1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

A New Estimate of the Average Earth Surface Land Temperature Spanning 1753 to 2011
Submitted to JGR The Third Santa Fe Conference on Global and Regional Climate Change Manuscript # 2012JD018202

Robert Rohde1, Richard A. Muller1,2,3, Robert Jacobsen2,3, Elizabeth Muller1, Saul Perlmutter2,3, Arthur Rosenfeld2,3, Jonathan Wurtele2,3,Donald Groom3, Charlotte Wickham4

Corresponding Author Address: Richard A. Muller Berkeley Earth Project 2831 Garber St. Berkeley CA 94705 email: [email protected]

Berkeley Earth Land Temperature Project, Novim Group; 2University of California,

Berkeley; 3Lawrence Berkeley National Laboratory; 4Oregon State University

22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42

Abstract:

We report an estimate of the Earths average land surface temperature for the period 1753 to 2011. To address issues of potential station selection bias, we used larger sampling of stations than had prior studies. For the period post 1880, our estimate is similar to those previously reported by other groups, although we report smaller error uncertainties. The land temperature rise from the 1950s decade to the 2000s decade is 0.87 0.05 C (95% confidence). Both maximum and minimum daily temperatures have increased during the last century. Diurnal variations decreased from 1900 to 1987, and then increased; this increase is significant but not understood. The period of 1753 to 1850 is marked by sudden drops in land surface temperature that are coincident with known volcanism; the response function is approximately 1.5 0.5 C per 100 Tg of atmospheric sulfate. This volcanism, combined with a simple proxy for anthropogenic effects (logarithm of the CO2 concentration), can account for much of the variation in the land surface temperature record; the fit is not improved by the addition of a solar forcing term. Thus, for this very simple model, solar forcing does not appear to contribute to the observed global warming of the past 250 years; the entire change can be accounted for by a sum of volcanism and anthropogenic proxies. The residual variations include interannual and multi-decadal variability very similar to that of the Atlantic Multidecadal Oscillation (AMO).

43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

Introduction and Methods

The average Earth surface temperature is a key indicator of climate change. Previous estimates have been reported by three groups: NOAA, NASA GISS, and HadCRU; the analyses of these groups share many common features [Hansen and Lebedeff, 1987; Hansen et al., 1999; Hansen et al., 2010; Jones et al., 1986; Jones and Moberg, 2003; Brohan et al., 2006; Smith and Reynolds, 2005; Smith et al., 2008]. According to the summary provided by the by the Intergovernmental Panel on Climate Change (IPCC), the mean global surface temperature (including land and oceans) has increased 0.64 0.13 C from 1956 to 2005 at 95% confidence [Trenberth et al., 2007]. The IPCC did not provide a similar review of land-only mean temperatures; however, the three groups ranged from 0.81 to 0.93 C when estimating the increase in land temperatures during the 2000s decade relative to the 1950s decade. As described below, we estimate this change as 0.869 0.048 C (95% confidence).

In this paper we present results for the Earths land surface temperature only, based on analysis of monthly averages at each station. We gathered and merged monthly and daily thermometer measurements from 14 databases to arrive at a collection of 14.4 million mean monthly temperature observations from 44,455 sites. These data have now been posted online in a uniform format at www.BerkeleyEarth.org, along with a description of the merging method. For stations that report only daily data (and not their own monthly average) we performed the average. When we removed very short records (less than 1 year) and records from sites with missing or highly uncertain location metadata we were left with 36,866 stations that we used in our analysis. 3

67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 To perform the average, the surface of the Earth was divided into 15984 elements of equal area and weighted by the percentage of land at each spot; 5326 of them had > 0.1 land. 4 In order to minimize statistical uncertainties, we developed a computer program we call BerkeleyAverage that could take advantage of all 179,928 record fragments. The Matlab program that implements BerkeleyAverage is available on www.BerkeleyEarth.org. Our analysis approach differed from that of the previous groups in several ways. Rather than adjust (homogenize) individual records for known and presumed discontinuities (e.g. from instrument changes and station moves), we split the records at such points, creating essentially two records from one. This procedure was completely automated to reduce human bias; we call this approach the scalpel. The detection of such breakpoints followed procedures similar to those used by existing groups, but the traditional adjustment step was omitted in favor of simply dividing the time series into two pieces at any apparent breakpoints. We also split records when there was a gap in record continuity greater than 1 year in duration, and at times when changes in station location or time of observation were documented. This approach is possible because our reconstruction method depends less on long duration samples than do the methods applied by prior groups. The 36,866 records were split, on average, 3.9 times to create 179,928 record fragments. When we detected other problems (e.g. undocumented changes from Celsius to Fahrenheit) we flagged the changes; the raw uncorrected data are available online in a separate file. Seasonality was eliminated from each record by removing cycles with 1year periods and higher harmonics; the unadjusted data are also available on the website.

91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114

For each month, BerkeleyAverage creates an estimated temperature field for the entire land surface of the globe using Kriging to interpolate the available temperature data. The Kriging process, also known as Gaussian Process Regression, is the best linear unbiased predictor of the underlying field provided the temperature fluctuations at each site are approximately normally distributed and the correlation structure between sites can be estimated. [Krige, 1958; Cresse 1990]. This method uses the correlations between temperature stations as a function of distance to estimate the temperature at any intermediate point and naturally avoids the bias that might come from overweighting highly clustered stations (e.g. in the United States and Europe). The average temperature Tavg is then the simple average of the interpolated land temperature field over land during each month.

To obtain the best estimate for the Earth surface land temperature, BerkeleyAverage performs a process analogous to iterative least-squares minimization. The initial variables are (1) an offset variable (baseline temperature) for each of the 179,928 temperature time series, referenced to values that take into account latitude and altitude, and (2) the set of Tavg numbers (one variable for each month) that will represent our land global average; the values of these variables necessary to minimize the least-squares differences are calculated exactly using matrix inversion. Berkeley Average then compares the interpolated temperature field obtained from Kriging to the actual observations to detect observations and stations trends that have large departures from that expected based on the regional average field. A weight adjustment is applied to stations to reduce the effect of the largest outliers, the variables are recomputed, and then the procedure is iterated. The weighting procedure was applied to reduce the effects of individual measurements that 5

115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135

were erroneously recorded (e.g. typographic errors) as well as individual stations that have spurious trends due to effects such as the urban heat island, poor station siting, and undocumented changes in stations or instruments that were not caught by our scalpel method.

Statistical uncertainties were estimated by randomly subdividing the 179,928 scalpeled stations into 8 smaller sets, calculating BerkeleyAverage for each of those, and then comparing the results using the jackknife statistical method [Tukey, 1958; Miller, 1974]. Spatial sampling uncertainties were estimated by simulating poorly sampled periods (e.g. 1800 to 1850) with modern data (1960 to 2010) for which the Earth coverage was better than 97% complete, and measuring the departure from the full site average when using only the limited spatial regions available at early times. This empirical approach implicitly assumes that the spatial relationships between different climate regions remain largely unchanged, and it underestimates the uncertainties if that assumption is false. This could happen, for example, in the period 1750 to 1850 when our evidence shows a strong influence from large volcanic eruptions, a phenomenon for which there are only weak analogs in the 20th century. Thus although we report global land surface results, it should not be forgotten that in the earliest periods that we cover (especially prior to 1850) that sampling is poor, and our results accurate only to the extent that the spatial structure of temperature does not change significantly with time.

136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155

Temperature Change

The 1-year and 10-year running averages of our estimated Earth land surface average temperature is shown in Figure 1, along with both the 1- and 2- standard-deviation uncertainties that combine those from statistical and spatial sampling effects. The land temperature estimates of the three prior groups are shown for comparison. Although some of those estimates lie outside of our uncertainty bands, they all fall within the uncertainty bands reported by those prior groups; thus there is no statistically significant disagreement. Note that we have extended the estimates back to 1753, although with increased uncertainties. In these early years Earth coverage was minimal; as discussed earlier, the uncertainties were determined by tests using modern data (post 1950) to see the accuracy we obtain for known global land changes when the program is restricted to sparse coverage.

We find that the global land mean temperature (comparing the average of the period 20002010 with that of 1950-1960) has increased by 0.87 0.05 C (95% confidence). This change is consistent with the global land-surface warming results previously reported, but with reduced uncertainty.

156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 The early part of the record infers, as best as is possible, the fluctuations in global landonly temperature from a sparse network of primarily European and North American observations. This is possible, albeit with the relatively large uncertainties, because the European and North American annual average anomalies are observed to remain within 0.5 C of the global land average 95% of the time during the 20th century. The BerkeleyAverage procedure allows us to use the sparse network of observations from the very longest monitoring stations (10, 25, 46, 101, and 186 sites in the years 1755, 1775, 1800, 1825, and 1850 respectfully) to place limited bounds on the early average. These limits imply an Earth that was colder than today during nearly all of the period since 1755. 8 Figure 1. Land temperature with 1- and 10-year running averages. The shaded regions are the one- and two-standard deviation uncertainties calculated including both statistical and spatial sampling errors. Prior land results from the other groups are also plotted. The NASA GISS record had a land mask applied; the HadCRU curve is the simple land average, not the hemispheric-weighted one.

173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 There are large differences among the surface temperatures of the hemispheres, and among the seven continents, as shown in Figure 2. For the period 1950 to the present, the temperature increase in the Southern Hemisphere land areas is lower than that in the Northern; this agrees with prior assessments [Trenberth et al., 2007]. Snow-albedo feedbacks and greater average distance from large bodies of water have been previously implicated in the greater rate of warming in the North. The large swings seen in the period from the earliest parts of our reconstruction prior to 1850 are statistically significant, though of rather uncertain amplitude, but because of our sparse coverage, may be more closely linked to regional climate changes in Europe and North America rather than to global processes. Similar swings were reported in O-18 measurements in Greenland ice by Chylek et al. [2011] who associated them with the Atlantic Multidecadal Oscillation (AMO); we discuss this similarly further later in this paper. The largest excursions also occur at times of significant volcanic events as discussed below.

189 190 9

191 192 193 194 195 196 197

Figure 2. The temperature records of the land divided by hemisphere, and by 7 continents, shown as a 10-year running average. All records were normalized to zero mean for 1950 to 1980.

Figure 3 shows a map of our estimated temperature change from the decade 1951-1960 to the decade 2001-2010 for all land regions.

198 199 200 201 202 203 204 205 206 207 In addition to data for Tavg, for most sites we have also analyzed data on Tmax and Tmin, the daily maximum and minimum temperatures respectively. The reconstructed global land-average temperature time series for Tmax and Tmin are similar to Tavg. The summary values for the change from 1950s to 2000s is found to be 0.867 0.053 C and 10 Diurnal Range Figure 3: Map of the estimated temperature change from the 1950s decade to the 2000s decade for all land regions.

208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231

1.046 0.066 C for Tmax and Tmin respectively. In general, these time series differ only slightly from Tavg and from each other; however, we find that the diurnal range (i.e. Tmax minus Tmin) has some surprising properties.

Some of the climate models predict that the diurnal temperature range, that is, the difference between Tmax and Tmin, should decrease due to greenhouse warming. The physics is that greenhouse gases have more impact at night when they absorb infrared and reduce the cooling, and that this effect is larger than the additional daytime warming. This predicted change is sometimes cited as one of the fingerprints that separates greenhouse warming from other effects such as solar variability. Previous studies [Karl et al., 1991; Easterling et al., 1997; Braganza et al. 1998; Jones et al. 1999] reported significant decreases in the diurnal temperature range over the period 1948 to 1994. Jones et al. [1999] for example described the decrease as 0.08C per decade for the period 1950 to 1993.

To determine diurnal range changes we used two approaches. The first was to compute global land averages for Tmax and Tmin completely independently and then take their difference. The second approach was to construct a diurnal range time series as each station equal to the local difference between Tmax and Tmin and then apply the BerkeleyAverage methodology to these diurnal range time series. These two approaches produced qualitatively the same results, however the second approach led to significantly smaller uncertainty estimates, presumably because taking the difference locally could sometimes eliminate inhomogeneities that affected both Tmax and Tmin.

11

232 233 234 235 236 237 238 239 240 241 242 243 244 245

The result of this calculation is shown in Figure 4. The solid line represents the annual average of the diurnal range, and the dashed line shows the 10-year running average. The 1- and 2-standard deviation error uncertainties are shown with the two grey bands for the 10-year average. The behavior of the diurnal range is not simple; it drops from 1900 to 1987, and then it rises. The rise takes place during a period when, according to the IPCC report, the anthropogenic effect of global warming is evident above the background variations from natural causes.

Although the post-1987 rise is not sufficient to undo the drop that took place from 1901 to 1987, the trend of 0.86 0.13 C/century is distinctly upwards with a very high level of confidence. This reversal is particularly odd since it occurs during a period when the rise in Tavg was strong and showed no apparent changes in behavior

12

246 247 248 249 250 251 252 253 254 255 256 257 From 1950 to 2010, because of the recent rise, the net change we observe is -0.043 0.011 C/decade. We are not aware of any global climate models that predicted the reversal of slope that we observe. For comparison, Jones [1999] reported a drop of 0.08C per decade for the period 1950 to 1993 for the global diurnal change. Figure 4. Earth average diurnal range, calculated from records of D = Tmin Tmax using the BerkeleyAverage estimation program. Shown are both the yearly averages and the 10-year averages, and the 1- and 2- standard deviation uncertainty estimates for the 10-year averages.

13

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281

The Volcanic Era

As has been widely discussed [Hansen et al., 1992; Kelly et al., 1996], several of the cool periods in the early 1800s have been associated with large volcanic eruptions. Since our work extends the world temperature estimates further back in time than the previous Earth land averages, we can make a somewhat more global comparison than was previously possible. Of course, since most of the oldest records are in the Northern Hemisphere; this reflects itself in our plots by the large uncertainty limits for the early data.

In order to explain the largest variations in the land-average temperature, we approximate it as a linear combination of several factors. Those factors include a term proportional to the measured volcanic sulfate emissions into the stratosphere from ice cores [Gao et al., 2008] and a representative anthropogenic term, for which we used the natural log of the carbon dioxide concentration, log(CO2). We note that there are many other greenhouse gases and types of anthropogenic forcing besides, CO2; however, nearly all such forcings follow a similar time evolution, and because our fit is so simple (a linear superposition of known curves) there was value in keeping the number of parameters small. For the analysis that follows it is not possible to distinguish between anthropogenic forcings that have similar time histories. (We tried, for example, including a historic term for methane; the resulting fit was virtually identical, and the data insufficiently precise to determine the relative components of CO2 and methane.) Consequently using log(CO2) as the representative anthropogenic term rather than CO2 concentration, or IPCC estimates of total greenhouse gas forcing or total anthropogenic forcing make no significant difference in the conclusions below. 14

282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 A linear combination of volcanic sulfates and CO2 changes were fit to the land-surface temperature history to produce Figure 5. As we will describe in a moment, the addition of a solar activity proxy did not significantly improve the fit. The large negative excursions are associated with volcanic sulfate emissions, with the four largest eruptions having all occurred pre-1850; thus our extension to the pre-1850 data proved useful for the observation of these events. To perform the fit, we adjusted the sulfate record by applying an exponential decay with a two year half-life following emission. The choice of twoyears was motivated by maximizing the fit, and is considerably longer than the 4-8 month half-life observed for sulfate total mass in the atmosphere (but plausible for reflectivity which depends on area not volume). It is likely that longer period reflects dynamic climate responses that are slower than the simple settling time of the total sulfate mass. We also tried fitting to other empirical functions (exponential, parabolic, population growth), but none of these fit the data nearly as well a the log(CO2 ) function.

297 298 299 Figure 5: The annual and decadal land surface temperature from the BerkeleyEarth average, compared to a linear combination of volcanic sulfate emissions (responsible 15

300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323

for the short dips) and the natural logarithm of CO2 (responsible for the gradual rise). Inclusion of a proxy for solar activity did not significantly improve the fit. The grey area is the 95% confidence interval.

Most of the eruptions with significant stratospheric sulfate emissions can be qualitatively associated with periods of low temperature. For most of these events, the volcano and year of emission is historically known. These include the eruptions of Laki in 1783, Tambora in 1815, and Cosiguina in 1835. The peak at 1809 does not have an associated historical event, and it might consist of two separate events [Yalcin et al., 2006; Cole-Dai, 2009]. The famous explosion of Krakatau in 1883 is smaller (measured in sulfates) than these other events.

The temperature forcing of volcanic aerosols is a complicated function of latitude, altitude, season, and particle size; see Kelly et al. [1996]. However, the fit presented here can provide a rough estimate. We observe a response of -1.5 0.5 C per 100 Tg of atmospheric sulfate emitted. The 95% confidence interval quoted here is primarily influenced by the uncertainties in the temperature data, however we also allowed the magnitude of each eruption to have a 1-sigma error of 15%. A more sophisticated analysis of the forcings and the details of the climate response may be able to improve upon the crude estimate offered here based solely on the linear combination fit.

In the simple linear combination, the anthropogenic term based on the logarithm of CO2 concentration has an effective response of 3.10 0.34 C at doubled CO2 (95% confidence). This is within the IPCC range for the equilibrium climate sensitivity at 16

324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347

doubled CO2 of 2-4.5 C; however, several important caveats apply. Firstly, the estimate provided here is based solely on land observations, and as such should be expected to overestimate the global change. Based on the last fifty years, IPCC observations suggest the land has warmed ~35% faster than the global average. Secondly, these numbers reflect the transient climate response to ongoing increases in CO2. Based on the global climate models runs reported by the IPCC, such a transient response will underestimate the equilibrium response by ~20-50%. Lastly, we reemphasize that CO2 is being used here as proxy for all anthropogenic effects, and is only reasonable because most anthropogenic effects are roughly proportionally to CO2. Despite these caveats, and the offsetting corrections for faster land warming and transient climate response, our very simple linear model is consistent with a climate sensitivity of around 3 C for an effective doubling CO2, consistent with the IPCC.

In producing the linear fit shown here, we also tested to see whether solar forcing played an important role. We find no significant correlation between the land surface temperature history of the last 250 years and the solar forcing history specified for use in the IPCC fifth assessment report global climate models. Specifically, a linear combination including solar produces a climate sensitivity to the IPCC solar forcing function of 0.03 0.49 C / (W/m2), which is consistent with zero but also subject to large uncertainty. (Note that for this function the W/m2 represent the variations assumed by the IPCC radiative forcing model, not actual measured W/m2.) We also attempted a fit where all parameters were decadally averaged to smooth over the large swings associated with the solar cycle in the IPCC forcing model. This also produced an insignificant result of 0.6 1.7 C / (W/m2) for solar forcing while CO2 and volcanic changes remained highly 17

348 349 350 351 352 353 354 355 356 357 358 359

significant. Based on these observations, we conclude that either solar forcing has not been a major component of the changes in the land-surface temperature record or that the assumed history for solar forcing is too inaccurate in shape to make a simple linear fit possible.

Decadal and sub-decadal variations

The short term (2 to 15 year) variations in the temperature record in the modern period, 1950 onward, have frequently been associated with El Nino events. Figure 6 shows the AMO Index along with the residual temperature record fluctuations after subtracting the fit curve shown in Figure 5.

360

18

361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384

Figure 6. The residual multi-decadal component (MDC) of the Earth land surface temperature compared to the Atlantic Multidecadal Oscillation (AMO). The upper panel shows both datasets on an annual scale (12-month moving average), while the lower panel applies a 10-year moving average to highlight decadal structure. The MDC was obtained by subtracting the fit curve shown in Figure 5. The MDC and the AMO have similar amplitude and structure for both the 2-3 year time scale and for the multi-decadal time scale. Error bands (95% confidence interval) reflect the temperature reconstruction uncertainties only; they do not include any uncertainty associated with the fit.

Although the agreement is not perfect, there is a remarkable correspondence in both the short-term 2-3 year variability, and also in the longer term multi-decadal oscillations that give the AMO its name. Although the variation has sometimes been described as a 70-year cycle (based on the single cycle seen in the data of Figure 6 from 1940 to 2010, its behavior does not seem that regular. Because it is seen in the land as well as the North Atlantic, we will refer to it simply as the multidecadal oscillation of the land surface temperature. Note that the amplitudes of the curves in Figure 6 have not been rescaled, and hence the apparent correspondence indicates that these fluctuations over land and ocean truly are of similar magnitude. Of course, it is not too surprising that the AMO a measure of temperatures in the North Atlantic match those of the Earth average land temperature, since 2/3 of the land is in the Northern Hemisphere.

It is unknown whether the observed decadal changes in AMO and land-average temperature are the results of natural variability, anthropogenic forcing, or some 19

385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406

combination of the two. Since the fit conducted in the previous section only used CO2 as a crude proxy of anthropogenic change it is entirely possible that a more complete understanding the anthropogenic forcings could explain these variations. However, it is doubtful whether the small temperature fluctuations we observe could be accounted for with any confidence given the relatively large uncertainties associated with many anthropogenic forcings. An alternative is to assume that some or all of these variations represent a form of natural variability. Using the curve from Figure 6, we can estimate that such variability on decadal scales is no more than 0.17 C, 95% of the time. This can be understood as a crude bound on the amount of temperature change that might potentially be ascribed to natural variability. Such a bound is small, but not trivial, compared to the 0.87 C warming since 1950s.

A cross-correlation between the residual multidecadal oscillation and the AMO is shown in Figure 7. A sharp peak occurs near zero lag, at 0.1 0.3 year; this is a result of the close correspondence between the decadal and subdecadal variations of the two signals. In addition to this, there is a lower but broad hump rising from lag of -15 yr and finally falling at a lag of +30 yr, with a center at about lag = +10 3 years. This broad peak indicates that the multidecadal structure in the AMO lags the land multidecadal structure by about a decade; however, the peak is broad, and the actual maximum occurs near zero lag. That can be understood as arising from two components, rapid part in which the ocean and the land vary together, and a slower one in which long term trends in ocean temperatures lags that on land.

20

407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 Discussion We have obtained an estimate of the Earth land surface temperature from 1753 unto the present. The limited land coverage prior to 1850 results in larger uncertainties in the 21 This behavior, with the two peaks (one at zero, the other at 10 years) can be understood physically if both land and ocean are responding to the same underlying process that has both sub-decadal and multi-decadal fluctuations . The subdecadal changes are rapid, but presumably only affect the shallow mixed layer of the oceans, so the land and the surface layers of the oceans respond essentially simultaneously. The multidecadal variations have time to penetrate into the deeper oceans, and that takes longer, giving rise to the lag between the quick land response and the slow oceanic temperature change. Figure 7. Correlation between the AMO and our observed multidecadal oscillaiton. The peak near zero represents the strong correlation between the decadal and subdecadal variations in the two sets; the broad peak centered near +10 years indicates that temperature changes in the north Atlantic lag those of the global land average for the multi-decadal time scale.

425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448

behavior of the record; despite these, we see behavior that is significant. Most dramatic are the large swings in the earliest period. These dips can be explained as the effect of large volcanic eruptions that took place during that period. The rapid changes in the Earths temperature at that time are remarkably swift, and at times even greater than the changes taking place in the last 50 years. Our records also show an average temperature during the early 1800s that is on the lower end of what had previously been estimated from proxy measurements, although there are large discrepancies between the values obtained by using different proxy sets.

The behavior changes in the early 1900s, and follows closely the results that had been previously reported by the three other major groups that analyze historic thermometer records, but with a smaller uncertainty than has been previously achieved. Since the 1950s, we observe a rise in the average land surface temperature of 0.87 0.05 C (95% confidence). This value is in the middle of the comparable values reported by other groups, but with an estimated uncertainty approximately twice as tight as prior reports.

We observe that the record of diurnal temperature range, Tmax - Tmin, follows an unexpected path, with a slow drop from the period 1900 to the late 1980s, followed by a rise up to the most recent period (2011). This change in direction is unexpected and not anticipated by existing climate models.

Many of the changes in land-surface temperature follow a simple linear combination of volcanic forcing (based on estimates of stratospheric sulfate injection) and an anthropogenic term represented here by the logarithm of the CO2 concentration. The best 22

449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471

fit an volcanic forcing term is -1.5 0.5 C per 100 Tg of atmospheric sulfate. The anthropogenic forcing parameter is 3.1 0.3 C for CO2 doubling (compared to preindustrial levels), broadly consistent with the IPCC estimate of ~3 C for the equilibrium warming at doubled CO2. When we included solar forcing we found that the solar variability record assumed by the IPCC did not contribute significantly to the fit of historic temperature. This could imply that any effect associated with solar variability is too small to be detected by our simple approach. It might also imply that the shape of solar forcing assumed by the IPCC during the last 250 years is too inaccurate for an effective comparison. However, if the shape of the solar forcing history is accurate, then the impact of solar variability on climate would have to be on the low side of present estimates, no more than 0.08 C since 1750.

After accounting for volcanic and anthropogenic effects, the residual variability in landsurface temperature is observed to closely mirror and for slower changes slightly lead variations in the Atlantic Multidecadal Oscillation Index. This is consistent with both the land and North Atlantic responding the same unknown process. That process may be partially anthropogenic and include effects whose time evolution is not proportional CO2. It might also include natural processes. If all of the residual evolution during the last 150 years is assumed to be natural, then it places an upper 95% confidence bound on the scale of decadal natural variability at 0.17 C. Though non-trivial, this number is small compared to the anthropogenic changes that appear to have occurred during the last century.

23

472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494

Acknowledgements

We thank David Brillinger for important guidance in statistical analysis, Zeke Hausfather, Steven Mosher, and Judith Curry for helpful discussions and suggestions. This work was done as part of the Berkeley Earth project, organized under the auspices of the Novim Group (www.Novim.org). We thank many organizations for their support, including the Lee and Juliet Folger Fund, the Lawrence Berkeley National Laboratory, the William K. Bowes Jr. Foundation, the Fund for Innovative Climate and Energy Research (created by Bill Gates), the Ann and Gordon Getty Foundation, the Charles G. Koch Charitable Foundation, and three private individuals (M.D., N.G. and M.D.). More information on the Berkeley Earth project can be found at www.BerkeleyEarth.org.

References

Alexander, L. V., X. Zhang, T. C. Peterson, J. Caesar, B. Gleason, A. M. G. Klein Tank, M. Haylock, D. Collins, B. Trewin, F. Rahimzadeh, A. Tagipour, K. Rupa Kumar, J. Revadekar, G. Griffiths, L. Vincent, D. B. Stephenson, J. Burn, E. Aguilar, M. Brunet, M. Taylor, M. New, P. Zhai, M. Rusticucci, and J. L. Vazquez-Aguirre (2006) Global observed changes in daily climate extremes of temperature and precipitation, Journal of Geophysical Research, v. 111, D05109. Braganza, K., D. J. Karoly, J. M. Arblaster, Diurnal temperature range as an index of global climate change during the twentieth century, 2004, Geophys. Res. Lett. vol 31, L13217, doi:10.1029/2004GL019998.

24

495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518

Brohan, P., J. J. Kennedy, I. Harris, S. F. B. Tett, and P. D. Jones (2006), Uncertainty estimates in regional and global observed temperature changes: A new data set from 1850, J. Geophys. Res., 111, D12106, doi:10.1029/2005JD006548. Chylek, Petr, Chris K. Folland, Henk A. Dijkstra, Glen Lesins, and Manvendra K. Dubey, Ice core data evidence for a prominent near 20 year time scale of the Atlantic Multidecadal Oscillation, Geophysical Research Letters,Vol. 38, L13704, doi:10.1029/2011GL047501, 2011 Cole-Dai, J., D. Ferris, A. Lanciki, J. Savarino, M. Baroni, M. H. Thiemens, 2009, Cold decade (AD 18101819) caused by Tambora (1815) and another (1809) stratospheric volcanic eruption, Geophys. Res. Lett., vol 36, L22703, doi:10.1029/2009GL040882, 2009 Cressie, N., 1990, The Origins of Kriging. Mathematical Geology, Vol. 22, No. 3, 1990. Easterling, D. R., B. Horton, P. D. Jones, T. C. Peterson, T. R. Karl, D. E. Parker, M. James Salinger, V. Razuvayev, N. Plummer, P. Jamason and C. K. Folland. Maximum and Minimum Temperature Trends for the Globe Science. Vol. 277 no. 5324 pp. 364-367. Gao, C., A. Robock, C. Ammann, 2008: Volcanic forcing of climate over the past 1500 years: An improved ice-core-based index for climate models. J. Geophys. Res., 113, D23111, doi:10.1029/2008JD010239. Data available at ftp://ftp.ncdc.noaa.gov/pub/data/paleo/climate_forcing/volcanic_aerosols/gao2008i vi2/ivi2totalloading501-2000.txt Hansen, J., Lacis, A,, Ruedy, R. and Sato, M. 1992. Potential climate impact of Mount Pinatubo eruption, Geophys. Res. Lett., 19, 215-218. 25

519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541

Hansen, J., R. Ruedy, J. Glascoe, and Mki. Sato, 1999: GISS analysis of surface temperature change. J. Geophys. Res., 104, 30997-31022, doi:10.1029/1999JD900835. Hansen, J., R. Ruedy, Mki. Sato, and K. Lo, 2010: Global surface temperature change. Rev. Geophys., 48, RG4004, doi:10.1029/2010RG000345. Hansen, J.E., and S. Lebedeff, 1987: Global trends of measured surface air temperature. J. Geophys. Res., 92, 13345-13372, doi:10.1029/JD092iD11p13345. Hansen, James, Makiko Sato, Reto Ruedy, 2012, Perceptions of Climate Change: The New Climate Dice (on line; submitted?) Kelly, P.M., P.D. Jones, J. Pengqun (1996), The Spatial Response of the Climate System to Explosive Volcanic Eruptions, International J. Climatology, vol. 16, 537-550. Jones, P. D., M. New, D. E. Parker, S. Martin, and I. G. Rigor (1999), Surface air temperature and its changes over the past 150 years, Rev. Geophys., 37(2), 173 199, doi:10.1029/1999RG900002. Jones, P. D., and A. Moberg (2003), Hemispheric and Large-Scale Surface Air Temperature Variations: An Extensive Revision and an Update to 2001, J. Clim., 16, 20623. Jones, P.D., T.M.L. Wigley, and P.B. Wright. 1986. Global temperature variations between 1861 and 1984. Nature322:430-434. Karl, Thomas, G. Kukla, V. Razuvayev, M. Changery, R. Quayle, R. Helm, D. Easterling, C. Fu, Global warming: Evidence for asymmetric diurnal temperature change, Geophysical Research Letters, vol. 18, no. 12, PP. 2253-2256, 1991 doi:10.1029/91GL02900

26

542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564

Krige, D.G, A statistical approach to some mine valuations and allied problems at the Witwatersrand, Master's thesis of the University of Witwatersrand, 1951. Miller, Rupert (1974), The Jackknife A review, Biometrika, v. 61, no. 1, pp. 1-15. Oppenheimer, Clive (2003). "Climatic, environmental and human consequences of the largest known historic eruption: Tambora volcano (Indonesia) 1815". Progress in Physical Geography 27 (2): 230259. Smith and Reynolds, 2005: A global merged land air and sea surface temperature reconstruction based on historical observations (18801997). J. Climate, 18, 2021 2036. Smith, T. M., et al. (2008), Improvements to NOAA's Historical Merged Land-Ocean Surface Temperature Analysis (1880-2006), J. Climate, 21, 2283-2293. Stothers, Richard B. (1984). "The Great Tambora Eruption in 1815 and Its Aftermath". Science 224 (4654): 11911198. Trenberth, K.E., P.D. Jones, P. Ambenje, R. Bojariu, D. Easterling, A. Klein Tank, D. Parker, F. Rahimzadeh, J.A. Renwick, M. Rusticucci, B. Soden and P. Zhai, 2007: Observations: Surface and Atmospheric Climate Change. In: Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change [Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor and H.L. Miller (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Tsonis, A., K. Swanson, S. Kravtsov, 2007, Geophys. Res. Lett. vol. 34, L13705, doi:10.1029/2007GL030288.

27

565 566 567 568 569 570 571 572

Tukey, J.W. (1958), Bias and confidence in not quite large samples, The Annals of Mathematical Statistics, 29, 614. Wagner, Sebastian and Eduardo Zorita (2005) The inuence of volcanic, solar and CO2 forcing on the temperatures in the Dalton Minimum (17901830): a model study, Climate Dynamics v. 25, pp. 205218. Yalcin, K., C. P. Wake, K.J. Kreutz, M. S. Germani, S. I. Whitlow, 2006, Ice core evidence for a second volcanic eruption around 1809 in the Northern Hemisphere, Geophs. Res. Lett. vol 33, L14706, doi:10.1029/2006GL026013, 2006

28

573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598

FIGURE CAPTIONS

Figure 1. Land temperature with 1- and 10-year running averages. The shaded regions are the one- and two-standard deviation uncertainties calculated including both statistical and spatial sampling errors. Prior land results from the other groups are also plotted. The NASA GISS record had a land mask applied; the HadCRU curve is the simple land average, not the hemispheric-weighted one. Figure 2. The temperature records of the land divided by hemisphere, and by 7 continents, shown as a 10-year running average. All records were normalized to zero mean for 1950 to 1980. Figure 3: Map of the estimated temperature change from the 1950s decade to the 2000s decade for all land regions. Figure 4. Earth average diurnal range, calculated from records of D = Tmin Tmax using the BerkeleyAverage estimation program. Shown are both the yearly averages and the 10-year averages, and the 1- and 2- standard deviation uncertainty estimates for the 10-year averages.

Figure 5: The annual and decadal land surface temperature from the BerkeleyEarth average, compared to a linear combination of volcanic sulfate emissions and the natural logarithm of CO2. It is observed that the large negative excursions in the early temperature records are likely to be explained by exceptional volcanic activity at this time. Similarly, the upward trend is likely to be an indication of anthropogenic changes.

29

599 600 601 602 603 604 605 606 607 608 609 610 611

Figure 6. The residual multi-decadal component (MDC) of the Earth land surface temperature compared to the Atlantic Multidecadal Oscillation (AMO). The upper panel shows both datasets on an annual scale (12-month moving average), while the lower panel applies a 10-year moving average to highlight decadal structure. The MDC was obtained by subtracting the fit curve shown in Figure 5. The MDC and the AMO have similar amplitude and structure for both the 2-3 year time scale and for the multi-decadal time scale. Error bands (95% confidence interval) reflect the temperature reconstruction uncertainties only; they do not include any uncertainty associated with the fit. Figure 7. Correlation between the AMO and the MDC. The peak near zero represents the strong correlation between the decadal and subdecadal variations in the two sets; the broad peak centered near +10 years indicates that temperature changes in the north Atlantic lag those of the global land average for the multi-decadal time scale.

30