This page presents a high-level insight into extreme events and how extreme events differ from mean climate. Extremes are often related to different physical processes than those that govern long-term means. While an average change in precipitation is primarily due to circulation changes, extremes are much more sensitive to the thermodynamic state and conditions during specific days. Therefore, it is important to compare and contrast trends and projections in means against those of rare events.
Extremes only occur in a conjunction of several preconditions. For example, extreme rainfall requires maximized (“potential”) moisture transport into the region, high temperatures (or large temperature gradients) and significant instability of the atmosphere. An alignment of these “ingredients” is relatively rare. Under climate change, however, some of these conditions might see a systematic increase in occurrence, which is particularly true for temperatures across the globe. If that one condition – higher temperatures – is more often fulfilled, then the chance for a combined occurrence can also increase. Warmer temperatures are especially important for precipitation because the Clausius-Clapeyron-Relationship dictates that for every 1ºC of increased air temperature, that air’s potential to carry moisture increases by 7%. Thus, the warmer the air, the much more moisture it “can” carry, and therefore if rain were to form, much more water could be tapped into.
Where exactly the most extreme precipitation then might happen is also somewhat uncertain as current local conditions over a broader region can dictate the dynamical process of triggering an event, although sometimes physical settings (e.g., topography) can lead to areas with higher likelihood of occurrence. Overall, extreme events have to be seen as requiring a set of pre-conditions tied with a probabilistic element of initiation. This is why extreme thunderstorms can affect one place, while a few kilometers away there is hardly any precipitation registered.
In summary: (1) Extreme precipitation events might show different signs and commonly larger magnitudes of change when compared to mean precipitation. (2) In a warmer world, the potential of air to carry moisture goes up exponentially, and thus the potential for heavier precipitation goes up. This means that intense events will likely recur more frequently, which can negatively affect the flooding risk. Only in areas where the occurrence of precipitation goes down significantly can the trend towards heavier rainfall be overcome and return periods of large events increase rather than decrease.
This page is designed to offer the user high-level insight into how CMIP6 models represent extreme events at global and country scales. The visualizations present different perspectives on historical conditions as well as future scenarios to enable understanding of shifts in extreme events. The baseline climate (Historical Period, 1985-2014, centered on 2000) can be compared with future time periods and scenarios, centered on 2025, 2050, 2075 as well as 2085 (using data to the end of the century). Please note, the presented extreme indicators offer qualitative projection results, which directly reflect global model output, and should not be mistaken for location-specific (“station-level”) extremes.
I – How Extremes Differ from Mean
Extreme events reflect rare (weather) events and therefore represent different characteristics of climate than the long-term means. Compare the maps to recognize potentially contrasting trends and different magnitudes of change between mean precipitation (top left) and the largest event during the period (top right) with the extreme statistics indicators for different return periods for the globe (bottom left) and country (bottom right).
Historical: Return Levels (precipitation amounts shown in "mm") are available for a select list of recurrence intervals (Return Periods). Equally, the spatial pattern of fixed precipitation magnitudes (return levels) can be inspected with respect to their corresponding Return Periods (shown in “years”). Sometimes the inverse of Return Period is needed as it represents the “Annual Exceedance Probability” (how many events can be expected per year).
Projections: The return levels established over the historical period are then used to determine possible changes in their associated recurrence interval (Return Period in “years”). Colors show, for example, if an event that is historically a 20-yr event (20-yr return level during historical period) is happening more (green) or less frequently (brown) in the future. Therefore, projections are shown as future return periods (in “years”) that indicate to the present-day reference. That potential shift in return period can also be formulated as a change in Future Annual Exceedance Probability. This product, shown as a factor, offers a view of how much more or less frequent an event might become.
Larger rainfall events are sometimes simulated in the tropics where large scale monsoonal rains and tropical cyclone-like processes lead to maximum rainfalls in global models. However, the global model data can not always properly resolve small-scale convective processes, and therefore there is an upper limit of magnitude of "single" events that can be simulated.
II – Distribution to Return Levels / Periods
Analyzing extreme events requires specialized statistical tools that make use of Extreme Value Theory (EVT). Means and standard deviations are not suited to describe the behavior at the tails (rare occurrence) of the data distribution. EVT employs flexible functions that estimate the frequency of rare, high magnitude events. The following graphs illustrate the process of (1) fitting a continuous generalized EV-function to the data as represented in form of a histogram (number of sample in different magnitudes bins), and (2) a comparison of the quantiles between empirical (actual model values) and their function-fitted representation (Density Plot). The closer the functional representation can follow the data, the better is the fit throughout the range of values.
While hard to see in the histogram, the QQ-Plot highlights the challenges towards the highest, most extreme values. There, it is normal that the fit starts to deviate somewhat. Confidence bands are important to recognize at what level of magnitude the fit starts to fail (where there is not enough data to offer robust estimates).
The Return Level Plot offers the link between frequency (return periods) and the magnitude of events (return levels). The further the extrapolation (note, these calculations used 30-years of data), the wider the confidence bands.