Decadal climate predictions have the objective to predict the development of the climate for the following years to decades. Numerical Earth system models are initialized with observational values, similar to the methodology applied in weather forecasting. Additionally, they are forced by boundary conditions, like greenhouse gas scenarios, to project the long term development. This thesis investigates decadal climate predictions with Earth system models and their further improvement.
A decadal prediction system is evaluated and investigated for sources of potential skill. Hence a systematic evaluation strategy is developed. It contains the assessment of accuracy of the ensemble mean and the ensemble spread, and compares decadal experiments with climatology, observations, and climate projections. This initialized reference system leads to good predictive skill in temperature and precipitation forecasts. The evaluation shows that the decadal prediction is scientifically sound, but it also has potential for improvement. The initialization with observed ocean data and the prediction with the ensemble mean of a larger ensemble size turn out to be sources of skill for decadal predictions. The entire assessment is performed within a novel evaluation system called Freva. This system is designed to complement climate modeling by a systematic and efficient assessment. Freva serves as a resource-efficient process framework between the data generation and its evaluation, to detect decadal climate prediction potential.
A new prediction technique called ’Ensemble Dispersion Filter’ is developed. It exploits two important climate prediction paradigms: the ocean’s heat capacity and the advantage of the ensemble mean. The Ensemble Dispersion Filter averages the ocean temperatures of the ensemble members every three months, uses this ensemble mean as a restart condition for each member, and further executes the prediction. The evaluation by the new verification framework shows that the Ensemble Dispersion Filter results in a significant improvement in the predictive skill compared to the unfiltered reference system. Even in comparison with prediction systems of a larger ensemble size and higher resolution, the Ensemble Dispersion Filter system performs better. In particular, the prediction of the global average temperature of the forecast years 2 to 5 shows a significant skill improvement. Compared to the observational climatology forecast, the Ensemble Dispersion Filter experiment has a Mean Squared Error Skill Score of 0.83, while the unfiltered reference system exceeds only 0.68. With major improvements over the Pacific and North Atlantic, the regional distribution of the Ensemble Dispersion Filter experiment is more accurate than the reference. In precipitation forecasts, improvements are seen over the continents. The prediction of the cyclone frequencies improves over the key region of the North Atlantic. Consequently, the thesis demonstrates a substantial advance in research on decadal climate predictions.