The Bayesian Processor of Forecasts on the probabilistic forecasting of long-range dependent variables using General Circulation Models

H. Tyralis, and D. Koutsoyiannis, The Bayesian Processor of Forecasts on the probabilistic forecasting of long-range dependent variables using General Circulation Models, Asia Oceania Geosciences Society (AOGS) 14th Annual Meeting, Singapore, HS20-A002, doi:10.13140/RG.2.2.15481.77922, Asia Oceania Geosciences Society, 2017.

[doc_id=1720]

[English]

We derive the distribution of the mean annual temperature and precipitation in the USA for the time period 2016-2100 conditional on observations from the time period 1916-2015 and ensembles from the phase 5 of the Coupled Model Intercomparison Project (CMIP5). To this end, we model the mean annual temperature and precipitation with the Hurst-Kolmogorov stochastic process (HKp, also known as Fractional Gaussian noise) to represent their long-range dependence (LRD). The HKp is a suitable model for climatic variables as has thoroughly been examined in the literature, while it can produce probabilistic forecasts conditional on historical observations. To improve the forecasts using the CMIP5 model ensembles, we apply the Bayesian Processor of Forecasts (BPF), which is a well-established technique used to forecast probabilistically weather and climatic variables conditional on a deterministic model output. The BPF is a general algorithm in the sense that it can be applied to any distribution and dependence pattern of the variables. However, it has been analysed theoretically and numerically solely for independent or Markov dependent variables. Here we extend its application to LRD dependent variables. The computation of uncertainties of climate projections is a mainstream subject in the climate literature and here we show that the BPF can be a sufficient solution.

PDF Full text (1895 KB)