Henrik Feddersen

Reduction of Model Systematic Error by Statistical Correction for Dynamical Seasonal Predictions

Abstract Singular value decomposition analysis (SVDA) is used to analyze an ensemble of three 34-yr general circulation model (GCM) simulations forced with observed sea surface temperature. It is demonstrated how statistical postprocessing based on the leading SVDA modes of simulated and observed fields, primarily precipitation, can be applied to improve the skill of the simulation. For a given limited prediction region, the GCM has the potential to nonlinearly transform the SST information from around the globe and produce a dynamic solution over the region that can be statistically corrected to account for such features as systematic shifts in the location of anomaly dipoles. This paper does not address the separate question of whether the current generation of GCMs contain information above that which could be extracted using linear statistical relationships with SST. For precipitation, examples are drawn from skillful tropical regions, as well as the moderate-to-low skill Pacific–North American and No...

A method for statistical downscaling of seasonal ensemble predictions

A model output statistics based method for downscaling seasonal ensemble predictions is outlined, and examplesof ensemble predictions of precipitation and 2-m temperature are verified against observing stations in Scandinavia, Europe, north-western America, the contiguous United States and Australia. The downscaling from seasonal ensemblepredictions from coupled ocean/atmosphere general circulation models to daily precipitation time series for individualobserving stations is performed in three steps: (i) a spatial downscaling of ensemble mean seasonal means from dynamicalmodel output to station level by means of patterns derived from a singular value decomposition analysis of model outputand observations; (ii) application of the downscaling transformation to the model output ensemble and subsequentcalibration of the downscaled ensemble; (iii) a stochastic generation of daily precipitation conditioned on predictionsof the probability of a wet day in the season and daily persistence. In the majority of the examples, the downscaling isfound to provide more skilful predictions than the raw dynamical model output.

Predictability of seasonal precipitation in the Nordic region

Predictability of seasonal precipitation in the Nordic countries (Denmark, Finland, Iceland, Norwayand Sweden) is investigated using a nine-member ensemble of atmospheric general circulation modelsimulations with prescribed sea-surface temperature from October 1950 until March 1999. The simulationsand corresponding observations from 65 stations in the Nordic countries are used to identifylarge-scale patterns of seasonal precipitation, the predictability of which is investigated. Subsequently, the identified large-scale patterns are used in a statistical downscaling of the model simulated precipitation.The downscaling, which is of the model output statistics type, yields seasonal predictions for theindividual stations. The model simulations of precipitation are compared to predictions of precipitationdirectly from observed sea-surface temperature using a statistical prediction method and no dynamicmodel. The two different methods give consistent results. It is demonstrated that seasonal precipitationin the Nordic region contains a weak predictable signal in several seasons. The most skilful predictionscan be made in spring, especially in the April’June season when precipitation appears to be influencedboth by tropical and North Atlantic sea surface temperature. In particular, the North Atlantic Oscillationin winter appears to influence the North Atlantic sea-surface temperature in spring, which in turn hasan effect on precipitation in Scandinavia.