Search

Large multiscenario multimodel ensembles (MMEs) of regional climate model (RCM) experiments driven by global climate models (GCMs) are made available worldwide and aim at providing robust estimates of climate changes and associated uncertainties. Due to many missing combinations of emission scenarios and climate models leading to sparse scenario–GCM–RCM matrices, these large ensembles, however, are very unbalanced, which makes uncertainty analyses impossible with standard approaches. In this paper, the uncertainty assessment is carried out by applying an advanced statistical approach, called QUALYPSO, to a very large ensemble of 87 EURO-CORDEX climate projections, the largest MME based on regional climate models ever produced in Europe. This analysis provides a detailed description of this MME, including (i) balanced estimates of mean changes for near-surface temperature and precipitation in Europe, (ii) the total uncertainty of projections and its partition as a function of time, and (iii) the list of the most important contributors to the model uncertainty. For changes in total precipitation and mean temperature in winter (DJF) and summer (JJA), the uncertainty due to RCMs can be as large as the uncertainty due to GCMs at the end of the century (2071–2099). Both uncertainty sources are mainly due to a small number of individual models clearly identified. Due to the highly unbalanced character of the MME, mean estimated changes can drastically differ from standard average estimates based on the raw ensemble of opportunity. For the RCP4.5 emission scenario in central–eastern Europe for instance, the difference between balanced and direct estimates is up to 0.8 ∘ C for summer temperature changes and up to 20 % for summer precipitation changes at the end of the century. [ABSTRACT FROM AUTHOR]

Deep learning techniques (in particular convolutional neural networks, CNNs) have recently emerged as a promising approach for statistical downscaling due to their ability to learn spatial features from huge spatiotemporal datasets. However, existing studies are based on complex models, applied to particular case studies and using simple validation frameworks, which makes a proper assessment of the (possible) added value offered by these techniques difficult. As a result, these models are usually seen as black boxes, generating distrust among the climate community, particularly in climate change applications. In this paper we undertake a comprehensive assessment of deep learning techniques for continental-scale statistical downscaling, building on the VALUE validation framework. In particular, different CNN models of increasing complexity are applied to downscale temperature and precipitation over Europe, comparing them with a few standard benchmark methods from VALUE (linear and generalized linear models) which have been traditionally used for this purpose. Besides analyzing the adequacy of different components and topologies, we also focus on their extrapolation capability, a critical point for their potential application in climate change studies. To do this, we use a warm test period as a surrogate for possible future climate conditions. Our results show that, while the added value of CNNs is mostly limited to the reproduction of extremes for temperature, these techniques do outperform the classic ones in the case of precipitation for most aspects considered. This overall good performance, together with the fact that they can be suitably applied to large regions (e.g., continents) without worrying about the spatial features being considered as predictors, can foster the use of statistical approaches in international initiatives such as Coordinated Regional Climate Downscaling Experiment (CORDEX). [ABSTRACT FROM AUTHOR]

This paper presents a novel data set of regional climate model simulations over Europe that significantly improves our ability to detect changes in weather extremes under low and moderate levels of global warming. The data set provides a unique and physically consistent data set, as it is derived from a large ensemble of regional climate model simulations. These simulations were driven by two global climate models from the international HAPPI consortium. The set consists of 100 × 10-year simulations and 25 × 10-year simulations, respectively. These large ensembles allow for regional climate change and weather extremes to be investigated with an improved signal-to-noise ratio compared to previous climate simulations. The changes in four climate indices for temperature targets of 1.5 °C and 2.0 °C global warming are quantified: number of days per year with daily mean near-surface apparent temperature of > 28 °C (ATG28); the yearly maximum 5-day sum of precipitation (RX5day); the daily precipitation intensity of the 50-yr return period (RI50yr); and the annual Consecutive Dry Days (CDD). This work shows that even for a small signal in projected global mean temperature, changes of extreme temperature and precipitation indices can be robustly estimated. For temperature related indices changes in percentiles can also be estimated with high confidence. Such data can form the basis for tailor-made climate information that can aid adaptive measures at a policy-relevant scales, indicating potential impacts at low levels of global warming at steps of 0.5 °C. [ABSTRACT FROM AUTHOR]

Deep learning techniques (in particular convolutional neural networks, CNNs) have recently emerged as a promising approach for statistical downscaling due to their ability to learn spatial features from huge spatio-temporal datasets. However, existing studies are based on complex models, applied to particular case studies and using simple validation frameworks, which makes difficult a proper assessment of the (possible) added value offered by these techniques. As a result, these models are usually seen as black-boxes generating distrust among the climate community, particularly in climate change problems. In this paper we undertake a comprehensive assessment of deep learning techniques for continental-scale statistical downscaling, building on the VALUE validation framework. In particular, different CNN models of increasing complexity are applied for downscaling temperature and precipitation over Europe, comparing them with a few standard benchmark methods from VALUE (linear and generalized linear models) which have been traditionally used for this purpose. Besides analyzing the adequacy of different components and topologies, we also focus on their extrapolation capability, a critical point for their possible application in climate change studies. To do this, we use a warm test period as surrogate of possible future climate conditions. Our results show that, whilst the added value of CNNs is mostly limited to the reproduction of extremes for temperature, these techniques do outperform the classic ones for the case of precipitation for most aspects considered. This overall good performance, together with the fact that they can be suitably applied to large regions (e.g. continents) without worrying about the spatial features being considered as predictors, can foster the use of statistical approaches in international initiatives such as CORDEX. [ABSTRACT FROM AUTHOR]

The prediction of climate on time scales of years to decades is attracting the interest of both climate researchers and stakeholders. The German Ministry for Education and Research (BMBF) has launched a major research programme on decadal climate prediction called MiKlip (Mittelfristige Klimaprognosen, Decadal Climate Prediction) in order to investigate the prediction potential of global and regional climate models (RCMs). In this paper we describe a regional predictive hindcast ensemble, its validation, and the added value of regional downscaling. Global predictions are obtained from an ensemble of simulations by the MPI-ESM-LR model (baseline 0 runs), which were downscaled for Europe using the COSMO-CLM regional model. Decadal hindcasts were produced for the 5 decades starting in 1961 until 2001. Observations were taken from the E-OBS data set. To identify decadal variability and predictability, we removed the long-term mean, as well as the long-term linear trend from the data. We split the resulting anomaly time series into two parts, the first including lead times of 1-5 years, reflecting the skill which originates mainly from the initialisation, and the second including lead times from 6-10 years, which are more related to the representation of low frequency climate variability and the effects of external forcing. We investigated temperature averages and precipitation sums for the summer and winter half-year. Skill assessment was based on correlation coefficient and reliability. We found that regional downscaling preserves, but mostly does not improve the skill and the reliability of the global predictions for summer half-year temperature anomalies. In contrast, regionalisation improves global decadal predictions of half-year precipitation sums in most parts of Europe. The added value results from an increased predictive skill on grid-point basis together with an improvement of the ensemble spread, i.e. the reliability. [ABSTRACT FROM AUTHOR]

The article discusses the importance of the COST Action 725 from 2004 to 2009 towards the collaboration of the European COoperation in the field of Scientific and Technical Research. The COST725 attained the common system, phenological observations, and utilizing data in scientific peer reviewed papers. It notes the accomplishments of the program including the phenological response to climate and the launching of project in 2010 under European network of meteorological services (EUMETNET/ECSN).