Ensuring Accuracy in Surface Temperature Measurements

12/22/20220 min read

Earth surface temperatures are calculated using a combination of measurements by thermometers fixed at sites on land, and measurements taken by moving vessels (Jones and Wigley 2010: 59). To enable accurate estimates of surface temperature to be derived, three potential areas of error require scrutiny – the degree of data homogeneity, the effect of bias adjustments on the data, and the effect of changes in spatial coverage due to more sporadic sampling in the past (Jones and Wigley 2010: 61).

Before conducting statistical analysis of any set of data, it is often crucial to ascertain its homogeneity; that is, the extent to which the data has been taken from a single population, and the degree to which the data has been unaffected by external factors during the period of sampling (Columbia University). Homogeneity can also be understood as uniformity among sampling units, or, those units possessing common or the same traits or characteristics (CAAF). The more homogeneous a population is, the more credible the conclusions drawn from a sample within that population (CAAF). Time series data from land stations or marine measurements is considered sufficiently homogeneous if the only variations relate to regional differences in weather and climate (Jones and Wigley 2010: 66). Inhomogeneity can be precipitated by causes such as instrument exposure (discussed below), the urbanization effect (discussed below), and changes in methods and instrumentation during the period of sampling (Jones and Wigley 2010: 66). With greater spatial magnitude, adjustment factors are more prone to cancel and mitigate issues of inhomogeneity (Jones and Wigley 2010: 66).

Biases are homogeneity-related issues which may arise from three sources – developments in the methods of measuring sea surface temperature (“SST”), exposure of thermometers pre-1870, and the urbanization effect (Jones and Wigley 2010: 62).

Different methods of SST measurement lead to bias which needs to be corrected prior to applying any statistical technique or analysis. Before 1900, SST was measured by collecting seawater in wooden buckets (Jones and Wigley 2010: 62). After 1900, there was a shift toward canvas buckets, and in 1941 these began to be replaced by thermometers installed in vessel intake pipes (Jones and Wigley 2010: 62), except in Britain where canvas buckets survived until 1960 (Jones and Wigley 2010: 63). Because different heat-retaining characteristics of wood, canvas, and engine pipes influence the rate of evaporative cooling between the time of sampling and the time of measurement, temperature measures taken by bucket must be raised compared to engine intake accordingly (Jones and Wigley 2010: 63). As for continuity, it was often the case that, when satellites and buoys took over in the 1970s as techniques of measurement, no or inadequate parallel measurements were taken (Jones and Wigley 2010: 63).

Prior to the invention of white louvred screens in 1870, thermometers were located at north-wall locations in the northern hemisphere (Jones and Wigley 2010: 63-64). For land-based datasets, therefore, it is critical to establish the date on which screens were adopted at a specific site, the extent of sunlight exposure prior to such adoption, and whether parallel measurements were made at the time of the change (Jones and Wigley 2010: 64).

The urbanization effect refers to likely higher rates of warming at sites which have, throughout the period of sampling, evolved from towns to cities (Jones and Wigley 2010: 65). To determine the influence of urbanization on the temperatures measured at those sites, one must then also consider long-term temperature trends in other rural and urban locations in the vicinity (Jones and Wigley 2010: 64).

Spatial coverage may also affect the accuracy of surface temperature estimates because there were fewer measurement sites before 1880 in the southern hemisphere (Jones and Wigley 2010: 67). Generally, to ensure precise area averages on larger scales, one would require at least 100 sampling points which are suitably distanced from each other and undisturbed by non-climatic elements (Jones and Wigley 2010: 60).

The respective significance of the three main areas of concern (i.e. homogeneity, bias, and coverage), depends according to the scale involved (Jones and Wigley 2010: 61). On a global level, bias is the most important, followed by coverage, and finally homogeneity; on a local level, the order is coverage, homogeneity, and then bias (Jones and Wigley 2010: 61).

References: