The RSS Climate Change Task Force has produced a series of explainers to help people understand the statistics and data underpinning our understanding of climate change. This explainer looks at when climate data needs to be adjusted, and how it is done.
1. What is a climate data record?
Climate data records are compiled by aggregating daily weather observations collected worldwide by national weather services. Historically, air temperature recordings were collected at locations where people lived or worked and sea temperatures from where they sailed, but broad geographic coverage is important and data now comes from varied sources, including tens of thousands of land weather stations and balloons, ships and buoys, radar and satellites. Spatial averaging ensures that areas of equal size contribute equally to the global average. Since locations that are closer together geographically tend to produce observations that are similar, location dispersion means that fewer observations are required overall to estimate the global mean temperature.
This explainer focuses on air temperature recordings over land, but the issues described similarly apply to sea surface and marine air temperatures, precipitation, snow cover and sea ice.
2. When does the climate record need to be adjusted?
Creating consistent climate records over many decades presents a range of challenges. It requires the digitisation of old paper-based records, as well as the identification and adjustment of inconsistencies created by changes in weather station location, changes in surrounding land use, technology development and random errors. Such changes can produce sudden shifts in the data that affect interpretation of the climate record over time, although changes can also be more gradual. If left uncorrected, these artefacts can lead to spurious warming or cooling trends in temperature data at individual stations.
The main reasons to adjust observed data include:
-
Location changes: Changes in station location and the height of the instruments above ground can lead to distributional shifts in annual mean temperature.
-
Increased urbanisation: Rapid increases in urbanisation can lead to upward shifts in the annual mean temperature measured at urban locations, warranting a location move. This would be so that the direct effect of increased urbanisation on local air temperature could be separated from the broader effects of urbanisation on the climate.
-
Instrument upgrades and failure: External damage, technological obsolescence, and instrument failure can all lead to data inconsistencies.
-
Observation schedule: Changes to the observation schedules and practices can produce shifts in annual mean temperature. While automation and the introduction of international standards have reduced the need for adjustments, archived data has been found to include different formulae for calculating certain mean temperatures.
3. How is a climate record adjusted?
To construct a complete climate record, adjustments may be needed to correct for instrument changes, missing data and measurement errors. The process of such adjustments seeks to answer important questions such as: What would the global long-term temperature trend look like if all observations were recorded at the current station with the current available technology?
For instance, if the location of a weather station moves, maintaining consistent long-term records for climate monitoring means an adjustment is required to account for the different climate conditions between the old and new location. The aim of this adjustment is to create a continuous record for that station despite the physical change in location. The adjusted data does not replace the old location record. Instead, it is flagged and appended to the observed record for the new location so that it creates a continuous long record for that station.
Figure 1 depicts a temperature record for a weather station at Port Lincoln, South Australia. The location of this station was moved from the town near the coast to an airport located to the north and several miles inland. Average daily temperatures (left panel) at the airport are lower than they are at the former location in town. Simply appending the new data to the old data would introduce bias to the temperature record and create what is described as a discontinuity; that is, a break or sudden shift in the time series that is an artefact caused by the location change. Without a suitable adjustment to the record, an increased frequency of low temperatures would result that would not accurately represent reality. The method to resolve this discontinuity is referred to as data homogenisation.
|
 Mean winter maximum daily temperatures at the two nearby locations, Port Lincoln (township vs airport).
|
 Homogenised temperature record: adjusted mean winter temperatures at airport, Port Lincoln.
|
Figure 1. Port Lincoln, South Australia temperature record (1981-2015) by location
Standard practice is to conduct a period of overlapping observations (for example, over several years) when a station location changes. Observations are recorded at both the old and the new locations to allow for comparisons between the two. Sometimes it is not possible to record overlapping observations when locations move and, in these cases, adjustments in datasets are made using data from several closely correlated reference stations in a region. This is done by first matching the old station location to a reference station and then matching the reference station to the new station location. Normally, a combination of at least 10 reference stations is used in this process.
4. What about adjusting for differences in the variability of the data at the different locations?
When a station has changed location, simply adjusting monthly or annual mean values is usually not sufficient to produce a consistent set of data, especially for higher-order statistical properties (such as the behaviour of extremes). For instance, scaling average temperatures alone for a station that has been moved inland from the coast will fail to capture the likely increased number of both very hot days and very cold days, since temperature extremes are often more variable when we move further inland (away from the ocean, and often at elevation). For this, we require a method that ‘homogenises’ data across the full range of the distribution of temperatures from season to season. One way to do this is to match the percentiles of temperature ranges across the distribution of temperatures at each location. This is referred to as percentile-matching.
To perform percentile matching, we first calculate daily temperature anomalies (i.e. differences) from the mean temperature over a set period, say 1961-1990. We then identify a suitable overlap period between the two locations and pair the observations. For each month of the year, we compare daily anomalies at the old and new locations and measure the range for each percentile of the temperature distribution. The distribution of temperatures at the new location is matched to the distribution at the old location based on their distance from each other. Values at the old location are converted to equivalent values at the new location to produce a composite record, with the new location taking precedence where data exist at both locations. The percentile mapping approach is illustrated in Figure 2 below.
Figure 2. Illustration of the percentile matching adjustment process from an old location to a new location with different temperature profiles through the year.
The use of the percentile matching algorithm for adjustment of temperatures does not greatly alter the effectiveness of homogenisation for mean temperatures compared with methods based on uniform monthly or annual adjustment but does improve the representation of extremes.
5. Will further adjustments be required?
There is inherent uncertainty associated with homogenisation adjustments, since no one method can be considered as the single version of truth. Care is taken to ensure estimates do not produce materially different results. The most important outcome is that the adjusted temperature dataset produces comparable results for temperature trends that appropriately represent the global mean.
Importantly, national meteorological agencies do not alter the original temperature data measured at individual stations. Rather, they create additional long, continuous and consistent (homogeneous) records for locations. The new data series are a complement to, not a replacement of, the original data. All adjustments should be identified and described when deployed.
In practice, very few century-long records exist that do not require adjustment, but adjustment of historical data is not unusual in other research areas - such as demography, biodiversity, population health, and economics. These areas also make adjustments to their historical records using a variety of methods to correct for errors and data mismatches, thus improving the reliability of