Estimation of global extreme temperature trends by merging Annual Maxima and Peaks Over Threshold

Annual Maxima (AM) and Peaks over Threshold (POT) are the two most common approaches to define extreme time series in hydroclimatic variables. Both methods present limitations. AM frequently fails to include significant extremes that occur during the same year. Conversely, POT may only include clustered values from a few years thus excluding many years from the analysis, especially when the threshold is set high. Additionally, a big challenge in POT is identifying the threshold which can markedly affect the results.

Here, we merge notions from both AM and POT, preserving the strengths of each approach, to extract extreme temperature series and estimate the trends in their frequency and magnitude. We select the values larger than or equal to the minimum of the AM series as high temperatures (HT) (lower than or equal to the maximum of the Annual Minima as the low temperatures – LT). Thus, each year of the HT, LT series has at least one extreme value (H1, L1). We apply the method to 4797 quality-controlled raw station observations from a global dataset of maximum and minimum temperatures over 1970-2019 when warming accelerates. To examine changes in H1-L1 frequency and magnitude, we estimate the ratio of observed to expected H1 (L1) annual occurrences, and the difference between the observed and expected mean H1 (L1) annual temperature values, respectively. We estimate the regression slopes of these ratios at the station level, regionally in 2°×2° grids, and globally. We then compare these trends with the ones obtained from AM and POT series. The proposed method adapts the threshold for each sample, and finds a compromise among all tested methods, thus being a flexible approach that can be applied to any non-intermittent variable.

 

Acknowledgment

This research was supported by a GWF Ph.D. Excellence Scholarship from the Global Institute for Water Security (GIWS), University of Saskatchewan