeling et al. (2001), blue) and PMIP3 assumed values (Schmidt et al.770
(2012), red). Note there is a 16ppb offset applied to the smoothed Law Dome methane con-771
centrations to produce a global mean as used by PMIP3 to account for the interhemispheric772
gradient. The 1720-1800 period is denoted by the grey shaded region in all panels. . . . . 38773
Fig. 2. Top panel: estimating global pre-industrial temperature using scaled radiative forcings774
(pink), using HadCRUT4 (black) as the reference. The grey shading represents the uncer-775
tainty in the regression. Estimated global temperature anomalies for 1765-1800 are given for776
all five global temperature datasets (right hand side, as labelled). Bottom panel: simulated777
global temperature anomalies in the Last Millennium Ensembles (LMEs) and estimates for778
the change since 1720-1800 for the range of ensemble members of CESM (blue), GISS779
(green) and MPI-ESM (orange). In both panels the blue horizontal bars indicate the period780
used for averaging. The 1986-2005 reference period is represented by the black dashed line. . 39781
Fig. 3. Estimating global pre-industrial temperature using scaled annual mean observations for CET782
scaled to HadCRUT4 (top) and BEST-Land scaled to BEST-GL (bottom), relative to 1986-783
2005 (dashed black). The dark grey shading (hardly visible) represents the uncertainty in the784
regressions and the light grey shading the uncertainty in the observations. The sets of five785
error bars on the right hand side use the different global temperature datasets, with the same786
ordering as in the top panel of Fig. 2, for the two different averaging periods as labelled.787
Note the vertical scale is different from Fig. 2. . . . . . . . . . . . . . . 40788
Fig. 4. Summarising the evidence for annual mean global temperature change from pre-industrial789
until 1986-2005 using each dataset. The horizontal bars represent the 5-95% uncertainty790
ranges for the different sources of evidence. Results for the radiative forcing approach are791
shown averaged over 1765-1800, and for 1750. The top row in the instrumental observations792
section shows the observed change since 1850-1900 (where available). For the instrumental793
data the longest timeseries during the pre-industrial period are used: CET and De Bilt (1720-794
1800), BEST-Land (1753-1800) and Central Europe (1760-1800). The light grey shading795
shows the assessed likely range and the dark grey line indicates the IPCC AR5 assessment796
(0.61C, Kirtman et al., 2013). . . . . . . . . . . . . . . . . . . 41797
Fig. 5. Global mean temperature relative to pre-industrial in six datasets, using the likely lower798
bound (0.60C) for warming from pre-industrial to 1986-2005. The change in the ERA-799
Interim reanalysis (Dee et al., 2011) relative to 1986-2005 is included with the five global800
temperature datasets discussed. The 1996-2015 period is 0.16 0.19C warmer than 1986-801
2005..........................42802
Fig. 6. Seasonal differences in warming rates. (a) Derived scaled warming from 1753-1800 to 1986-803
2005 (using Approach 3) for annual means (black) and for the extended seasons (April to804
September - AMJJAS, red, and October to March - ONDJFM, blue) for the different regional805
timeseries, all using annual mean HadCRUT4 as the reference dataset. (b) Seasonal warming806
derived from the CESM1 LME simulations for the global mean (crosses, with black lines807
linking the same ensemble members in each season) and for the ensemble mean of simulated808
CET(circles).......................43809
37
Volcanic aerosol optical depth
Crowley & Unterman (2013)
1560 1640 1720 1800 1880 1960
0
0.1
0.2
0.3
0.4
Sunspot number
Maunder
Minimum
Dalton
Min.
Lockwood et al. (2014)
1560 1640 1720 1800 1880 1960
0
40
80
120
160
200
Year
CO2 concentration [ppm]
Law Dome
Mauna Loa
PMIP3
1560 1640 1720 1800 1880 1960
280
300
320
340
360
380
400
Solar variations, volcanic eruptions & atmospheric greenhouse gas concentrations
Year
CH4 concentration [ppm]
Law Dome
Mauna Loa
PMIP3
1560 1640 1720 1800 1880 1960
0.6
0.8
1
1.2
1.4
1.6
1.8
FIG. 1. Historical natural forcings and greenhouse gas variations. Top left: annual sunspot number since
1612, with the Maunder Minimum and Dalton Minimum indicated (Lockwood et al., 2014). Top right: estimated
global volcanic aerosol optical depth (Crowley and Unterman, 2013). Bottom row: the Law Dome greenhouse
gas data (MacFarling Meure et al. (2006), black) for carbon dioxide (left) and methane (right), along with the
annual means from Mauna Loa (Keeling et al. (2001), blue) and PMIP3 assumed values (Schmidt et al. (2012),
red). Note there is a 16ppb offset applied to the smoothed Law Dome methane concentrations to produce a
global mean as used by PMIP3 to account for the interhemispheric gradient. The 1720-1800 period is denoted
by the grey shaded region in all panels.
810
811
812
813
814
815
816
817
38
HadCRUT4.5
Cowtan & Way
NOAA GlobalTemp
GISTEMP
BEST−GL
Estimating pre−industrial global temperatures using radiative forcings & simulations
Temperature anomaly [K]
1986−2005
1700 1750 1800 1850 1900 1950 2000
−1.4
−1.2
−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4 HadCRUT4.5
Scaled CMIP5 forcings
CESM
GISS
MPI−ESM
Temperature anomaly [K]
1700 1750 1800 1850 1900 1950 2000
−1.4
−1.2
−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4 HadCRUT4.5
FIG. 2. Top panel: estimating global pre-industrial temperature using scaled radiative forcings (pink), using
HadCRUT4 (black) as the reference. The grey shading represents the uncertainty in the regression. Estimated
global temperature anomalies for 1765-1800 are given for all five global temperature datasets (right hand side, as
labelled). Bottom panel: simulated global temperature anomalies in the Last Millennium Ensembles (LMEs) and
estimates for the change since 1720-1800 for the range of ensemble members of CESM (blue), GISS (green)
and MPI-ESM (orange). In both panels the blue horizontal bars indicate the period used for averaging. The
1986-2005 reference period is represented by the black dashed line.
818
819
820
821
822
823
824
39
Estimating pre−industrial global temperatures using scaled observations
Temperature anomaly [K]
(1765−1800)
(1720−1800)
1700 1750 1800 1850 1900 1950 2000
−1.6
−1.2
−0.8
−0.4
0
0.4
0.8 HadCRUT4.5
Scaled HadCET
Temperature anomaly [K]
(1765−1800)
1700 1750 1800 1850 1900 1950 2000
−1.6
−1.2
−0.8
−0.4
0
0.4
0.8 BEST−GL
Scaled BEST−Land
FIG. 3. Estimating global pre-industrial temperature using scaled annual mean observations for CET scaled to
HadCRUT4 (top) and BEST-Land scaled to BEST-GL (bottom), relative to 1986-2005 (dashed black). The dark
grey shading (hardly visible) represents the uncertainty in the regressions and the light grey shading the uncer-
tainty in the observations. The sets of five error bars on the right hand side use the different global temperature
datasets, with the same ordering as in the top panel of Fig. 2, for the two different averaging periods as labelled.
Note the vertical scale is different from Fig. 2.
825
826
827
828
829
830
40
HadCRUT4.5 1765−1800
1750
Cowtan & Way
1765−1800
1750
NOAA GlobalTemp
1765−1800
1750
GISTEMP
1765−1800
1750
BEST−GL
1765−1800
1750
AR5
Global (1850−1900)
Central England
De Bilt
BEST−Land
Central Europe
(1720−1800)
(1720−1800)
(1753−1800)
(1760−1800)
Warming from pre−industrial to 1986−2005 [K]
Instrumental data Forcings
0.4 0.5 0.6 0.7 0.8 0.9
FIG. 4. Summarising the evidence for annual mean global temperature change from pre-industrial until 1986-
2005 using each dataset. The horizontal bars represent the 5-95% uncertainty ranges for the different sources of
evidence. Results for the radiative forcing approach are shown averaged over 1765-1800, and for 1750. The top
row in the instrumental observations section shows the observed change since 1850-1900 (where available). For
the instrumental data the longest timeseries during the pre-industrial period are used: CET and De Bilt (1720-
1800), BEST-Land (1753-1800) and Central Europe (1760-1800). The light grey shading shows the assessed
likely range and the dark grey line indicates the IPCC AR5 assessment (0.61C, Kirtman et al., 2013).
831
832
833
834
835
836
837
41
Global temperature change since pre−industrial (likely lower bound)
Temperature anomaly [
oC]
Year
1980 1985 1990 1995 2000 2005 2010 2015
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
HadCRUT4.5
Cowtan & Way
NOAA GlobalTemp
GISTEMP
BEST−GL
ERA−Interim
FIG. 5. Global mean temperature relative to pre-industrial in six datasets, using the likely lower bound
(0.60C) for warming from pre-industrial to 1986-2005. The change in the ERA-Interim reanalysis (Dee et al.,
2011) relative to 1986-2005 is included with the five global temperature datasets discussed. The 1996-2015
period is 0.16 0.19C warmer than 1986-2005.
838
839
840
841
42
CET De Bilt BEST−Land C.Europe
0.2
0.4
0.6
0.8
1
1.2 ANNUAL AMJJAS ONDJFM
(a) Scaled instrumental observation estimates (1753−1800)
Change from pre−industrial [K]
ANNUAL AMJJAS ONDJFM
0.2
0.4
0.6
0.8
1
1.2
(b) CESM1 simulations
CESM1 Global
CESM1 CET
(ensemble mean)
FIG. 6. Seasonal differences in warming rates. (a) Derived scaled warming from 1753-1800 to 1986-2005
(using Approach 3) for annual means (black) and for the extended seasons (April to September - AMJJAS, red,
and October to March - ONDJFM, blue) for the different regional timeseries, all using annual mean HadCRUT4
as the reference dataset. (b) Seasonal warming derived from the CESM1 LME simulations for the global mean
(crosses, with black lines linking the same ensemble members in each season) and for the ensemble mean of
simulated CET (circles).
842
843
844
845
846
847
43
