«Robert Rohde1, Judith Curry2, Donald Groom3, Robert Jacobsen3,4, Richard A. Muller1,3,4, Saul Perlmutter3,4, Arthur Rosenfeld3,4, Charlotte Wickham5, ...»
time, and the land area sampled vs. time (calculated using the method described in equation ). The sudden drop in the number of stations ca. 1990 is largely a result of the methodology used in compiling the GHCN dataset; GHCN generally only accepts records for stations that explicitly issue a monthly summary report however many stations have stopped reporting monthly results and only reported daily ones. Despite this drop, Figure 4(c) shows that the coverage of the Earth’s land surface remained above 95%, reflecting the broad distribution of the stations that did remain.
Figure 4. (Upper) Station locations for the 7280 temperature stations in the Global Historical Climatology Network Monthly dataset.
(Lower Left) Number of active stations over time.
(Lower Right) Percentage of the Earth’s land area sampled by the available stations versus time, calculated as explained in the text. The transition during the mid 1950s corresponds to the appearance of the first temperature records on Antarctica.
We applied the Berkeley Average methodology to the GHCN monthly data. The results and associated uncertainties are shown in Figure 5. The upper plot shows the 12-month landonly moving average and its associated 95% uncertainty; the lower plot shows the result of applying a 10-year moving average. Applying the methods described here, we find that the average land temperature from Jan 1950 to Dec 1959 was 8.849 ± 0.033 C, and temperature average during the most recent decade (Jan 2000 to Dec 2009) was 9.760 ± 0.041 C, an increase of 0.911 ± 0.042 C. The trend line for the 20th century is calculated to be 0.733 ± 0.096 C/century, well below the 2.76 ± 0.16 C/century rate of global land-surface warming that we observe during the interval Jan 1970 to Aug 2011. (All uncertainties quoted here and below are 95% confidence intervals for the combined statistical and spatial uncertainty). Though it is sometimes argued that global warming has abated since the 1998 El Nino event (e.g. Easterling and Wehner 2009, Meehl et al. 2011), we find no evidence of this in the GHCN land data.
Applying our analysis over the interval 1998 to 2010, we find the land temperature trend to be
2.84 ± 0.73 C / century, consistent with prior decades. Meehl et al. (2011) associated the recent decreases in global temperature trends with increased heat flux into the deep oceans. The fact that we observe no change in the trend over land would seem to be consistent with the conclusion that any change in the total global average has been driven solely with oceanic processes.
Figure 5. Result of the Berkeley Average Methodology applied to the GHCN monthly data.
Top plot shows a 12-month land-only moving average and associated 95% uncertainty from statistical and spatial factors. The lower plot shows a corresponding 10-year land-only moving average and 95% uncertainty. This plot corresponds to the parameter in Equation 5. Our plotting convention is to place each value at the middle of the time interval it represents. For example, the 1991-2000 average in the decadal plot is shown at 1995.5.
In the section on the sampling method, we discussed the determination of statistical uncertainties by dividing the full data set into five subsamples. In Figure 6 below, we show the results of doing this for the GHCN data set. We show this primarily because the sampling method is more intuitive for many people than is the jackknife, and the charts in Figure 6 make it clear why the statistical uncertainties are small. The five completely independent subsamples produce very similar temperature history when processed via the Berkeley Average methodology.
Figure 6. Five independent temperature reconstructions each derived from a separate 20% of the GHCN stations.
The upper figure shows the calculation of the temperature record based on five independent subsamples. The lower plot shows their difference from the 100% result, and the expected 95% uncertainty envelope. The uncertainty envelope used here is scaled by √ times the statistical uncertainty reported for the complete Berkeley Average analysis. This reflects the larger variance expected for the 20% samples.
The spatial structure of the climate change during the last century is shown in Figure 7 and found to be fairly uniform, though with greater warming over the high latitudes of North America and Asia, consistent with prior results (Hansen et al. 2010). We also show the pattern of warming since the 1960s, as this is the period during which anthropogenic effects are believed to have been the most significant. Warming is observed to have occurred over all continents, though parts of South America are consistent with no change. No part of the Earth’s land surface shows appreciable cooling.
Figure 7. Maps showing the decadal average changes in the land temperature field.
In the upper plot, the comparison is drawn between the average temperature in 1900 to 1910 and the average temperature in 2000 to 2010. In the lower plot, the same comparison is made but using the interval 1960 to 1970 as the starting point. We observe warming over all continents with the greatest warming at high latitudes and the least warming in southern South America.
In Figure 8, we compare our land reconstruction to the land reconstructions published by the three other groups (results updated online, methods described by Brohan et al. 2006; Smith et al. 2008; Hansen et al. 2010). Overall our global land average is similar to those obtained by these prior efforts. There is some disagreement amongst the three groups, and our result is most similar overall to NOAA’s work. The differences apparent in Figure 8 may partially reflect difference in source data, but they probably primarily reflect differences in methodology.
The GHCN dataset used in the current analysis overlaps strongly with the data used by other groups. The GHCN was developed by NOAA and is the sole source of the land-based weather station data in their temperature reconstructions (but does not include the ocean data also used in their global temperature analyses). In addition, GISS uses GHCN as the source for ~85% of the time series in their analysis. The remaining 15% of GISS stations are almost exclusively US and Antarctic sites that they have added / updated, and hence would be expected to have somewhat limited impact due to their limited geographic coverage. HadCRU maintains a separate data set from GHCN for their climate analysis work though approximately 60% of the GHCN stations also appear in HadCRU.
Figure 8. Comparison of the Berkeley Average to existing land-only averages reported by the three major temperature groups.
The upper panel shows 12-month moving averages for the four reconstructions, and a gray band corresponding to the 95% uncertainty range on the Berkeley average. The lower panel shows each of the prior averages minus the Berkeley average, as well as the Berkeley average uncertainty. As noted in the text, there is a much larger disagreement among the existing groups when considering land-only data than when comparing the global averages. HadCRU and GISS have systematically lower trends than Berkeley and NOAA. In part, this is likely to reflect differences in how “land-only” has been defined by the three groups.
Berkeley is very similar to the NOAA result during the twentieth century and slightly lower than all three groups during the 19th century.
The GISS and HadCRU work produce lower land-average temperature trends for the late part of the 20th century. In this regard, our analysis suggests a degree of global land-surface warming during the anthropogenic era that is consistent with prior work (e.g. NOAA) but on the high end of the existing range of reconstructions. We note that the difference in land average trends amongst the prior groups has not generally been discussed in the literature. In part, the spread in existing land-only records may have received little attention because the three groups have greater agreement when considering global averages that include oceans (Figure 1). We strongly suspect that some of the difference in land-only averages is an artifact of the different approaches to defining “land-only” temperature analyses. Our analysis and that produced by NOAA explicitly construct an average that only considers temperature values over land.
However, that is not the only possible approach. The literature suggests that the GISS “landonly” data product may be generated by measuring the “global” temperature fields using only data reported over land. In this scenario temperature records in coastal regions and on islands would be extrapolated over the oceans to create a “global” field using only land data. Whether or not this approach was actually used is unclear from the literature, but it would result in an overweighting of coastal and oceanic stations. This would in turn lead to a reduction in the calculated “land” trend in a way that is qualitatively consistent with the difference observed in Figure 8.
Though we are similar to NOAA for most of the 20th century, we note that we have somewhat lower average temperatures during the period 1880-1930. This gives us a slightly larger overall trend for the 20th century than any of the three groups. Most of that difference comes from the more uncertain early period. In previous work, it has been argued that instrumentation changes may have led to an artificial warm bias in the early 1900s (Folland et al.
2001, Parker 1994). To the degree that our reconstruction from that era is systematically lower than prior work (Figure 8), it could be that our methods are more resistant to biases due to those instrumental changes.
As is shown in Figure 5, we extend our record all the way back to 1800, including 50 more years than HadCRU and 80 more years than NOAA and GISS. We feel this extension is justifiable though obviously, any such reconstruction will have large uncertainties. Our analysis technique suggests that temperatures during the 19th century were approximately constant (trend
0.20 ± 0.25 C/century) and on average 1.48 ± 0.13 C cooler than the interval 2000-2009. Circa 1820 there is a negative temperature excursion that happens to roughly coincide with both the 1815 eruption of Mount Tambora and the Dalton Minimum in solar activity. The Mount Tambora eruption was the largest eruption in the historical era and has been blamed for creating the “year without a summer” (Oppenheimer 2003; Stothers 1984). It was preceded by an additional large eruption in 1809 (Wagner and Zorita 2005). The Dalton Minimum in solar activity from circa 1790 to 1830 includes the lowest 25 year period of solar activity during the last 280 years, but this is considered to have produced only minor cooling during this period, while volcanism was the dominant source of cooling (Wagner and Zorita 2005). Though the uncertainties are very large, the fact that this temperature excursion is well-established in the historical record and motivated by known climate forcings gives us confidence that the ~1820 excursion is a reflection of a true climate event. However, we will note that our early data is heavily biased towards North America and Europe, so we cannot draw conclusions about the regional versus global extent of the event.
As discussed above, the uncertainty in our result is conceptually divided into two parts, the “statistical uncertainty” which measures how well the temperature field was constrained by data in regions and times where data is available, ( ⃑ ), and the “spatial uncertainty” which measures how much uncertainty has been introduced into the temperature average due to the fact that some regions are not effectively sampled, ( ⃑ ). These uncertainties for the GHCN analysis are presented in Figure 9.
Figure 9. The 95% uncertainty on the Berkeley Average (red line) and the component spatial (blue) and jackknife statistical (green) uncertainties for 12-month moving land averages.