⚒️ Our methodology is based on looking for weather conditions similar to those that caused the extreme event of interest with physics-informed machine-learning methodologies. We focus on the satellite era, namely the period since 1979, when widespread observations of climate variables from satellites have become available.
🗺️ The object studied (i.e. "the event") is a surface-pressure pattern over a certain region and averaged over a certain number of days, that has lead to a extreme weather conditions.
🛰️ We consider the early decades of the satellite era (1979–2000, "past") and the more recent decades (2001–2022, "present") separately. We use data from MSWX. We then compare how the selected weather conditions have changed between the two periods, and whether such changes are likely due to natural climate variability or anthropogenic climate change.
📊We use historical data and do not rely on numerical model simulations. This makes the framework rapid and reproducible. However, it also comes with disadvantages, since in some cases the extreme events result from very unusual weather situations that have not previously occurred, which hinders our analysis. Full details of our methodology are provided in the following freely-accessible peer-reviewed paper.
📄 Faranda, D., Bourdin, S., Ginesta, M., Krouma, M., Noyelle, R., Pons, F., Yiou, P., and Messori, G.: A climate-change attribution retrospective of some impactful weather extremes of 2021, Weather Clim. Dynam., 3, 1311–1340, 2022.
👇 Below is a graphical explanation of ClimaMeter. More detailed information is provided in the FAQs.
FAQs on the Methodology
1) How do you download the data?
2) How do you define your event of interest?
The object studied is a surface-pressure pattern over a certain region and averaged over a certain number of days, that has lead to a extreme weather conditions.
More specifically, after obtaining the data, we define the length of the event in terms of days when the largest impacts from extreme weather conditions (heat, wind, rain) have been reported and we select the geographical region to analyse based on the meteorological phenomena that caused the event. For example, in the case of a summer heatwave we would select a region including the high pressure causing the heatwave. The length of the event is specified in the title of the Climameter the region is exactly the one shown in the map.
3) Which data you download and how you pre-process them?
We download surface pressure, near-surface temperature, total precipitation and wind-speed data from MSWX with a daily time resolution. In order to account for the seasonal cycle in surface pressure and temperature data, we remove at each grid point and for each day the average of the pressure and temperature values for all the corresponding calendar days.
For surface pressure this removes the effect of varying surface elevation in space. Total precipitation and wind-speed data are not preprocessed. If the duration of the event is greater than one day, we perform a moving average of the length of the event duration on all datasets.
Note that there can be local discrepancies between local station observations and the gridded product that is used in ClimaMeter.
4) How do you find similar past events to the one you have defined?
The search of similar past events is based on defining analogues of the identified surface pressure patterns over the chosen spatio-temporal domain (see FAQ 2). We then divide the MSWX surface pressure data set into two periods: 1979-2000 ("past") and 2000-2021 ("present") each consisting of 21 years of daily data. We consider the first period as representative of a past world with a weaker anthropogenic influence on climate than the second period, which represents the present world affected by anthropogenic climate change. The analogues search is only performed on surface pressure data. Results reported for temperature, precipitation and wind-speed data are always associated with surface pressure analogues.
5) How many similar past events do you select?
For each period, we examine all daily surface pressure data and select the best 15 analogues (*), i.e. the data minimizing the Euclidean distance to the event itself. The number of 15 corresponds approximately to the smallest 1‰ Euclidean distances in each subset of our data. We tested the extraction of 10 to 20 analogues, without finding qualitatively important differences in our results. For the present period, as is customary in attribution studies, the event itself is excluded. In addition, we do not search for analogues within a window of 21 days centered on the date of the event (central date for events that last for longer than one day).
(*) For longer events such as floods occurring on an entire season, we may use a larger sample of analogues. If this is the case, this will be specified in the event description.
6) How do you take into account the role of the natural climate variability?
Here, we assume that 21 years is a long enough period to average out high-frequency interannual climate variability. To account for the possible influence of low-frequency modes of natural variability in explaining the differences between the two periods, we also consider the possible roles of the El Niño-Southern Oscillation (ENSO), the Atlantic Multidecadal Oscillation (AMO), and the Pacific Decadal Oscillation (PDO). We perform this analysis using monthly indices produced by NOAA/ERSSTv5. Data for ENSO and AMO are retrieved from the Royal Netherlands Meteorological Institute (KNMI) Climate Explorer, while the PDO time series is downloaded from the NOAA National Centers for Environmental Information (NCEI).
7) How do you determine the role of the natural variability VS human-induced climate change in the top-left gauge of the ClimaMeter?
The gauge can take values between 0% (pointing to the left) and 100% (pointing to the right). We look on whether the analogues occurred in a statistically significant different phase of El Niño-Southern Oscillation (ENSO), the Atlantic Multidecadal Oscillation (AMO), and/or the Pacific Decadal Oscillation (PDO). Whenever a statistically significant differences between phases of ENSO, AMO or PDO is observed we subtract 30% from the gauge value starting from 95%. We do not use 0% or 100% to acknowledge data and analysis uncertainties.
8) How do you determine the uniqueness of the event displayed in the top-right gauge of the ClimaMeter?
The gauge can take values between 0% (pointing to the left) and 100% (pointing to the right). We define several quantities to support our interpretation of analogue-based assignment, including the analogue quality Q, which is the average Euclidean distance of a given day from its closest analogues. If for both the periods, the value of Q for the event is below the 75th percentile of the distribution of Qs computed for all days in each period, we assign the gauge the value 5% (similar events have occurred in the past). If instead, for both periods the value of Q is below the 95th percentile, we assign 30%. If for one of the two periods this condition does not hold, we assign 60%. Finally, if the value of Q for the extreme event exceeds the 95th percentile for both periods, we assign the gauge the value 95% (the event is unique). We do not use 0% or 100% to acknowledge data and analysis uncertainties.
9) What do you mean by anomalies in the maps displayed the upper central panels of the ClimaMeter?
For pressure, we display the difference between the average surface pressure in the region for the duration of the event minus the average surface pressure for those same calendar days in the whole period 1979 to present. For example, if an extreme event happens on the 20th-27th July 2023, the pressure anomaly at a given location is the average pressure on the 20th to 27th July 2023 minus the average pressure on all 20th to 27th Julys from 1979 to 2023. For temperature, we do as for pressure. We do not compute anomalies for precipitation and windspeed because of their very noisy nature.
10) What do you mean by changes in the maps displayed in the lower central panels of the ClimaMeter?
The pressure map displays the difference in the average pressure for all analogues in the present period minus the average pressure for all analogues in the past period. The same is done for temperature. To determine significant changes between the two periods, we adopt a bootstrap procedure which consists of pooling the dates from the two periods together, randomly extracting 15 dates from this pool 100 times, creating the corresponding difference maps and marking as significant only grid point changes more than two standard deviations above or below the mean of the bootstrap sample.