What is the Necessary Ensemble Size for a Sub-Seasonal Timescale Simulation of Regional Climate?

Determining the minimum ensemble size to characterize climate is important yet challenging. While there can be various ways to do it, we propose a new method to determine what ensemble size is required to represent intraseasonal fluctuations of the atmospheric circulation and their regional dependence, as illustrated here for the 300-hPa geopotential height in the cold season.

The method is first applied to ERA 5 satellite-era (1979-2021) reanalysis data. For each grid point, we randomly select N sample years of 3-hourly time series from November to January among the 43 year-long dataset. We then establish the correlation coefficient r between the mean of these N time series and the mean time series of the complete 43 years, which is considered as the target truth. This procedure is repeated 100 times, and the mean of the resulting 100 correlation coefficients is used to quantify the similarity between the N-year climatology and the 43-year climatology. For a given correlation threshold, one can determine the sample size N needed to generate a representative control climate.

With an example threshold of r=0.99, it is found that the North Atlantic, northern Eurasia, northeastern North Pacific, and the northwestern corner of North America are the regions that require the greatest number of samples to generate a representative climate, ranging from 35 to 42 ensemble members (or years of average). The northeastern corner of North America, however, requires 30 to 35 samples. Over the Tibetan Plateau, East Asia, and the southwestern side of the North Pacific, a representative climate can be obtained with even a smaller number of sample sizes, ranging from 20 to 25 samples. Similar regional characteristics are obtained with an extended ERA datasets, and high-resolution atmosphere simulations, stressing that the necessary ensemble size to simulate a representative seasonal climate has a strong regional dependency.