Regional climate downscaling over Europe: perspectives from the EURO-CORDEX community

Regional Environmental Change, Год журнала: 2020, Номер 20(2)

Опубликована: Апрель 23, 2020

Abstract The European CORDEX (EURO-CORDEX) initiative is a large voluntary effort that seeks to advance regional climate and Earth system science in Europe. As part of the World Climate Research Programme (WCRP) - Coordinated Regional Downscaling Experiment (CORDEX), it shares broader goals providing model evaluation projection framework improving communication with both General Circulation Model (GCM) data user communities. EURO-CORDEX oversees design coordination ongoing ensembles projections unprecedented size resolution (0.11° EUR-11 0.44° EUR-44 domains). Additionally, inclusion empirical-statistical downscaling allows investigation much larger multi-model ensembles. These complementary approaches provide foundation for scientific studies within research community others. value ensemble shown via numerous peer-reviewed its use development services. Evaluations also show benefits higher resolution. However, significant challenges remain. To further understanding, two flagship pilot (FPS) were initiated. first investigates local-regional phenomena at convection-permitting scales over central Europe Mediterranean collaboration Med-CORDEX community. second impacts land cover changes on across spatial temporal scales. Over coming years, looks forward closer other communities, new advances, supporting international initiatives such as IPCC reports, continuing basis adaptation

Язык: Английский

---

Deep Learning for Daily Precipitation and Temperature Downscaling

Water Resources Research, Год журнала: 2021, Номер 57(4)

Опубликована: Апрель 1, 2021

Abstract Downscaling is a critical step to bridge the gap between large‐scale climate information and local‐scale impact assessment. This study presents novel deep learning approach: Super Resolution Deep Residual Network (SRDRN) for downscaling daily precipitation temperature. approach was constructed based on an advanced convolutional neural network with residual blocks batch normalizations. The data augmentation technique utilized address overfitting that due highly imbalanced nonprecipitation days sparse extremes. Synthetic experiments were designed downscale maximum/minimum temperature from coarse resolutions (25, 50, 100 km) high resolution (4 km). results showed that, during validation period, SRDRN not only captured spatial temporal patterns remarkably well, but also reproduced both extremes in different locations time at local scale. Through transfer learning, trained model one region directly applied another environment, notable improvement compared classic statistical methods. outstanding performance of stemmed its ability fully extract features without suffering degradation issues incorporations blocks, normalizations, augmentations. thus powerful tool can potentially be leveraged any hydrologic, climate, earth system data.

Язык: Английский

---

Deep learning in statistical downscaling for deriving high spatial resolution gridded meteorological data: A systematic review

ISPRS Journal of Photogrammetry and Remote Sensing, Год журнала: 2024, Номер 208, С. 14 - 38

Опубликована: Янв. 9, 2024

Язык: Английский

---

Estimating pros and cons of statistical downscaling based on EQM bias adjustment as a complementary method to dynamical downscaling

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 3, 2025

Язык: Английский

---

Testing bias adjustment methods for regional climate change applications under observational uncertainty and resolution mismatch

Atmospheric Science Letters, Год журнала: 2020, Номер 21(7)

Опубликована: Апрель 20, 2020

Abstract Systematic biases in climate models hamper their direct use impact studies and, as a consequence, many statistical bias adjustment methods have been developed to calibrate model outputs against observations. The application of these change context is problematic since there no clear understanding on how may affect key magnitudes, for example, the signal or trend, under different sources uncertainty. Two relevant uncertainty, often overlooked, are sensitivity observational reference used method and effect resolution mismatch between observations ( downscaling effect). In present work, we assess factors temperature precipitation considering marginal, temporal extreme aspects. We eight standard state‐of‐the‐art (spanning variety regarding nature—empirical parametric—, fitted parameters trend‐preservation) case study Iberian Peninsula. quantile trend‐preserving (namely delta mapping (QDM), scaled distribution (SDM) from third phase ISIMIP‐ISIMIP3) preserve better raw signals indices variables considered (not all preserved by construction). However, they rely largely dataset calibration, thus presenting larger observations, especially intensity, spells indices. Thus, high‐quality datasets essential comprehensive analyses (continental) domains. Similar conclusions hold experiments carried out at high (approximately 20 km) low 120 spatial resolutions.

Язык: Английский

---

A comparative study of convolutional neural network models for wind field downscaling

Meteorological Applications, Год журнала: 2020, Номер 27(6)

Опубликована: Ноя. 1, 2020

We analyze the applicability of convolutional neural network (CNN) architectures for downscaling short-range forecasts near-surface winds on extended spatial domains. Short-range wind (at 100 m level) from European Centre Medium Range Weather Forecasts ERA5 reanalysis initial conditions at 31 km horizontal resolution are downscaled to mimic high (HRES) (deterministic) 9 resolution. evaluate quality four exemplary CNN and compare these against a multilinear regression model. conduct qualitative quantitative comparison model predictions examine whether predictive skill CNNs can be enhanced by incorporating additional atmospheric variables, such as geopotential height forecast surface roughness, or static high-resolution fields, like land–sea mask topography. further propose DeepRU, novel U-Net-based architecture, which is able infer situation-dependent structures that cannot reconstructed other models. Inferring target field low-resolution input fields over Alpine area takes less than 10 ms our graphics processing unit compares favorably an overhead in simulation time minutes hours between low- simulations.

Язык: Английский

---

Statistical downscaling with the downscaleR package (v3.1.0): contribution to the VALUE intercomparison experiment

Geoscientific model development, Год журнала: 2020, Номер 13(3), С. 1711 - 1735

Опубликована: Апрель 1, 2020

Abstract. The increasing demand for high-resolution climate information has attracted growing attention to statistical downscaling (SDS) methods, due in part their relative advantages and merits as compared dynamical approaches (based on regional model simulations), such much lower computational cost fitness purpose many local-scale applications. As a result, plethora of SDS methods is nowadays available scientists, which motivated recent efforts comprehensive evaluation, like the VALUE initiative (http://www.value-cost.eu, last access: 29 March 2020). systematic intercomparison large number techniques undertaken VALUE, them independently developed by different authors modeling centers variety languages/environments, shown compelling need new tools allowing application within an integrated framework. In this regard, downscaleR R package covers most popular (model output statistics – including so-called “bias correction” perfect prognosis) state-of-the-art techniques. It been conceived work primarily with daily data can be used framework both seasonal forecasting change studies. Its full integration climate4R (Iturbide et al., 2019) makes possible development end-to-end applications, from retrieval building, validation, prediction, bringing scientists practitioners unique development. article main features are showcased through replication some results obtained placing emphasis technically complex stages perfect-prognosis calibration (predictor screening, cross-validation, selection) that accomplished simple commands extremely flexible tuning, tailored needs users requiring easy interface levels experimental complexity. open-source framework, freely necessary scripts fully replicate experiments included paper also provided companion notebook.

Язык: Английский

---

Comparison of statistical downscaling methods for climate change impact analysis on precipitation-driven drought

Hydrology and earth system sciences, Год журнала: 2021, Номер 25(6), С. 3493 - 3517

Опубликована: Июнь 21, 2021

Abstract. General circulation models (GCMs) are the primary tools for evaluating possible impacts of climate change; however, their results coarse in temporal and spatial dimensions. In addition, they often show systematic biases compared to observations. Downscaling bias correction model outputs is thus required local applications. Apart from computationally intensive strategy dynamical downscaling, statistical downscaling offers a relatively straightforward solution by establishing relationships between small- large-scale variables. This study compares four methods (BC), change factor mean (CFM), quantile perturbation (QP) an event-based weather generator (WG) assess impact on drought end 21st century (2071–2100) relative baseline period 1971–2000 station Uccle located Belgium. A set drought-related aspects analysed, i.e. dry day frequency, spell duration total precipitation. The applied 28-member ensemble Coupled Model Intercomparison Project Phase 6 (CMIP6) GCMs, each forced future scenarios SSP1–2.6, SSP2–4.5, SSP3–7.0 SSP5–8.5. 25-member CanESM5 GCM also used significance signals comparison internal variability climate. performance reveals that QP method outperforms others reproducing magnitude monthly pattern observed indicators. While all good agreement precipitation, differ quite largely frequency length spells. Using methods, projected increase significantly summer months, with up 19 % At same time, precipitation decrease 33 these months. Total increases winter, as it driven significant intensification extreme rather than change. Lastly, spells 9 %.

Язык: Английский

---

An evaluation framework for downscaling and bias correction in climate change impact studies

Journal of Hydrology, Год журнала: 2023, Номер 622, С. 129693 - 129693

Опубликована: Май 30, 2023

Climate change impact studies commonly use models (such as hydrological or crop models) forced with corrected climate input data from global models. A range of downscaling and bias correction methods have been developed to increase the spatial resolution remove systematic biases in model outputs be applied before Many focused on evaluating such approaches for variables they aim correct. However, due nonlinear error propagation there can large remaining outputs, even when ingesting forcings. Here we propose an impact-centric evaluation framework used risk assessments. This evaluates compares strengths limitations domain, highlighting that lead reduced interest. We demonstrate context assessing projections Australia. Our results show although all evaluated perform adequately variables, their errors vary markedly is modelled. proposed involves selecting a number key performance metrics, ranking four compute overall ranking, best-performing each statistical metric approach. present application this approach using metrics relevant applications, relating mean biases, variability, heavy precipitation peak runoff days, dry conditions. For related find multi-variate considers cross-correlations, temporal auto-correlations at multiple time scales (daily annual) performs best reducing output wide applications where are required, including impacts agricultural production, wildfires, energy generation, human health, ecosystem functioning, water resource management.

Язык: Английский

---

High-resolution downscaling of CMIP6 Earth system and global climate models using deep learning for Iberia

Geoscientific model development, Год журнала: 2024, Номер 17(1), С. 229 - 259

Опубликована: Янв. 12, 2024

Abstract. Deep learning (DL) methods have recently garnered attention from the climate change community for being an innovative approach to downscaling variables Earth system and global models (ESGCMs) with horizontal resolutions still too coarse represent regional- local-scale phenomena. In context of Coupled Model Intercomparison Project phase 6 (CMIP6), ESGCM simulations were conducted Sixth Assessment Report (AR6) Intergovernmental Panel on Climate Change (IPCC) at ranging 0.70 3.75∘. Here, four convolutional neural network (CNN) architectures evaluated their ability downscale, a resolution 0.1∘, seven CMIP6 ESGCMs over Iberian Peninsula – known hotspot, due its increased vulnerability projected future warming drying conditions. The study is divided into three stages: (1) evaluating performance CNN in predicting mean, minimum, maximum temperatures, as well daily precipitation, trained using ERA5 data compared Iberia01 observational dataset; (2) further ensemble against Iberia01; (3) constructing multi-model CNN-based downscaled projections temperature precipitation 0.1∘ throughout 21st century under Shared Socioeconomic Pathway (SSP) scenarios. Upon validation satisfactory evaluation, DL demonstrate overall agreement magnitude sign changes. Moreover, advantages high-resolution are evident, offering substantial added value representing regional Iberia. Notably, clear trend observed Iberia, consistent previous studies this area, increases 2 ∘C, depending scenario. Regarding robust decreases western southwestern particularly after 2040. These results may offer new tool providing information adaptation strategies based prior next European branch Coordinated Regional Downscaling Experiment (EURO-CORDEX) experiments.

Язык: Английский

---