Reducing uncertainties in the historical record of Earth’s surface temperature can improve scientists’ ability to understand and explain changes in the climate over the past 150 years. This is particularly important for the early part of the twentieth century, because the cause of observed warming at that time remains fiercely debated1. Writing in Nature, Chan et al.2 demonstrate an innovative approach to account for differences in how sea surface temperature was measured in the early twentieth century. Their results suggest modestly less warming in the North Atlantic Ocean and substantially greater warming in the North Pacific Ocean during the period from 1908 to 1941, relative to previous estimates. Such findings indicate that intrinsic climate variability has a smaller impact on regional warming rates than was thought.
Improving historical temperature estimates has long been a key focus for climate researchers. Until the past few decades, most temperature measurements on both land and ocean were not aimed at detecting long-term climate changes. Rather, they were mainly intended to document average climate conditions or were for shorter-term meteorological purposes3. Adjustments to measurement methods that introduced biases of a few tenths of a degree Celsius were common. Although these biases were of little concern at the time, they become substantially more relevant when trying to detect long-term changes in global temperature of about 1 °C over the past 150 years.
The record of global surface temperature is produced by combining measurements of sea surface temperature (SST) with measurements of air temperature over land and ice. The largest remaining uncertainties in the global temperature record are associated with the SST estimates. Specifically, changes in observational instrumentation and techniques over time, coupled with patchy metadata (information about data) and sparse sampling in some regions complicate the interpretation of the historical record4.
Initially, SST estimates were made using wooden buckets that were thrown over the sides of ships, filled with water and hauled up. The temperature of the water in the buckets was then measured using a thermometer. While the buckets were being hoisted up, evaporative cooling and exposure to ambient conditions would often reduce the temperature of the water by a few tenths of a degree Celsius.
This bias was exacerbated by a transition to poorly insulated canvas buckets in the late nineteenth century, and these buckets continued to be the main means of SST measurement until the period of the Second World War. Accounting for the cold bias in bucket measurements is the single largest adjustment to the ocean (and global) temperature record. Without the adjustment, the estimated rate of ocean warming from 1850 to the present would be about 30% higher5.
A bucket measurement can be affected by a wide range of factors. These include the height of the ship, the composition and size of the bucket, how long it remains in the sea, whether the water is stirred before measurement and how long the thermometer is left in the water. Little of this information was recorded in a form that has survived to the present day. As a result, researchers have often had to inaccurately treat many bucket measurements as having the same magnitude of bias.
Chan and colleagues found a clever way to tackle this problem. They looked at the difference between SST measurements that were made within 300 kilometres and 2 days of one another, producing a data set of 6 million measurement pairs between 1908 and 1941. Ships were grouped by national origin, on the assumption that ships from the same country would tend to have similar measurement practices at any given time. The authors found sizeable offsets in SST estimates between ship groups, ranging from −0.3 °C to +0.6 °C.
Digging further into these differences, Chan et al. realized that measurements from Japanese ships in the North Pacific suddenly became about 0.35 °C cooler after 1930 when compared with measurements from other countries. This change was caused by the Japanese switching from recording temperatures in whole-degrees Fahrenheit to taking readings in degrees Celsius and then dropping any numbers after the decimal point. The authors identified a similarly large change in the North Atlantic that is associated with German readings, but the cause of this change is less clear.
Chan and colleagues’ results suggest that scientists have been overestimating warming in the North Atlantic and substantially underestimating warming in the North Pacific during the early twentieth century because of not fully accounting for biases in bucket measurements (Fig. 1). These findings bring the difference in estimated warming between the two regions in line with projections from climate models. However, there are still large differences between models and observations in the overall rate of global ocean warming during this period.
The authors’ approach of comparing groups of proximate-ship measurements is conceptually similar to that used in identifying problems in the land temperature record, whereby each weather station is compared with its neighbours to find and remove localized biases6. The method offers an innovative solution to the lack of good ship metadata during the early twentieth century and provides a major advance in our understanding of historical ocean measurements.
This study, and recent major updates to the SST record at the UK Met Office’s Hadley Centre7, provide a useful reminder that large systematic biases might remain in our observational temperature records. Improved quantification of these biases is still a key technical challenge for researchers, and will help to address questions about the performance of climate-model simulations of the past and the role of intrinsic climate variability in historical temperature change.
Nature 571, 328-329 (2019)