Decadal climate prediction and predictability for climate services

Abstract

[eng] Climate variations at annual to decadal time scales affect many regions around the globe, causing direct impacts on the economy, ecosystems and society. Knowing these variations in advance allows for implementing measures to adapt, mitigate and build resilience to the consequences of a changing climate. At decadal time scale, climate variations are caused by both external forcings and internal climate variability. While climate projections incorporate external forcing information based on different socio-economic scenarios to project possible pathways the climate system would follow, decadal predictions also incorporate information on the current climate state through a model initialisation procedure. Forecast quality assessment, which involves comparing hindcasts (retrospective predictions) to past observations to evaluate their degree of agreement, is an essential step to ensure that such predictions are trustable and beneficial for decision-making. Climate hindcasts also allow for applying post-processing techniques such as calibration and downscaling methods, as well as for selecting the best climate information for each specific variable, region and forecast period. This Ph.D. thesis has focused on evaluating the forecast quality of variables and indicators relevant for decision-making in several sectors, with a particular focus on agriculture. The evaluation has been performed globally, for individual models and multi-model ensemble, and different forecast periods to identify windows of opportunity for which the climate predictions show skill to support decision-making. Furthermore, historical forcing simulations (retrospective climate projections) have been used to estimate the impact of initialisation. First, the quality of multi-model forecasts of temperature, precipitation, the Atlantic Multidecadal Variability index (AMV) and Global near-Surface Air Temperature (GSAT) generated from all decadal hindcasts contributing to the Coupled Model Intercomparison Project Phase 6 (CMIP6) were evaluated, finding high skill for predictions of temperature, AMV and GSAT, and limited skill for predictions of precipitation. The multi-model ensemble was also compared to individual forecast systems, finding that the best system for each particular location usually outperforms the multi-model ensemble. However, the multi-model provides higher skill than at least half of the systems. The decadal predictions were also compared to the historical simulations, finding an added value from model initialisation over several ocean and land regions for temperature, and for AMV and GSAT. The full multi-model was compared to a sub-ensemble of predictions generated from forecast systems that provided timely forecasts to assess the impact of the ensemble size in an operational climate services context. Second, the representation and prediction of the Euro-Atlantic weather regimes by the EC-Earth3 model were assessed by identifying the dominating atmospheric circulation patterns in this region. The Euro-Atlantic weather regimes are the positive and negative phases of the North Atlantic Oscillation (NAO+ and NAO-, respectively), Blocking, and Atlantic Ridge in winter; and the NAO-, Blocking, Atlantic Ridge, and Atlantic Low in summer. The EC-Earth3 correctly represents the spatial patterns and climatological frequencies of all weather regimes. However, the skill in predicting the annual to decadal variations of the weather regimes’ frequency of occurrence is low, and the model initialisation does not improve such prediction skill. Third, the multi-model forecast quality of the CMIP6 decadal hindcasts is evaluated for multi-annual predictions of a set of indices related to the frequency and intensity of daily temperature and precipitation extremes. The multi-model ensemble is skillful in predicting temperature extremes over most land regions, while the quality is lower for precipitation extremes. Comparing the skill with that for mean temperature and precipitation, extremes are predicted with lower skill, especially those related to the most extreme days. Compared to the historical forcing simulations, decadal predictions show only small and region-dependent skill improvements from model initialisation. Finally, this Ph.D. thesis presents the applications of the research within several European projects and a contract with a private company for which prototypes of climate services have been created. For instance, prototypes of forecast products have been developed for the Southern African Development Community region. These prototypes consist of annual and multi-annual forecast products of temperature, precipitation and drought conditions

[spa] Las variaciones climáticas de uno a diez años impactan a la economía, ecosistemas y sociedad. Anticipar dichas variaciones permite implementar medidas de adaptación y mitigación a consecuencias de un clima variable. La variabilidad decadal está causada tanto por forzamientos externos como por variabilidad interna. Los modelos climáticos permiten estudiar la dinámica y anticipar las variaciones. Mientras que las proyecciones climáticas incorporan forzamientos externos basados en distintos escenarios socioeconómicos, las predicciones decadales también incorporan información del estado actual del sistema climático a través del proceso de inicialización del modelo. Esta tesis doctoral se centra en evaluar la calidad de las predicciones comparándolas con observaciones del pasado con el objetivo de asegurar que son útiles y beneficiosas para la toma de decisiones. Esta evaluación permite identificar variables, regiones y periodos en los cuales la calidad de la información climática se puede utilizar para llevar a cabo decisiones. También, se han aplicado técnicas de postprocesado (calibración, multi-modelo y regionalización) con el fin de mejorar las predicciones y adaptarlas a las necesidades de los usuarios. Las predicciones se han comparado con las simulaciones de forzamiento histórico con el fin de estimar el impacto de la inicialización del modelo en la calidad de las predicciones. La tesis consta de tres publicaciones científicas en peer-reviewed revistas. El primer estudio se centra en la evaluación de predicciones multi-modelo para temperatura, precipitación, Atlantic Multidecadal Variability index (AMV) y temperatura global. También se estima el beneficio del uso del multi-modelo en lugar de modelos individuales, impacto de la inicialización, impacto de calibración, e impacto del número de modelos. El segundo estudio evalúa la representación espacial y predictibilidad temporal de tipos de tiempo europeos (por ejemplo, el bloqueo). El tercer estudio se enfoca a la predicción de extremos climáticos de temperatura y precipitación dada su importancia en la sociedad y sectores vulnerables a las variaciones climáticas. Finalmente, se muestran ejemplos de aplicaciones de la investigación llevada a cabo junto con prototipos de servicios climáticos, con particular enfoque a la agricultura.