Guest post by Dr. Willie Soon, Dr. Ronan Connolly & Dr. Michael Connolly
Recently, a new paper which we co-authored with five other researchers was published in Earth-Science Reviews entitled, “Comparing the current and early 20th century warm periods in China”. The paper is paywalled, but the journal has kindly allowed free access to the article until 20th July 2018 at this link here. If you’re reading this post after that date, you can download a pre-print here:Soon et al, 2018 ESR – China SAT trends (PDF)
The Supplementary Information and data for the paper is available here (Excel file) :Soon et al, 2018 ESR – China SAT trends – SI
The paper is quite technical and focuses specifically on Chinese temperature trends. But, we think that it will still be of interest to many readers here, especially anybody who is interested in any of the following topics:
- Urbanization bias
- The homogenization of temperature data
- The “early 20th century warm period” found in many parts of the Northern Hemisphere, and
- Comparing temperature proxies to instrumental records
Background to the project
Most studies of regional temperature trends for China have found at least two warm periods for the last century – the current warm period, and an early warm period from the 1920s until the 1940s. As part of our 2015 study of Northern Hemisphere temperature trends (abstract here; preprint here), we did our own analysis of Chinese temperature trends. We found that – for rural China – the 1940s warm period was actually hotter than present.
This was quite surprising in terms of the current human-caused global warming explanation. In the 1940s, the atmospheric concentration of carbon dioxide (CO2) at 0.03% was still relatively close to the pre-industrial concentrations of 0.027-0.029%, while currently the concentration is about 0.04%. For that reason, the current Global Climate Models calculate that (a) there shouldn’t have been an early 20th century warm period, and (b) the current warm period should be the hottest on record. This can be seen from the plot below which shows the mean “hindcast” for China of all 41 of the “CMIP5” climate models. These are the models which were used for the most recent UN IPCC reports:
In our 2015 paper, we argued that the most likely reason for the poor performance of the climate models is that they are underestimating the role of natural climate change and overestimating the role of carbon dioxide. [See also Soon et al., 2011 (abstract here; pdf here)].
However, while we were finding a very warm 1940s period for China, other groups were finding that the recent warm period was the “hottest on record” by far. For instance, the figure below is adapted from Liu et al., 2017 (abstract, paywalled unfortunately) – a study entitled, “Unprecedented warming revealed from multi-proxy reconstruction of temperature in southern China for the past 160 years”:
Liu et al., 2017 did find a slight warm period in the 1940s, but they found that the recent warm period was much hotter.
When we combed through the literature on Chinese temperature trends, we found that there was actually a lot of debate over how the early-20th century and current warm periods compared to each other. Some studies found they were both quite similar, e.g., Ren et al., 2017 (abstract here; pdf here). But, others concluded that the earlier “warm period” was just a temporary localised blip, e.g., Li et al., 2017 (open access).
Why were different studies getting such different answers?
Along with two other colleagues (Prof. Hong Yan and Peter O’Neill), we decided to collaborate with three of the scientists who had reached the opposite conclusion to us (Prof. Jingyun Zheng, Prof. Quansheng Ge, Prof. Zhixin Hao) and investigate.
Prof. Zheng and Prof. Hao were both co-authors of the Liu et al., 2017 study mentioned above, i.e., the one claiming the current warm period was “unprecedented” in the last 160 years. And along with Prof. Ge, they have co-authored many papers which often reached very different conclusions from us, e.g., see Wang et al., 2018 (Open access). However, even though we each have different views on this subject, we all agreed that it is important to establish the reasons for the disagreements. Our new paper describes the results of this collaboration.
Summary of our key findings
1. Limitation of data for pre-1950s period
Probably, the biggest challenge for comparing the early 20th century and current warm periods in China is the shortage of long-term records. After the People’s Republic of China was founded in 1949, a nation-wide network of weather stations was installed across much of China, and so there is a relatively large amount of data for the post-1950s period. However, most of these stations were not available during the 1940s, i.e., at the time of the earlier warm period!
There is a lot of ongoing work to try and track down and digitize more data for China in the pre-1950s period, e.g., the so-called “ACRE” project – see Williamson et al., 2017 (Open access). But, for now, the available data is quite limited.
Some of you may have noticed that there was a surprising sudden drop in station numbers after 1990 for version 3 of both datasets. This was particularly pronounced for the GHCN dataset, which had a relatively large number of stations for the 1961-1990 period. There are several legacy reasons for this odd fall off, e.g., when GHCN was initially compiled, several of their main data sources had only been updated to 1990. However, with version 4 of both datasets, this particular problem seems to have been mostly resolved.
2. The urbanization bias problem
A related problem is that most of those stations with data for the early-20th century are in urban areas. This makes sense in that it is harder to staff a weather station continuously for decades if it is located in an isolated spot far from any urban area. However, urban areas tend to be warmer than the surrounding countryside due to the “Urban Heat Island” (UHI) effect. Since China has experienced a dramatic increase in urbanization over the last few decades, many of these urban station records have experienced a strong “urban warming” due to the growth of the local UHI.
This urban warming is a real climatic effect, and it is a growing problem for the Chinese population, since most people in China now live in urban areas. However, urban areas account for less than 1% of the land area of China. So, if you want to study the actual climatic trends for China, you need to correct for this “urbanization bias” problem.
To study this problem, we ranked all 494 Chinese stations in the GHCN version 4 dataset based on how urbanized they are today. We then divided the stations into five subsets according to these rankings. Below are the results for the most urban and most rural subsets:
We found that the more urbanized the subset, the hotter the recent warm period seemed to be. This suggests that at least some of the recent “warming” is an artefact of urbanization bias. However, unfortunately, the more rural the subset, the less data there was for the early 20th century – leading to greater error bars. So, if we want to compare the two periods, we can’t just rely on the rural data.
Also, we found that most of the urban stations were in eastern China, while most of the rural stations were from western and central-China.
3. The homogenization debate
Urbanization bias is by no means the only non-climatic bias associated with the temperature records. There are many different ways in which changes to the thermometer station and its environment can artificially alter the values of the “measured temperatures” independently of the real climate. For example, station moves, changes in the time of observation, changes in the immediate surroundings of the thermometer, etc.
Several groups have come up with different statistically based “homogenization” algorithms to adjust the station records to attempt to correct for such “step change” non-climatic biases, such as the Menne & Williams, 2009 algorithm used by NOAA (Open access).
Meanwhile, studies such as Venema et al., 2012 (Open access) have tested these algorithms by artificially adding synthetic biases to unbiased data, and seeing how good they are at removing them. On these tests, the main algorithms seem to do quite well. This has led many people to assume that the “homogenized” temperature datasets are more reliable than the “raw” (unhomogenized) datasets.
As it happens, when the Chinese temperature datasets are homogenized, this tends to slightly reduce the warmth of the early 20th century period and increase the warmth of the current period. This has led some researchers to conclude that the 1940s warm period was at least partially an artefact of non-climatic biases, e.g., Li et al., 2017 (Open access).
However, in Sections 3.2.3-3.2.5 of the paper, we point out that it ain’t necessarily so. We show that the current homogenization algorithms have a theoretical flaw called the “urban blending problem”. Whenever there are a lot of urban stations, and the urban stations are used as neighbours for homogenizing rural stations, some of the urban heat island of the urban stations gets added into the “rural” station records. This means that after homogenization, all of the stations have some urban warming in their records – even if they had originally been rural!
We can see the results in the following plot which shows the effects of homogenization on a sample of 10 stations near Beijing. The plot is adapted from He and Jia, 2012 (Open access).
We can see that before homogenization, there was a very strong correlation between the 1978-2008 warming trend and the rate of urbanization. That is, most of the warming trend was a result of urbanization bias. After homogenization, the differences in trends between the stations are reduced, and all of the stations have similar rates of warming. But, while the most heavily urbanized station has been cooled, the rural stations have been warmed. Instead of the urbanization bias being removed by homogenization, it has been merely spread among all stations (rural and urban).
This urban blending effect seems to be the main reason why homogenization “cools” the 1940s warm period for China. So, before we can rely on the homogenized data for China, we have to avoid urban blending. However, because there are so few Chinese rural stations for the early warm period, nobody has yet managed to construct a homogenized Chinese dataset for the early 20th century which only uses rural stations.
4. Inconsistency of temperature proxies
One possible approach to overcoming the urbanization bias problem could be to use so-called “temperature proxies”. These are datasets constructed from temperature-influenced measurements such as tree rings and ice cores. Although these are usually only indirect measurements of temperature, if they are calibrated against local thermometer records, they can be used to estimate long-term temperature trends for the region. The records often cover hundreds (or even thousands) of years. They also tend to be located in isolated areas, such as in the mountains, and therefore are usually unaffected by urbanization.
We think that temperature proxies are a very powerful tool for the climate science community and several of our co-authors on this paper do a lot of work with temperature proxies. However, they also have their limitations and should be used cautiously. We have noticed that there has been a tendency for researchers to focus mostly on the bits where different proxy series agree with each other. But, in Section 3.5, we point out that we should be comparing and contrasting the proxy series, rather than just focusing on the points of agreement.
Below are four different proxies for the same region, i.e., northeast China (although Wiles et al., 2014 is technically for northeast Asia). All of them show warming and cooling periods, and it is possible to find points of agreement between any two series. However, when you try to directly compare any two specific periods, e.g., the 1920s-1940s and 1980s-present, the results depend on which series you pick!
For instance, Zhu et al., 2015 implies that both periods were similarly warm (and that there were also similar warm periods in the 19th century). However, Zhu et al., 2016 implies that the 1940s were one of the coldest decades since the 19th century!
Therefore, we believe more research is needed before temperature proxies can be used for directly comparing periods such as the early 20th century and recent warm periods.
So, what can we say? The main conclusions of our study are as follows:
- The 1940s warm period seems to have been a real phenomenon in China. However, the data is probably still too limited to establish exactly how warm it was compared to the current warm period.
- The current climate models are unable to reproduce this early-20th century warm period, and can only simulate the recent warm period. The models attribute the recent warming to greenhouse gas emissions, and calculate that greenhouse gas emissions should not have caused an early-20th century warm period. This suggests that the earlier warm period was probably due to natural climate change – and that the current models are underestimating natural climate change.
- But, given that the models can’t explain the earlier warm period, it is quite plausible that some (or even most) of the recent warm period could also be due to similar natural climate changes to the ones which caused the 1940s warm period. That would imply that the current models are also overestimating the role of greenhouse gases in the recent warm period.
Given the fact that conspiracy theories abound in climate science, we should stress that all of the research for this collaborative paper was carried out during our free time and at our own expense. None of us received any funding for this research. The motivation for all eight of us in writing this paper was to try to advance the science on this topic.