Historical climate data contains inhomogeneities, for example due to changes in the instrumentation or the surrounding. Removing these inhomogeneities to get more accurate estimates of how much the Earth has actually warmed is a really interesting problem. I love the statistical homogenization algorithms we use for this; I am a sucker for beautiful algorithms. As an observationalist it is great to see the
historical instruments, read how scientists understood their measurements better and designed new instruments to avoid errors.
Still for science it would be better if future climatologists had an easier task and could work with more accurate data. Let's design a climate-change-quality network that is a stable as we can humanly get it to study the ongoing changes in the climate.
Especially now that the climate is changing, it is important to accurately predict the climate for the coming season, year, decade and beyond at a regional and local scale. That is information (local) governments, agriculture and industry needs to plan, adapt, prepare and limit the societal damage of climate change.
Historian Sam White argues that the
hardship of the Little Ice Age in Europe is not just about cold, but also about the turbulent and unpredictable weather. Also the coming century much hardship can be avoided with better predictions. To improve decadal climate prediction of regional changes and to understand the changes in extreme weather we need much better measurements. For example, with a
homogenized radiosonde dataset, the improvements in the German decadal prediction system became much clearer than with the old dataset.
We are performing a unique experiment with the climate system and the experiment is far from over. It would also be scientifically unpardonable not to measure this ongoing change as well as we can. If your measurements are more accurate, you can see new things. Methodological improvements that lead to smaller uncertainties is one of the main factors that brings science forward.
A first step towards building a global climate reference network is agreeing on a concept. This modest proposal for preventing inhomogeneities due to poor observations from being a burden to future climatologists is hopefully a starting point for this discussion. Many other scientists are thinking about this. More formally there are the
Rapporteurs on Climate Observational Issues of the Commission for Climatology (CCl) of the World Meteorological Organization (WMO). One of their aims is to:
Advance specifications for Climate Reference Networks; produce a statement of guidance for creating climate observing networks or climate reference stations with aspects such as types of instruments, metadata, and siting;
Essential Climate Variables
A few weeks ago Han Dolman and colleagues wrote a call to action in Nature Goescience titled "
A post-Paris look at climate observations". They argue that while the political limits are defined for temperature, we need climate quality observations for all essential climate variables listed in the table below.
We need continuous and systematic climate observations of a well-thought-out set of indicators to monitor the targets of the Paris Agreement, and the data must be made available to all interested users.
I agree that we should measure much more than just temperature. It is quite a list, but we need that to
understand the changes in the climate system and to monitor the changes in the atmosphere, oceans, soil and biology we will need to adapt to. Not in this list, but important are biological changes, especially ecology needs support for
long-term observational programs, because they lack the institutional support the national weather services provide on the physical side.
Measuring multiple variables also helps in understanding measurement uncertainties. For instance, in case of temperature measurements, additional observations of insolation, wind speed, precipitation, soil temperature and albedo are helpful. The US Climate Reference Network measures this wind speed at the height of the instrument (and humans) rather than at the meteorologically typical height of 10 meter.
Because of my work, I am mainly thinking of the land surface stations, but we need a network for many more observations. Please let me know where the ideas do not fit to the other climate variables.
Table. List of the Essential Climate Variables; see original for footnotes.
| Domain | GCOS Essential Climate Variables |
| Atmospheric (over land, sea and ice) | Surface: Air temperature, Wind speed and direction, Water vapour, Pressure, Precipitation, Surface radiation budget.
Upper-air: Temperature, Wind speed and direction, Water vapour, Cloud properties, Earth radiation budget (including solar irradiance).
Composition: Carbon dioxide, Methane, and other long-lived greenhouse gases, Ozone and Aerosol, supported by their precursors. |
| Oceanic | Surface: Sea-surface temperature, Sea-surface salinity, Sea level, Sea state, Sea ice, Surface current, Ocean colour, Carbon dioxide partial pressure, Ocean acidity, Phytoplankton.
Sub-surface: Temperature, Salinity, Current, Nutrients, Carbon dioxide partial pressure, Ocean acidity, Oxygen, Tracers. |
| Terrestrial | River discharge, Water use, Groundwater, Lakes, Snow cover, Glaciers and ice caps, Ice sheets, Permafrost, Albedo, Land cover (including vegetation type), Fraction of absorbed photosynthetically active radiation, Leaf area index, Above-ground biomass, Soil carbon, Fire disturbance, Soil moisture. |
Comparable networks
There are comparable networks and initiatives, which likely shape how people think about a global climate reference network. Let me thus describe how they fit into the concept and where they are different.
There is the
Global Climate Observing System (
GCOS), which is mainly an undertaking of the World Meteorological Organization (WMO) and the Intergovernmental Oceanographic Commission (IOC). They observe the entire climate system; the idea of the above list of essential climate variables comes from them (Bojinski and colleagues, 2014). GOCS and its member organization are important for the coordination of the observations, for setting standard so that measurements can be compared and for defending the most important observational capabilities against government budget cuts.
Especially important from a climatological perspective is a new program to ask governments to
recognize centennial stations as part of the world heritage. If such long series are stopped or the station is forced to move, a unique source of information is destroyed or damaged forever. That is comparable to destroying ancient monuments.
A subset of the meteorological stations are designated as
GCOS Surface Network measuring temperature and precipitation. These stations have been selected for their length, quality and to cover all regions of the Earth. Its monthly data is automatically transferred to global databases.
National weather services normally take good care of their GCOS stations, but a global reference network would have much higher standards and also provide data at better temporal resolutions than monthly averages to be able to to study changes in extreme weather and weather variability.
There is already a global radiosonde
reference network, the
GCOS Reference Upper-Air Network (
GRUAN, Immler and colleagues, 2010). This network provides measurements with well characterized uncertainties and they make extensive parallel measurements when they transition from one radiosonde design to the next. No proprietary software is used to make sure it is know exactly what happened to the data.
Currently they have about 10 sites, a similar number is on the list to be certified and the plan is not make this a network of about 30 to 40 stations; see map below. Especially welcome would be partners to start a site in South America.
The observational system for the ocean
Argos is, as far as I can see, similar to GRUAN. It measures temperature and salinity (Roemmich and colleagues, 2009). If your floats meet the specifications of Argos, you can participate. Compared to land stations the measurement environment is wonderfully uniform. The instruments typically work a few years. Their life span is thus between a weather station and a one-way radiosonde ascent. This means that the instruments may deteriorate somewhat during their lifetimes, but maintenance problems are more important for weather stations.
A wonderful explanation of how Argos works for kids:
Argos has almost four thousand floats. They are working on a network with
spherical floats that can go deeper.
Finally there are a number of climate reference networks of land climate stations. The best known is probably the
US Climate Reference Network (
USCRN, Diamond and colleagues, 2013). It has has 131 stations. Every station has 3 identical
high quality instrument, so that measurement problems can be detected and the outlier attributed to a specific instrument. To find these problems quickly all data is relayed online and checked at their main office. Regular inspections are performed and everything is well documented.
The USCRN has selected new locations for its stations, which are expected to be free of human changes of the surroundings in the coming decades. This way it takes some time until the data becomes climatologically interesting, but they can already be compared with the normal network and this gives some confidence that its homogenized data is okay for the national mean; see below. The number of stations was sufficient to compute a national average in 2005/2006.
Other countries, such as Germany and the United Kingdom, have opted to make existing stations into a national climate reference network. The
UK Reference Climatological Stations (
RCS) have a long observational record spanning at least 30 years and their distribution aims to be representative of the major climatological areas, while the locations are unaffected by environmental changes such as urbanisation.
German Climate Reference Station which was founded in 1781 in Bavaria on the mountain Hohenpeißenberg. The kind of weather station photo, WUWT does not dare to show.In
Germany the climate reference network are
existing stations with a very long history. Originally they were the stations where conventional manual observations continued. Unfortunately, they will now also switch to automatic observations. Fortunately, after making a long parallel measurement to see what this does to the climate record*.
An Indian scientist proposes an
Indian Climate Reference Network of about 110 stations (Jain, 2015). His focus is on precipitation observations. While temperature is a good way to keep track on the changes, most of the impacts are likely due to changes in the water cycle and storms. Precipitation measurements have large errors; it is very hard to make precipitation measurements with an error below 5%. When these errors change, that produces important inhomogeneities. Such jumps in precipitation data are hard to remove with
relative statistical homogenization because the correlations between stations are low. If there is one meteorological parameters for which we need a reference network, it is precipitation.
Network of networks
For a surface station Global Climate Reference Network, the current US Climate Reference Network is a good template when it comes to the quality of the instrumentation, management and documentation.
A Global Climate Reference Network does not have to do the heavy lifting all alone. I would see it as the temporally stable backbone of the much larger climate observing system. We still have all the other observations that help to make sampling errors smaller and provide the regional information you need to study how energy and mass moves through the climate system (natural variability).
We should combine them in a smart way to benefit from the strengths of all networks.
The Global Climate Reference Network does not have to be large. If the aim is to compute a global mean temperature signal, we would need just as many samples as we would need to compute the US mean temperature signal. This is in the order of 100 stations. Thus on average, every country in the world would have one climate reference station.
The figure on the right from Jones (1994) compares the temperature signal from 172 selected stations &mdsh; 109 in the Northern Hemisphere. 63 in the Southern Hemisphere. &mdash with the temperature signal computed from all available stations. There is nearly no difference, especially with respect to the long term trend.
Callendar (1961) used 80 only stations, but his temperature reconstruction fits quite well to the modern reconstructions (Hawkins and Jones, 2013).
Beyond the global means
The number of samples/stations can be modest, but it is important that all climate regions of the world are sampled; some regions warm/change faster than others. It probably makes sense to have more stations in especially vulnerable regions, such as mountains, Greenland, Antarctica. We really need a stable network of buoys in the Arctic, where changes are fast and these changes also influence the weather in the mid-latitudes.
Crew members and scientists from the US Coast Guard icebreaker Healy haul a buoy across the sea ice during a deployment. In the lead an ice bear watcher and a rescue swimmer.
To study changes in precipitation we probably need more stations. Rare events contribute a lot to the mean precipitation rate. The threshold to get into the news seems to be the rain sum of a month falling in on one day. Enormous downpours below that level are not even newsworthy. This makes the precipitation data noisy.
To study changes in extreme events we need more samples and might need more stations as well. How much more depends on how strong the synergy between the reference network and the other networks is and thus how much the other networks could then be used to produce more samples. That question needs some computational work.
The idea to use 3 redundant instruments in the USCRN is something we should also use in the GCRN and I would propose to also to create clusters of 3 stations. That would make it possible to detect and correct inhomogeneities by making comparisons. Even in a reference network there may still be inhomogeneities due to changes in the surrounding or management (which were not noticed).
We should also carefully study whether is might be a problem to only use pristine locations. That could mean that the network is no longer representative for the entire world. We should probably include stations in agricultural regions, that is a large part of the surface and they may respond differently from natural regions. But agricultural practices (irrigation, plant types) will change.
Starting a new network at pristine locations has as disadvantage that it takes time until the network becomes valuable for climate change research. Thus I understand why Germany and the UK have opted to use locations where there are already long historical observations. Because we only need 100+ stations it may be possible to select existing locations from the 30 thousand stations we have that are and likely stay pristine in the coming century. If not, I would not compromise and use a new pristine location for the reference network.
Finally, when it comes to the number of stations, we probably have to take into account that no matter how much we try some stations will become unsuitable due to war, land-use change and many other unforeseen problems. Just look back a century and consider all the changes we experienced, the network should be robust against such changes for the next century.
Absolute values or changes
Argos (ocean) and GRUAN (upper air) do not specify the instruments, but set specification for the measurement uncertainties and their characterization. Instruments may thus change and this change has to be managed. In case of GRUAN they perform many launches with multiple instruments.
For a climate reference land station I would prefer to keep the instruments exactly the same design for the coming century.
To study changes in the climate climatologists look at the local changes (compute anomalies) and average those. We had a temperature increase of about 1°C since 1900 and are confident it is warming. This while the uncertainty in the average absolute temperature is of the same order of magnitude. Determining changes directly is easier than first estimating the absolute level and then look whether it is changing. By keeping the instruments the same, you can study changes more easily.
This is an extreme example, but how much thermometer screens weather and yellow before they are replaced depends on the material (and the climate). Even if we have better materials in the future, we'd better keep it the same for stable measurements.For GRUAN managing the change can solve most problems. Upper air measurements are hard; the sun is strong, the air is thin (bad ventilation) and the clouds and rain make the instruments wet. Because the instruments are only used once, they cannot be too expensive. On the other hand, each time starting with a freshly calibrated instrument makes the characterization of the uncertainties easier. Parallel measurements to manage changes are likely more reliable up in the air than at the surface where two instruments measuring side by side can legitimately measure a somewhat different climate, especially when it comes to precipitation, where undercatchment strongly depends on the local wind or for temperature when cold air flows at night hugging the orography.
Furthermore, land observations are used to study changes in extreme weather, not just the mean state of the atmosphere. The uncertainty of the rain rate depends on the rain rate itself. Strongly. Even in the laboratory and likely more outside where also the influence factors (wind, precipitation type) depend on the rain rate. I see no way to keep undercatchment the same without at least specifying the outside geometry of the gauge and wind shield in minute detail.
The situation for temperature may be less difficult with high-quality instruments, but is similar. When it comes to extremes also the response time (better: response function) of the instruments becomes important and how much out-time the instrument experiences, which is often related to severe weather. It will be difficult to design new instruments that have the same response functions and the same errors over the full range of values. It will also be difficult to characterize the uncertainties over the full range of values and velocity of changes.
Furthermore, the instruments of a land station are used for a long time while not being observed. Thus weather, flora, fauna and humans become error sources. Instruments which have the same specifications in the laboratory may thus still perform differently in the field. Rain gauges may be more or less prone to getting clogged by snow or insects, more or less attractive for drunks to pee in. Temperature screens may be more or less prone to be blocked by icing or for bees to build their nest in. Weather stations may be more or less attractive to curious polar bears.
This is not a black and white situation. It will depend on the quality of the instruments which route to prefer. In the extreme case of an error free measurement, there is no problem with replacing it with another error free instrument. Metrologists in the UK are building an instrument that acoustically measures the temperature of the air, without needing a thermometer, which should have the temperature of the air, but in practice never has. If after 2 or 3 generations of new instruments, they are really a lot better in 50 years and we would exchange them, that would still be a huge improvement of the current situation with an inhomogeneity every 15 to 20 years.
The software of GRUAN is all open source. So that when we understand the errors better in future, we know exactly what we did and can improve the estimates. In case we specify the instruments, that would mean that we need Open Hardware as well. The designs would need to be open and specified in detail. Simple materials should be used to be sure we can still obtain them in 2100. An instruments measuring humidity using the dewpoint of a mirror will be easier to build in 2100 than one using a special polymer film. These instruments can still be build by the usual companies.
If we keep the instrumentation of the reference network the same, the normal climate network, the GCOS network will likely have better equipment in 2100. We will discover many ways to make more accurate observations, to cut costs and make the management more easy. There is no way to stop progress for the entire network, which in 2100 may well have over 100 thousand stations. But I hope we can stop progress for a very small climate reference network of just 100 to 200 stations. We should not see the reference network as the top of hierarchy, but as the stable backbone that complements the other observations.
Organization
How do we make this happen? First the scientific community should agree on a concept and show how much the reference network would improve our understanding of the climatic changes in the 21st century. Hopefully this post is a step in this direction and there is an article in the works. Please add your thoughts in the comments.
With on average one reference station per country, it would be very inefficient if every country would manage its own station. Keeping the high metrological and documentation standards is an enormous task. Given that the network would be the same size as USCRN, the GCRN could in principle be managed by one global organization, like USCRN is managed by NOAA. It would, however, probably be more practical to have regional organizations for better communication with the national weather services and to reduce travel costs for maintenance and inspections.
Funding
The funding of a reference network should be
additional funding. Otherwise it will be a long hard struggle in every country involved to build a reference station. In developing countries the maintenance of one reference station may well exceed the budget of their current network. We already see that some meteorologists fear that the millennial stations program will hurt the rest of the observational network. Without additional funding, there will likely be quite some opposition and friction.
In the Paris climate treaty, the countries of the world have already pledged to support climate science to reduce costs and damages. We need to know how close we are to the 2°C limit as feedback to the political process and we need information on all other changes as well to assess the damages from climate change. Compared to the economic consequences of these decisions the costs of a climate reference network is peanuts.
Thus my suggestion would be to ask the
global climate negotiators to provide the necessary funding. If we go there, we should also ask the politicians to agree on the international sharing of all climate data.
Restrictions to data is holding climate research and climate services back. These are necessary to plan adaptation and to limit damages.
The World Meteorological Organization had its congress last year.
The directors of the national weather services have shown that they are not able to agree on the international sharing of data. For weather services selling data is often a large part of their budget. Thus the decision to share data internationally should be made by politicians who have the discretion to compensate these losses. In the light of the historical responsibility of the rich countries, I feel a global fund to support the meteorological networks in poor countries would be just. This would compensate them for the losses in data sales and would allow them to better protect themselves against severe weather and climate conditions.
Let's make sure that future climatologists can study the climate in much more detail.
Think of the children.
Related information
Hillary Rosner in the NYT on the global greenhouse gas reference network:
The Climate Lab That Sits Empty
Free our climate data - from Geneva to Paris
Congress of the World Meteorological Organization, free our climate data
Climate History Podcast with Dr. Sam White mainly on the little ice age
A post-Paris look at climate observations. Nature Geoscience (
manuscript)
Why raw temperatures show too little global warming
References
Bojinski, Stephan, Michel Verstraete, Thomas C. Peterson, Carolin Richter, Adrian Simmons and Michael Zemp, 2014: The Concept of Essential Climate Variables in Support of Climate Research, Applications, and Policy.
Journal of Climate, doi:
10.1175/BAMS-D-13-00047.1.
Callendar, Guy S., 1961: Temperature fluctuations and trends over the earth.
Quarterly Journal Royal Meteorological Society,
87, pp. 1–12. doi:
10.1002/qj.49708737102.
Diamond, Howard J., Thomas R. Karl, Michael A. Palecki, C. Bruce Baker, Jesse E. Bell, Ronald D. Leeper, David R. Easterling, Jay H. Lawrimore, Tilden P. Meyers, Michael R. Helfert, Grant Goodge, Peter W. Thorne, 2013: U.S. Climate Reference Network after One Decade of Operations: Status and Assessment.
Bulletin of the American Meteorological Society, doi:
10.1175/BAMS-D-12-00170.1.
Dolman, A. Johannes, Alan Belward, Stephen Briggs, Mark Dowell, Simon Eggleston, Katherine Hill, Carolin Richter and Adrian Simmons, 2016:
A post-Paris look at climate observations. Nature Geoscience,
9, September, doi:
10.1038/ngeo2785. (
manuscript)
Hawkins, Ed and Jones, Phil. D. 2013: On increasing global temperatures: 75 years after Callendar.
Quarterly Journal Royal Meteorological Society,
139, pp. 1961–1963, doi:
10.1002/qj.2178.
Immler, F.J., J. Dykema, T. Gardiner, D.N. Whiteman, P.W. Thorne, and H. Vömel, 2010:
Reference Quality Upper-Air Measurements: guidance for developing GRUAN data products. Atmospheric Measurement Techniques,
3, pp. 1217–1231, doi:
10.5194/amt-3-1217-2010.
Jain, Sharad Kumar, 2015:
Reference Climate and Water Data Networks for India. Journal of Hydrologic Engineering,
10.1061/(ASCE)HE.1943-5584.0001170, 02515001. (
Manuscript)
Jones, Phil D., 1994:
Hemispheric Surface Air Temperature Variations: A Reanalysis and an Update to 1993. Journal of Climate, doi:
10.1175/1520-0442(1994)007<1794:HSATVA>2.0.CO;2.
Pattantyús-Ábrahám, Margit and Wolfgang Steinbrecht, 2015: Temperature Trends over Germany from Homogenized Radiosonde Data.
Journal of Climate, doi:
10.1175/JCLI-D-14-00814.1.
Roemmich, D., G.C. Johnson, S. Riser, R. Davis, J. Gilson, W.B. Owens, S.L. Garzoli, C. Schmid, and M. Ignaszewski, 2009:
The Argo Program: Observing the global ocean with profiling floats. Oceanography,
22, p. 34–43, doi:
10.5670/oceanog.2009.36.
* The transition to automatic weather stations in Germany happened to have almost no influence on the annual means, contrary to what Klaus Hager and the German mitigation sceptical blog propagandise based on badly maltreated data.
** The idea to illustrate the importance of smaller uncertainties by showing two resolutions of the same photo comes from metrologist Michael de Podesta.
