Recent Global Warming Estimates Vary More Than Twofold

Estimates of how quickly Earth has warmed since 2011 differ by more than a factor of two, and the method used to measure the trend affects projections for the decades ahead. A recent study places the faster rate among its estimates; a comparison using three statistical approaches shows why that result should be read alongside substantial uncertainty.

The study, described in the paper, identifies 2011 as a point when the warming rate increased. It estimates separate trends for 1970–2010 and 2011–2025, using the Theil-Sen method. That approach calculates a slope from pairwise comparisons in the data and is less sensitive to individual unusually warm or cool years than a standard least-squares fit.

There’s a consequential modeling choice in fitting those two periods separately: the estimated temperature line can jump at 2011 as well as change slope. The jump is a feature of the fitted model, not a claim that Earth’s temperature rose abruptly at that moment. When researchers choose a changepoint and fit each side independently, the two estimated lines aren’t required to meet.

Three ways to estimate the recent trend

A comparison using HadCRU yearly temperature averages tests three ways to describe the period from 2011 onward. Least-squares regression fits one straight line to that interval. Theil-Sen estimates its slope using the same method as the study. A piecewise linear fit, or PLF, allows the slope to change at a chosen point while requiring the line to remain continuous there.

That continuity requirement matters. The least-squares and Theil-Sen lines can reflect a jump at the changepoint; the PLF cannot. The fits therefore answer related but not identical questions about how to represent the data. In the comparison, PLF gives the smallest warming-rate estimate, while Theil-Sen gives the largest.

The accompanying charts show the yearly HadCRU averages and fitted lines, first across the broader record and then in a closer view of the period since 2011. A third chart compares estimated rates and 95% confidence ranges across four temperature data sets.

Why the range matters

Across all four data sets, the ordering is consistent: Theil-Sen produces the highest rate estimates and PLF the lowest. The PLF confidence ranges are narrower in this comparison. Requiring continuity constrains the model, which can make its estimated rate more precise under that setup. A narrower confidence range, however, doesn’t account for every source of uncertainty in choosing the time period, changepoint or model form.

The estimates span roughly 0.025°C to 0.06°C per year, equivalent to 0.25°C to 0.6°C per decade. That is a wide spread for a trend used to discuss when global temperature levels might be reached. The article argues that even the lower rate would take warming beyond 2°C by 2050, while the upper rate would imply passing 2.5°C by then. These are conditional projections based on extending the estimated rates; they should not be mistaken for precise forecasts that account for every future influence on temperatures.

Recent warming also varies from year to year, so short-period trend estimates can shift when the start or end year changes. The method comparison makes that sensitivity visible: the data support a faster recent trend, but the precise size depends in part on how the line is fitted. The reported confidence ranges describe uncertainty within each method and data set; they don’t resolve all the choices behind the analysis.

For readers tracking climate impacts, the practical signal is that a faster sustained rate would bring temperature thresholds closer, while the gap between the low and high estimates leaves meaningful uncertainty about timing. The key evidence to watch is whether future observations continue to support the post-2011 increase, and how estimates compare when researchers use different trend methods and temperature records.