What Makes Northern Hemisphere Heatwave Regimes Typical

Methods

Methods.

Limitations and scope for future work.

Limitations and scope for future work. There are several limitations to this work. Notably, while we note that the AA-derived events are almost always associated with extremes, not all extremes are associated with AA-derived events (see, for instance, the discussion of the August 2010 case study in Section 3). We hypothesize that this is due to the fact that the events in our catalogue are developed from the entire Northern Hemisphere, meaning that only the largest-scale atmospheric regimes are captured, and events without a global imprint are likely to be.

Atmospheric Reanalysis

Atmospheric Reanalysis The primary data source used here is the output from the Japanese 55-Year Reanalysis project (JRA-55, Kobayashi et al. (2015) ), with the extended output from January 1st 1958 until the 31st of December 2023, a total of 65 years. We make use of the daily mean surface air temperature (at 2m above the ground or sea-surface), as well as the 500hPa geopotential height and 200-hPa wind velocities. We restrict our attention to an extended Northern Hemisphere (boreal) summer, from the 1st of May until the 30th of.

Archetype Analysis

Archetype Analysis Archetype Analysis (Cutler and Breiman, 1994; Hannachi and Trendafilov, 2017 ) (herein AA), is a matrix factorisation method that can extract extreme or outlying configurations from a dataset of finite but otherwise arbitrary dimension. Consider a data matrix X = X s×t describing a spatial-temporal field where s corresponds to the set of spatial dimensions (e.g. latitude and longitude) and t is time. AA approximates X as the convex combination of a set of p archetypal spatial patterns Z s×p , X s×t ≈ Z s×p S p×t.

Interpreting the Output of Archetype Analysis

Interpreting the Output of Archetype Analysis At time of writing, AA is not yet in widespread use in geophysics. To aid the reader, we provide a brief explanation of the utility of AA and its interpretation. AA approximates the data as a number p << t of spatial patterns Z, each with an associated affiliation probability timeseries S p×t . At every time step, the affiliation probability gives the probability that the spatial pattern associated with archetype k ∈ p is expressed. As such, when S kt → 1, the.

Event Identification and Regime Determination

Event Identification and Regime Determination We use the AA to determine ‘events’ -time periods when a particular regime is strongly expressed. We refer to a collection of events as a ‘catalogue’. To define the events that make up a catalogue, we first compute the discrimination score (Risbey et al., 2021) that describes the dominance, or otherwise, of a single archetype at a given time step. First, note that due to the convexity constraints on S, the sum over all p at time t is 1 (i.e. p k S kt.

Extreme Regime Patterns for Detrended Data

Extreme Regime Patterns for Detrended Data AA is applied to detrended JRA-55 reanalysis daily-mean surface temperature anomalies T 2m for the period 1958-2023, over the extend boreal summer (May-September). We compute archetypes from p = 2 to p = 20, inclusive, in order to assess the effects of the number of potential regimes identified by the method. Careful assessment of results has lead us decide on using 8 archetypes for the majority of this study -which provides a good trade-off between simplification of the data and discrimination between distinct regimes.

Figures Explained

The paper’s visual material highlights the workflow and the main system components.

• Figure 1: -180.0 W -120.0 W -60.0 W 0.0 W 60.0 W 120.0 W 180.0 W -120.0 W -60.0 W 0.0 W 60.0 W 120.0 W 180.0 W -120.0 W -60.0 W 0.0 W 60.0 W 120.0 W 180.0 W -120.0 W -60.0 W 0.0 W 60.0 W 120.0 W 180.0 W -120.0 W -60.0 W 0.0 W 60.0 W 120.0 W 180.0 W -120.0 W -60.0 W 0.0 W 60.0 W 120.0 W 180.Anomal y ( C).
• Figure 1 :: Figure1: Archetypal Patterns and Affiliation computed from detrend JRA-55 daily maximum surface temperatures anomalies: left panels spatial patterns for each of the 8 “regimes” showing T ′ 2m max (colours) and z500 (contour lines). The contour interval is 50m, with a maximum value of 300m, dashed contour lines show negative anomalies, while solid contour lines show positive anomalies; right affiliation probability (grey lines) and C matrix weights (orange bars) corresponding to each spatial pattern.
• Figure 2 :: Figure2: Temporal distribution of events from the automated detection procedure: a A calendar showing the distribution of events, coloured by the relevant archetype, by year (y-axis) and day of year (x-axis). White regions show times when no event was detected.; b The total event days for each regime occurring in each year; c The annual cycle for the total event days per year, for each regime. The colour legend is shown in the bottom right of the page. The method for specifying events is described in section 11.21.2.4.
• Figure 3 :: Figure3: The percentage of event days that exceed the surface temperature 90th percentile: For each of the events associated with each individual archetypal pattern in the catalogue, the number of days that exceed the 90th percentile, divided by the total number of event days. The events associated with archetypal patterns 1 through 8 are shown down the page. Solid red contour shows the 10% level, which we might expect if days were drawn randomly.12.
• Figure 4 :: Figure 4: Three example events from the automated detection procedure: left spatial patterns of T 2m max (colours) and z500 (contour lines) for the date falling halfway between event onset and termination; and right areas where the surface temperature anomaly exceeds the 80th, 90th or 95th percentile. The date for each plot is indicated in the in-panel text. These dates correspond to (a,b) the 2003 western European (French) heatwave; (c,d) the 2010 central European (Russian) heatwave; and the 2021 western North American “heatdome” heatwave.