Normals, normals, normals. A climate normal is an average of a particular quantity – temperature, rainfall, humidity, whatever – over an extended period of time, usually 30 years. If you’ve ever been on holiday to a foreign or unfamiliar place and wondered what to wear, you’ve probably made use of a climate normal for the city or region you were visiting – how warm is it in May? how much rain can I expect? do I need to pack an umbrella or oilskins? – without realising it. They have all manner of other uses too and are widely used throughout climate science/services/communications/etc/etc.
WMO Guidelines about the use of “normals” stipulate a double approach: 1991-2020 (or whatever the latest thirty-year period ending in a ‘0’ year is) for almost everything and 1961-1990 for long-term climate assessment. I’ve always found this a slightly confusing combination as it means 1991-2020 is to be used for climate monitoring and 1961-1990 is to be used for long-term climate monitoring. 1961-1990 also excludes a wide range of data sets from use for “long-term climate assessment” – reanalyses and satellite products don’t all extend back to the early 1960s but are vital sources of information. Actually… the use of 1961-1990 doesn’t exclude them because in practice people – being people – use whatever’s practical.
Preserving 1961-1990 has something going for it nonetheless. It’s helpful to have a fixed point, specified by convention and general agreement, against which change can be measured in perpetuity (or until, as the guidelines say, there is a compelling scientific reason to shift it). A fixed point might be desirable because a shifting baseline, particularly one that is as up to date as possible, can appear to erase global warming. When there are suddenly lots of blues in the maps where once there was red, the impression of warming is lessened (a point to which I shall return). It’s value is in consistency over time. Much effort has been invested across National Meteorological and Hydrological Services to gather and process normals for the period 1961-1990 as well as producing products (maps, datasets, reports, etc) that use this baseline. Updating these to a new baseline leads to extra work for NMHSs and other organisations.
Against this, there are of course other arguments for gradually deprecating 1961-1990 in addition to the aforementioned exclusion of useful data sets. I list a few below:
1. long-term climate assessment isn’t something that stands separately from all other uses of climate data. Using a separate baseline for this one particular use, and a different baseline for all other uses can and does lead to confusion and inconsistency. This defeats the purpose of standardisation. I’ve worked on things where we had to provide information on three different baselines (1850-1900, 1981-2010 and 1961-1990) just to manage all the different anticipated use cases and satisfy all interested parties (potential or actual). Which brings up another difficulty: it rapidly becomes unmanageably confusing, particularly when dealing with multiple data sets of varying length, which may or may not overlap the chosen baseline periods and one is then compelled to provide all other related information on the same range of baselines, which…. ugh.
2. For assessing global temperature change, the current favoured baseline is 1850-1900, which aligns (nearly) with the Paris Agreement’s use of change since “pre-industrial” conditions. One might question whether this is strictly a baseline, or whether it is merely a number that can be calculated from global temperature data sets. When I say that, I can hear the high-pitched sound of hairs splitting, but it is a period used in a very specific way that is rather different from other uses of normals. The use of 1850-1900 as a “baseline” has led to many people asking what the local temperature changes are relative to this period, often in the form of “has country X exceeded 1.5C yet?”. For some parts of the world we just don’t know because there are so few data. But it’s also not a meaningful question in relation to the Paris Agreement, which refers specifically to the global mean and also to long-term climate change. All of these words – global, long-term, change – are important and doing a lot of work behind the scenes that is often glossed over and which is also hard to apply to local temperature change. Even for global temperature, only two regularly updated data sets extend back to 1850. For other variables, 1850-1900 is a complete non-starter.
3. Earlier baselines are more uncertain due to sparser coverage, less advanced instrumentation, accumulation of homogenisation uncertainty, and so on. Expressing something as anomalies relative to an early and very uncertain baseline can therefore give the impression that recent anomalies and hence recent change are more uncertain or variable than they are. This is obviously true for a very early baseline like 1850-1900, but it is also true (at least for global temperature) for 1961-1990, a period that saw large and less-well understood changes in the way that ocean temperatures were measured.
4. Many of the potential problems of retiring the 1961-1990 baseline are problems of communication and/or related to the way that climate is talked about and presented. The first thing to note is that there is now a solid base of expertise that can be drawn on to help understand the difficulties and advantages of making such transitions. A number of institutions and NMHSs have made the shift between baselines.
In other cases, there are simple changes that can be made. One example, is the choice of charts to present large-scale averages of quantities. For example this plot from Copernicus…
Here, the baseline (1991-2020) is strongly emphasised. There is a horizontal line at zero and the colours and bars emphasise this at every single point. Changing the baseline on this graph gives a very different visual impression – more blues, less reds or vice versa – which is entirely due to the form of the graph. You can see this by flicking between baselines on one of the Copernicus Climate bulletins. In shifting the baseline, nothing important has changed, so a graph format that makes it appear otherwise might be considered a poor choice. In contrast, another plot from Copernicus largely avoids this problem by showing temperatures as a line graph instead:
In this case, the left and right-hand y-axes show anomalies relative to two different periods (1991-2020 on the left and 1850-1900 on the right), emphasising the fact that the choice of baseline is somewhat immaterial for the important content of the graph, which shows clearly the long-term change. (That’s not to say that the choice of baseline is completely immaterial in this case).
5. Which leads on to a related point, which is that baselines are, in a sense, arbitrary. There is nothing special about 1981-2010 (for example) other than it ends in a ‘0’ year and it was at one time designated as a standard by the WMO. That’s not say that such a designation is not useful or without any advantage (see above), simply that there was nothing special about the climate of 1981-2010 (or 1991-2020) that makes it better suited for this purpose than any other period. If communications around climate information are not provided with sufficient context, it can be very hard to extract any meaning from them. If I tell you that the average temperature anomaly in a country for 2021 was 1.3degC above normal, it’s probably hard for you to make sense of that. What is normal, you might ask? How does 2021 compare to other years? How much do temperatures vary? Is there a long-term change in temperature? What is it? How does 2021 relate to that? Context is key. If the baseline – the arbitrary baseline – is doing a lot of work in the comms then it might be time to reconsider what information is being shared and how. Note, here that 1850-1900 comes with its own context which is provided indirectly by the Paris Agreement. Even there, we should be careful about quoting individual years relative to 1850-1900 as the Agreement refers to long-term change and there are concerns that this can mislead.
6. As we shift from a situation where the focus was on convincing people that climate change exists, to one where the emphasis is more on adaptation and mitigation, the role of normals can and will change. The use of a single period will help with integration between products – monitoring reports, seasonal forecasts, decadal forecasts, etc. A more up-to-date normal period will be more relevant as things like the recent rate of change, the rapidity of changes and interactions with natural variability will become the focus. A more modern baseline is also generally thought to be more relatable. That is, people naturally compare current events to their memory of an extended, but recent, past and not to some now-remote earlier period.
7. Last, and certainly not least, as this is central to the practicality and fairness of the guidance: 1961-1990 excludes some countries that don’t have good long term digital records. When we surveyed countries during the development of guidance on calculating National Climate Monitoring Products (NCMPs), a larger number of countries (particularly those who didn’t already generate NCMPs) said they could make use of a more modern baseline than they could 1961-1990. The insistence on using 1961-1990 for assessing long-term change is liable to leave grey areas of “missing data” on the map where there needn’t be. The data are not necessarily missing, they may simply have been excluded by the choice of baseline. As has been noted in the context of attribution studies, it is often the areas with the least data that are most vulnerable to climate change.
So ends my lengthy ramble about normals. It’s a topic that creates a lot of discussion (to which, of course, I have just added) and little in the way of solid conclusions. I’ve been following the process of creating practical guidance for over ten years and in that time I have spent many, many hours – days even – discussing, pondering, reading and writing about normals. To me, it seems that the chief value in specifying a standard, is to maximise consistency between different parts of a larger system – in this case a sprawling ecosystem of climate services and products – so that they all work together neatly without extensive interactions. To that end, preserving a special use case for 1961-1990, injects an inconsistency into proceedings.
The current guidelines require a “scientific” reason for effecting a change. The above points may, or may not constitute “scientific” reasons for simplifying the dual-normal system to a single-normal system. As the original reasons for preserving 1961-1990 were not especially scientific, it seems an unfair hurdle to have to cross.
The practicalities, however, cannot be wholly ignored. Guidance is only useful if it can be followed. I do think that some of the practical comms-related reasons for maintaining 1961-1990 can be managed otherwise, by thinking more carefully about what it is one actually wants to show. What is harder, perhaps, to countenance, in so far as guidelines are binding, is the extra work then required to update existing products.
To this end – the difficulty associated with updating products – some thought must be given. What changes need to be made to make this a simpler task? How can the generation of products against different baselines be automated or otherwise eased? In the software that was produced to calculate NCMPs, the baseline was user specified (though it defaulted at the time to 1981-2010). Nonetheless, many people faced with the possibility of running the software expressed something like dread at the thought of updating to a modern baseline. So, another question, why the dread?
Some other thoughts: what are the “scientific” reasons envisioned by the drafters of the guidance? What software is out there to make the re-calculation of baselines a simpler task? What guidance can be given to help with communication aspects? Are there libraries for making better plots?
More questions than answers, as ever.
Ian Harris said:
Another issue is availability of data. For observations, (mean / min / max temperature, precip, humidity, cloud / sun, etc), the optimum cover is found in 1961-1990. A certain well-used multivariate dataset would lose cover almost everywhere if a new normalisation period were imposed.
A fair point.
To what extent is 1961-1990 the period with the best coverage due to the fact that 1961-1990 has been used for so long now? Is that actually the period with the largest number of stations open simultaneously? Or is it due to things like limitations around data sharing and incomplete digitization of records?
Ian Harris said:
It’s the latter – station numbers peaked in the late C20th, then declined as funding was lost. This is a worldwide problem – see Fig. 3 of Harris et al (2014).
Ian Harris said:
I guess that’s part of the overall trade off: what are the relative utilities of having a modern period which is more representative of current conditions and having a better estimate for an earlier period.