Statistical Learning Techniques Help Extract Climate Change Fingerprint

In October 2019, Utah-based weather scientists measured the lowest temperature of −37.1 °C in the United States (except for Alaska). This temperature was the lowest to be ever recorded in the month of October.

A record cold day does not call climate change into question. On the contrary: in daily weather, climate researchers have been able to detect the fingerprint of climate change since 2012. Image Credit: Colourbox.

The earlier low-temperature record for the same month was −35 °C, which made the public wonder about the trends of climate change.

So far, climate researchers have asserted that climate is different from weather. The former is what people anticipate in the long term, while the latter is what people experience in the short term. As local weather conditions are extremely variable, it can be quite cold in one place for a short time in spite of long-term global warming. In brief, the local weather variations conceal the long-term trends in the global climate.

A Paradigm Shift

But now a research team headed by Reto Knutti, an ETH professor, has carried out a recent study on temperature measurements as well as models. The researchers inferred that the paradigm that weather is not the same as climate is no longer relevant in that form.

The climate signal—in other words, the long-term warming trend—can truly be detected in day-to-day weather data, like humidity and surface air temperature, if worldwide spatial patterns are taken into consideration, stated the scientists.

To put it in simpler terms, this implies that—in spite of global warming—the United States may experience a record low temperature in the month of October. However, if it is concurrently warmer than average in other areas, such a deviation is virtually fully eliminated.

Uncovering the climate change signal in daily weather conditions calls for a global perspective, not a regional one.

Sebastian Sippel, Study Lead Author, ETH Zurich

Sippel is a postdoc working in Knutti’s research team. The study was recently published in the Nature Climate Change journal.

Extracting the Fingerprint of Climate Change

To identify the climate signal in day-to-day weather records, Sippel and his collaborators employed statistical learning methods to integrate simulations with climate models as well as data obtained from measuring stations.

Such statistical learning methods are capable of deriving a climate change “fingerprint” from the combination of temperatures at different areas and the ratio of predicted warming and variability. Systematic assessment of the model simulations can help detect the climate fingerprint in the worldwide measurement information on any given day since the period of spring 2012.

Upon comparing the variability of both global and local everyday mean temperatures, it was found why the global standpoint is crucial. While locally quantified day-to-day mean temperatures can vary broadly (even after the removal of a seasonal cycle), the global day-to-day mean values demonstrate an extremely narrow range.

The two distributions, or bell curves, will hardly overlap if the distribution of global day-to-day mean values from 1951 to 1980 is subsequently compared with the global day-to-day mean values from 2009 to 2018.

Therefore, the climate signal is important in the global values but masked in the local values, as the distribution of day-to-day mean values overlaps rather significantly in the two periods, that is, from 1951 to 1980 and from 2009 to 2018.

Application to the Hydrological Cycle

The study outcomes can have wide implications for climate science.

Weather at the global level carries important information about climate. This information could, for example, be used for further studies that quantify changes in the probability of extreme weather events, such as regional cold spells.

Reto Knutti, Professor, ETH Zurich

Knutti continued “These studies are based on model calculations, and our approach could then provide a global context of the climate change fingerprint in observations made during regional cold spells of this kind. This gives rise to new opportunities for the communication of regional weather events against the backdrop of global warming.”

The research is the outcome of an association between the Swiss Data Science Center (SDSC) and ETH scientists. SDSC is jointly operated by ETH Zurich with its sister university EPFL.

The current study underlines how useful data science methods are in clarifying environmental questions, and the SDSC is of great use in this.

Reto Knutti, Professor, ETH Zurich

Data science techniques enable scientists to show the power of the human “fingerprint” and also allow them to demonstrate where in the world climate change is specifically clear and discernable at an initial stage.

This fact is extremely significant in the hydrological cycle, where there are extremely huge natural variations from day to day and year to year.

In future, we should therefore be able to pick out human-induced patterns and trends in other more complex measurement parameters, such as precipitation, that are hard to detect using traditional statistics,” concluded Knutti.


Tell Us What You Think

Do you have a review, update or anything you would like to add to this news story?

Leave your feedback
Your comment type