By Judith Curry
It is therefore suggested to use either the more robust tropospheric temperature or ocean surface temperature in studies of climate sensitivity. – Cederlof, Bengtsson, Hodges
The discrepancies among temperature datasets, not to mention the temporal changes in individual datasets, fuels much debate about exactly how much the Earth is warming. Previous CE posts on this topic can be found [here].
An important new paper on this topic has just been published by Tellus:
Assessing atmospheric temperature data sets for climate studies
Magnus Cederlof, Lennart Bengtsson, Kevin Hodges
Abstract. Observed near-surface temperature trends during the period 1979–2014 show large differences between land and ocean, with positive values over land (0.25–0.27 °C/decade) that are significantly larger than over the ocean (0.06–0.12 °C/decade). Temperature trends in the mid-troposphere of 0.08-0.11 °C/decade, on the other hand, are similar for both land and ocean and agree closely with the ocean surface temperature trend. The lapse rate is consequently systematically larger over land than over the ocean and also shows a positive trend in most land areas. This is puzzling as a response to external warming, such as from increasing greenhouse gases, is broadly the same throughout the troposphere. The reduced tropospheric warming trend over land suggests a weaker vertical temperature coupling indicating that some of the processes in the planetary boundary layer such as inversions have a limited influence on the temperature of the free atmosphere. Alternatively, the temperature of the free atmosphere is influenced by advection of colder tropospheric air from the oceans. It is therefore suggested to use either the more robust tropospheric temperature or ocean surface temperature in studies of climate sensitivity. We also conclude that the European Centre for Medium-Range Weather Forecasts Reanalysis Interim can be used to obtain consistent temperature trends through the depth of the atmosphere, as they are consistent both with near-surface temperature trends and atmospheric temperature trends obtained from microwave sounding sensors.
Citation: Tellus A 2016, 68, 31503 [link to full manuscript]
The most significant aspect of this paper is incorporation of the ECMWF Interim Reanalysis dataset:
An independent approach is to compare temperature from operational analyses as done in numerical weather prediction. As was originally proposed by Bengtsson and Shukla (1988), this requires a dedicated data assimilation system to avoid systematic biases. In this study, we make use of recent re-analyses from the European Centre for Medium-Range Weather Forecasts (ECMWF). A main objective of this study is to explore whether the ECMWF Interim Reanalysis (ERAI) data set can reproduce credible temperature trends over a significant period of time.
However, global temperature records of the troposphere can only be used for limited time periods as data are only available globally with suitable accuracy since 1979 because of limited upper air observations from the Southern Hemisphere (SH) and the tropics before this date. Radiosonde data on their own have not been considered, because except when controlled and integrated into data assimilation, these data are subject to significant network and instrumental changes. For this reason, we do not intend to use observations from the free atmosphere directly but instead use re-analyses, though we will also use temperatures derived from microwave sounders for comparison.
As part of the re-analysis process, the observational data undergo an advanced data bias control. Satellite and aircraft data, assimilated by the re-analyses, have undergone systematic evaluation for the period after 1979, and we therefore believe that the re-analysis data can be considered as a reasonably independent robust source of tropospheric data
An alternative to using the tropospheric temperatures is to use sea surface temperatures (SSTs). The atmospheric temperature approximately 2 m above the ocean surface on average does not differ from the SST in a significant way, and temperature trends calculated over many years are expected to be the same as that of the SST.
The main results are summarized in the figures and tables below:
Fig. 1. Global mean surface temperature trends for the period 1979–2013 for ERAI, GISSTEMP and HadCRUT4.
Fig. 3. Global mean upper air temperature trends for the period 1979–2013 for ERAI, UAH and RSS.
From the Discussions and Conclusions:
The results show that surface air temperature changes over land are significantly larger than those over the oceans. This is to be expected because of the limited heat capacity of land surfaces compared to the ocean as was already demonstrated in early climate simulation studies. Other possibilities could be associated with changes in surface albedo such as reduced snow cover in winter that will act as a positive feedback factor. There is also the possibility that urbanisation effects have been underestimated as has been suggested from some studies. It could also be that ERAI has a systematic cold bias in the free atmosphere over land, but this is not very likely, as we do not see this over the oceans.
The surface temperature trend over land stands out. It is about twice as large as the temperature trend of the mid-troposphere. In the mid-troposphere, the trend is similar to that over the ocean. A possible explanation could be the drying out of the land surface leading to reduced fluxes of water vapour from the ground accompanied by a larger lapse rate.
Another area with a large warming trend is in the Arctic, most likely due to reduced sea ice cover in summer and autumn. The Arctic warming trend is most pronounced in ERAI with the largest values in the Russian sector. Such values are consequently not a direct effect of increasing greenhouse gases. It is most likely due to reduced sea ice in summer and autumn that in turn can be a secondary effect of climate warming but with no apparent warming response at upper levels.
It is also interesting to note that the trend in the lapse rate is also increasing, meaning that the temperature difference between the surface and the mid-troposphere is increasing during the period. This increase occurs over most land areas including higher latitudes, the Arctic and Antarctic regions. Weather situations in high latitudes with reduced inversions could add to such a development. Typical of the Arctic climate are pronounced boundary layer inversions that at low solar angles often persist during the day. A more detailed examination of the vertical structure of the trend shows that the near-surface temperature trend is approximately 2.5 times larger than that in the mid-troposphere. The difference is largest over ocean suggesting an additional contribution from reduced sea ice coverage. The reason for the enhanced warming of the boundary layer is not clear but is probably a combination of circulation changes and surface boundary conditions.
In the situation of a sustained warming or cooling of the climate caused by changes in radiative forcing (greenhouse gases, solar irradiation or volcanic eruptions), the temperature change through the troposphere should stay approximately the same for all vertical levels because of the strong vertical coupling due to fast atmospheric processes such as convection and constant large-scale horizontal mixing. In the case of net positive forcing, the tropospheric warming is expected to be slightly larger in the upper troposphere. This is because of the influence of the moist adiabatic lapse rate at higher tropospheric temperatures. However, with present minor temperature changes and data limitations, this cannot be uniquely determined.
The minimum temperature increase in the mid-troposphere, as suggested from ERAI and most clearly indicated over land, is somewhat puzzling. Comparison with GCM simulations, to be reported elsewhere, shows that this does not occur in model simulations. A possible explanation might be that model simulations have difficulties to handle the magnitude of different convective processes over land including the effect of limited horizontal resolution preventing a more realistic parameterization of mixed dry and moist convection.
The ERAI data are not free from systematic errors, which may affect trends. However, the large amount of different observations now used in the ERAI data assimilation suggest that the biases caused by changing observations over time are unlikely to corrupt the trend calculations and in any case not more than trends calculated from one set of specific observations.
We have highlighted in this article the problem with surface temperature trends over land. However, it is important to point out that SST trends are also problematic, particularly prior to the availability of reliable satellite observations. Available data sets have been composed by merging different types of observations in a partly subjective way that is not possible to fully reproduce.
We have also used available independent MSU data provided by UAH and RSS. There are minor differences between the two data sets as well as the corresponding values calculated from the vertical temperature profiles of ERAI. Comparing TLT of UAH and RSS with that calculated from ERAI shows a close agreement with the exception of the ocean TLT trends for UAH that are lower than the other two.
Tropospheric temperature trends are affected by gradual changes mainly in space observations both with respect to quality and coverage, but further improvements are expected with new re-analyses having more advanced bias control. We therefore strongly suggest that tropospheric temperature trends from re-analyses should replace surface temperature trends in future climate validation studies. If we use the temperature trend of the layer 700–400 hPa or any other similar measure, instead of the surface temperature trend, then this is probably a better representation of the global tropospheric temperature and presumably a more robust quantity to assess climate change.
JC reflections
In principle, the global reanalyses provide the best approach for developing truly global temperature datasets. The assimilation of multiple different types of data not only improves spatio-temporal coverage, but reduces the biases that would be associated with individual datasets. The reanalysis process provides a dynamically consistent way for providing a truly global dataset that does not rely on kriging, extrapolation or other infilling methods.
The ECMWF reanalysis agrees pretty well with the UAH and RSS tropospheric temperature analysis. While ECMWF uses the same basic datasets, it assimilates radiances from the satellite (rather than the temperature retrievals) as well as the radiosonde data. Hence this constitutes independent verification of the UAH/RSS analyses of tropospheric temperatures.
The differences between ECMWF and the land surface temperature datasets are striking.
ECMWF reanalyses currently don’t add much elucidation to the sea surface temperature debate, since sea surface temperatures are a boundary condition for the reanalysis process. ECMWF SST analysis relies heavily on satellite and buoy data in recent years.
The most interesting point (to me) in the paper is this:
Another area with a large warming trend is in the Arctic, most likely due to reduced sea ice cover in summer and autumn. The Arctic warming trend is most pronounced in ERAI with the largest values in the Russian sector. Such values are consequently not a direct effect of increasing greenhouse gases. It is most likely due to reduced sea ice in summer and autumn that in turn can be a secondary effect of climate warming but with no apparent warming response at upper levels.
This is a particularly interesting result, since the direct retrievals of upper level temperatures by RSS and UAH are not deemed to be reliable at high latitudes. The ECMWF reanalysis of upper level temperatures is arguably more reliable in the Arctic, but how reliable is a subject requires more investigation. The absence of upper level warming in the Arctic leads the authors to conclude that the surface warming trend in the Arctic is not a direct consequence of increasing greenhouse gases. Any GHG impact on Arctic sea ice is lost in the decadal scale variability of ocean heat advection and cloudiness changes (which may have a component from GHG warming), further supporting my contention (shared by the IPCC) that we do not have confidence in attributing a substantial fraction of the recent Arctic sea ice decline to GHG warming. This important point is lost in the public alarm surrounding the decline of the Arctic sea ice.
This paper is also important in that it establishes the ECMWF Interim Reanalysis as a useful data set for examining regional and temporal climate variability in recent decades. Of particular relevance are plans for ERA-CLIM2:
The five main objectives for the ERA‐CLIM2 project are:
- Production of an ensemble of 20th-century reanalyses at moderate spatial resolution, using a coupled atmosphere-ocean model, which will provide a consistent data set for atmosphere, land, ocean, cryosphere, including, for the first time, the carbon cycle across these domains;
- Production of a new state-of-the-art global reanalysis of the satellite era at improved spatial resolution, which will provide a climate monitoring capability with near-real time data updates;
- Further improvement of earth-system reanalysis capability by implementing a coherent research and development program in coupled data assimilation targeted for climate reanalysis;
- Continued improvement of observational data sets needed for reanalysis, in-situ as well as satellite-based, with a focus on temporal consistency and reduction of uncertainties in estimates of essential climate variables;
- Development of tools and resources for users to help assess uncertainties in reanalysis products.
IMO, this is where true progress lies in terms of understanding the global temperatures (not to mention the water and carbon budgets).Filed under: Data and observations