Verification of Temperature and Sea Ice in the MiKlip Decadal Climate Predictions with the ESMValTool

Abstract

Decadal climate predictions, that aim at predicting the time horizon of the next10-30 years, are a relatively new field of research. An open science topic is whetherthe initialization of the climate model simulations with observations of the slowly-varying components of the climate system results in more accurate near-term pre-dictions compared to uninitialized long-term simulations. To address this sciencequestion, Goddard et al. [2013] introduced a verification system for decadal ex-periments that enables a quantitative assessment of the model performance fromthe decadal predictions compared to observations and to uninitialized long-termsimulations.The goal of this thesis is to assess the possible additional predictive skill for near-surface temperature and sea-ice concentrations in the decadal simulations of theMax Planck Institute Earth System Model (MPI-ESM) compared to the uninitial-ized long-term simulations. To allow this assessment, the verification frameworkfrom Goddard et al. (2013) is implemented into the Earth System Model Valida-tion Tool (ESMValTool). The ESMValTool is a software tool developed by multipleinstitutions that aims at improving routine Earth system model (ESM) evaluation.For this work, in particular the anomaly correlation skill, reliability and accuracy ofthe simulations are evaluated and tested against each other, the model’s uninitial-ized long-term simulations, and observations.No further prediction skill in global mean near-surface temperature is found fordecadal hindcasts (i.e., retrospective forecasts) in comparison to the long-termsimulations, except for the initialization year 1. In the following years, the decadalhindcasts drift to their preferred biased model state resulting in a prediction skillthat is similar to that of the long-term simulations. On regional scales however,certain areas such as southwest of the South American continent and the NorthAtlantic Ocean show a significantly higher predictive skill. This regionality alsotranslates to sea ice.Further studies are required that expand the proposed metrics and include dif-ferent variables and additional climate models to provide a concluding answer tothe question of whether the initialization of climate models can lead to a higherpredictability of near-future climate change.