Comparison of Potential Seasonal Predictability of Temperature in Statistical Methods

This study compares estimates of potential seasonal predictability from a single time series using four methods: a spectral method proposed by Madden (MN), an Analysis of Variance procedure proposed by Shukla and Gutzler (SG), a bootstrap method and an Analysis of Covariance (ANOCOVA) method proposed by the authors. The comparison is conducted using the time series from the Monte Carlo experiments, atmospheric general circulation model (AGCM) simulations, and reanalysis data. The results indicate that SG systematically underestimates weather noise variance and is considered as a less useful method. MN tends to produce the least biased estimates of noise variance, but it has a higher probability of identifying insignificant predictability than the other methods. ANOCOVA generally gives more accurate, but biased, estimates of weather noise compared to MN. The bootstrap appears to lie between MN and ANOCOVA in certain scenarios or inferior to them in other situations. We conclude that no simple, universally corrected statements can be made regarding the relative performances of MN, ANOCOVA and bootstrap. The reanalysis-based potential predictability of seasonal mean 2-m temperature derived from four methods reveals high predictability in the tropics and low estimates in the extratropics. Omitting SG, the remaining three methods consistently identify about 80% of the globe as significantly predictable, 5% of the globe as insignificantly predictable, and 15% of the globe with inconsistent assessments of potential predictability, mostly over extratropical land areas.