Sources of Uncertainty in Global Climate Change Estimates

Bit of a technical post, but here is a chart showing the sources of uncertainty in one of the modern estimates of global climate change since 1850.

For most of the last 170 years, ocean temperature uncertainties are actually more impactful that land uncertainties. pic.twitter.com/BjAriML5bF

— Dr. Robert Rohde (@RARohde) March 3, 2023

When we talk about temperature uncertainties, we are talking about the little error bars added to charts like this, based on a scientific assessment of the "known unknowns".

The assessed uncertainties are much smaller than the long-term trend. pic.twitter.com/qnXKUkDecy

— Dr. Robert Rohde (@RARohde) March 3, 2023

Assessments of uncertainty are complicated, and even labelling them is rarely black & white.

One person's "measurement noise" might be assessed by someone else as related to "systematic biases".

— Dr. Robert Rohde (@RARohde) March 3, 2023

Similarly, assessments of uncertainty due to incomplete coverage depend on what interpolation methods may have been used, and often create trade-offs.

Interpolations that can reduce coverage biases often increase the impact of measurement & systematic uncertainties.

— Dr. Robert Rohde (@RARohde) March 3, 2023

Modern weather observing networks are very extensive, but it hasn't always been that way.

Past gaps give rise to coverage uncertainties when we try to estimate how much variation might be missed in regions that weren't observed at the time. pic.twitter.com/EDilr2rP98

— Dr. Robert Rohde (@RARohde) March 3, 2023

Typically, coverage uncertainty is assessed by taking modern, spatially-complete weather fields and masking them to resemble past spatially-incomplete observations.

The difference between the original average and the masked average gives an estimate of the coverage uncertainty.

— Dr. Robert Rohde (@RARohde) March 3, 2023

In addition to coverage gaps, systematic biases are a major concern.

Systematic biases arise when the methods and instruments used to measure temperature change over time in such a way past data and present data are no longer completely comparable.

— Dr. Robert Rohde (@RARohde) March 3, 2023

For example, originally ocean-going ships measured water temperature by hauling buckets of water onto the deck.

Later, they started measuring engine intake water.

Engine intake water tends come from a bit below the surface, and hence is slightly cooler than the previous method.

— Dr. Robert Rohde (@RARohde) March 3, 2023

Other examples of systematic biases include changes in the time of measurement or location of a weather station, replacing liquid-in-glass thermometers with thermistors, development of an urban heat island, etc.

— Dr. Robert Rohde (@RARohde) March 3, 2023

Many systematic biases are partially correctable.

For example, by comparing the bucket method to the engine intake method, we can estimate the typical difference.

Such corrections reduce error, but questions about the magnitude of such effects gives rise to bias uncertainties.

— Dr. Robert Rohde (@RARohde) March 3, 2023

Bias uncertainties associated with changing methods of ocean measurement are often the largest uncertainty in the global average.

This is particular pronounced during World War II, when few ships collected data and those that did often used non-standard or unclear techniques. pic.twitter.com/DBsWk7thh5

— Dr. Robert Rohde (@RARohde) March 3, 2023

As it turns out, ship captains worried about dodging German U-Boats, didn't really bother to collect weather data. Shocking, I know.

During the war years, reported ocean data volumes dropped 90%.

— Dr. Robert Rohde (@RARohde) March 3, 2023

The other problem with ocean biases is that they can be harder to detect and correct for.

On land, we can often detect weather station biases by comparing one station to its neighbors over many years.

However, ships don't stay still and many don't have long time series.

— Dr. Robert Rohde (@RARohde) March 3, 2023

And, of course, the Earth's surface is more than 2/3 water.

So, any uncertainty in the ocean average necessarily has a proportionally larger impact on the global average uncertainty.

— Dr. Robert Rohde (@RARohde) March 3, 2023