Climate model downscaling in central Asia: a dynamical and a neural network approach

General circulation models (GCMs) are important tools for the study of climate change, but their resolutions are too coarse for station-scale impact assessments. Statistical and dynamical downscaling methods are widely used to translate the predictions of GCMs to the finer spatial scale and it is important to understand the difference between statistical and dynamical downscaling methods in different climatic zones and time periods. Moreover, statistical downscaling can be used on both GCM and regional climate model (RCM) outputs. In this study, two sets of GCM precipitations were dynamically and statistically downscaled and their performances were evaluated against the observed precipitation from 308 stations distributed throughout the Yellow, Yangtze, and Pearl River basins. These stations have distinct climatic characteristics from the historical period (1961–2000) and future period (2031–2050). Results suggest dynamically downscaled GCM precipitation does not present lower biases when comparing observed site-specific precipitation to GCM outputs, and biases of the initial dynamically downscaled GCM outputs decreased in areas with higher humidity. This demonstrates that statistical downscaling can improve GCM and RCM outputs, and the statistical downscaling method can reproduce local-scale precipitation satisfactorily without dynamical downscaling. However, statistical downscaling reduced spatial regularity of the biases that exist in GCM and RCM outputs between the observations and simulation. Additionally, the spatial discrepancy between statistically downscaled GCM and RCM precipitations was very small. In the future period, discrepancies between statistically downscaled RCM and GCM precipitations in the two climate scenarios were larger than the historical period for all climate zones.

High-resolution climate projections are crucial for assessing the future impacts of climate change. Statistical, dynamic, or hybrid climate data downscaling is often employed to create the datasets required for impact modelling. In this study, we utilize the COSMO-CLM (CCLM) version 6.0, a regional climate model, to investigate the advantages of dynamically downscaling a general circulation model (GCM) from CMIP6, with a focus on Central Asia (CA). The CCLM, running at a 0.22° horizontal resolution, is driven by the MPI-ESM1-2-HR GCM (at 1° spatial resolution) for the historical period 1985–2014 and projections for 2019–2100 under three shared socioeconomic pathways (SSPs): SSP1-2.6, SSP3-7.0, and SSP5-8.5 (Fallah et al., 2025). Using the CHIRPS gridded observation dataset for evaluation, we assess the performance of the CCLM driven by ERA-Interim reanalysis over the historical period.The added value of CCLM, particularly over mountainous areas in CA, is evident, with a reduction in mean absolute error and bias of climatological precipitation by 5 mm/day for summer and 3 mm/day for annual values (Fallah et al., 2024). While no error reduction is achieved for winter, the frequency of extreme precipitation events improves in the CCLM simulations. Future projections indicate an increase in the intensity and frequency of extreme precipitation events in CA by the century’s end, particularly under the SSP3-7.0 and SSP5-8.5 scenarios. The number of days with more than 20 mm of precipitation increases by more than 90, and the annual 99th percentile of total precipitation increases by over 9 mm/day in mountainous areas.A convolutional neural network (CNN) is also trained to map GCM simulations to their dynamically downscaled CCLM counterparts. The CNN successfully emulates the GCM-CCLM chain across large areas of CA but demonstrates reduced skill when applied to other GCM-CCLM chains. This downscaling approach and CNN architecture provide an alternative to traditional methods and could be a valuable tool for the scientific community involved in downscaling CMIP6 models (Harder et al., 2023).In future work, we aim to extend this approach by training a neural network model to map the available GCM-RCM model chains for CORDEX-EU and applying the trained model to decadal prediction ICON simulations. This will enable the production of CORDEX-EU-like regional ICON simulations, bridging the gap between global and regional climate information on decadal timescales. By integrating decadal predictions into the framework, we aim to enhance the usability of regionalized climate data for short-term climate planning and decision-making.

An improved deep learning procedure for statistical downscaling of climate data

Statistical and dynamical downscaling predictions of changes in surface temperature and precipitation for 2080–2100, relative to pre-industrial conditions, are compared at 976 European observing sites, for January and July. Two dynamical downscaling methods are considered, involving the use of surface temperature or precipitation simulated at the nearest grid point in a coupled ocean–atmosphere general circulation model (GCM) of resolution ∼300 km and a 50 km regional climate model (RCM) nested inside the GCM. The statistical method (STAT) is based on observed linear regression relationships between surface temperature or precipitation and a range of atmospheric predictor variables. The three methods are equally plausible a priori, in the sense that they estimate present-day natural variations with equal skill. For temperature, differences between the RCM and GCM predictions are quite small. Larger differences occur between STAT and the dynamical predictions. For precipitation, there is a wide spread between all three methods. Differences between the RCM and GCM are increased by the meso-scale detail present in the RCM. Uncertainties in the downscaling predictions are investigated by using the STAT method to estimate the grid point changes simulated by the GCM, based on regression relationships trained using simulated rather than observed values of the predictor and the predictand variables (i.e. STAT_SIM). In most areas the temperature changes predicted by STAT_SIM and the GCM itself are similar, indicating that the statistical relationships trained from present climate anomalies remain valid in the perturbed climate. However, STAT_SIM underestimates the surface warming in areas where advective predictors are important predictors of natural variability but not of climate change. For precipitation, STAT_SIM estimates the simulated changes with lower skill, especially in January when increases in simulated precipitation related to a moister atmosphere are not captured. This occurs because moisture is rarely a strong enough predictor of natural variability to be included in the specification equation. The predictor/predictand relationships found in the GCM do not always match those found in observations. In January, the link between surface and lower tropospheric temperature is too strong. This is also true in July, when the links between precipitation and various atmospheric predictors are also too strong. These biases represent a likely source of error in both dynamical and statistical downscaling predictions. For example, simulated reductions in precipitation over southern Europe in summer may be too large. Copyright © 2000 British Crown Copyright

Statistical and dynamical downscaling predictions of changes in surface temperature and precipitation for 2080–2100, relative to pre-industrial conditions, are compared at 976 European observing sites, for January and July. Two dynamical downscaling methods are considered, involving the use of surface temperature or precipitation simulated at the nearest grid point in a coupled ocean–atmosphere general circulation model (GCM) of resolution ∼300 km and a 50 km regional climate model (RCM) nested inside the GCM. The statistical method (STAT) is based on observed linear regression relationships between surface temperature or precipitation and a range of atmospheric predictor variables. The three methods are equally plausible a priori, in the sense that they estimate present-day natural variations with equal skill. For temperature, differences between the RCM and GCM predictions are quite small. Larger differences occur between STAT and the dynamical predictions. For precipitation, there is a wide spread between all three methods. Differences between the RCM and GCM are increased by the meso-scale detail present in the RCM. Uncertainties in the downscaling predictions are investigated by using the STAT method to estimate the grid point changes simulated by the GCM, based on regression relationships trained using simulated rather than observed values of the predictor and the predictand variables (i.e. STAT_SIM). In most areas the temperature changes predicted by STAT_SIM and the GCM itself are similar, indicating that the statistical relationships trained from present climate anomalies remain valid in the perturbed climate. However, STAT_SIM underestimates the surface warming in areas where advective predictors are important predictors of natural variability but not of climate change. For precipitation, STAT_SIM estimates the simulated changes with lower skill, especially in January when increases in simulated precipitation related to a moister atmosphere are not captured. This occurs because moisture is rarely a strong enough predictor of natural variability to be included in the specification equation. The predictor/predictand relationships found in the GCM do not always match those found in observations. In January, the link between surface and lower tropospheric temperature is too strong. This is also true in July, when the links between precipitation and various atmospheric predictors are also too strong. These biases represent a likely source of error in both dynamical and statistical downscaling predictions. For example, simulated reductions in precipitation over southern Europe in summer may be too large. Copyright © 2000 British Crown Copyright

Application of relative importance metrics for CMIP6 models selection in projecting basin-scale rainfall over Johor River basin, Malaysia

Accurate climate projections are critical for various applications and impact assessments in environmental science and management. This study presents HiCPC (High-resolution CMIP6 downscaled daily Climate Projections over China), a novel dataset tailored to China’s specific needs. HiCPC leverages outputs from 22 global climate models (GCMs) from the Coupled Model Intercomparison Project Phase 6 (CMIP6). To address inherent biases in daily GCM simulations, an advanced Bias Correction and Spatial Disaggregation (BCSD) method is employed, using the China Meteorological Forcing Dataset (CMFD) as a reference. HiCPC offers detailed daily precipitation and temperature data across China at an enhanced spatial resolution of 0.1° × 0.1°. It covers both the historical period (1979–2014) and future projections (2015–2100) based on four CMIP6 Shared Socioeconomic Pathways (SSPs) scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5). Upon validation, HiCPC demonstrates good performance, surpassing CMIP6 GCMs across the historical period. This reinforces its significance for essential research in climate change evaluation and its associated implications within China.

Cold winters in Eurasia considerably affect transportation, agriculture, energy, and public health. This study utilizes 31 global climate models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) and 33 CMIP5 models to evaluate the historical surface air temperature, sea level pressure, 500‐hPa geopotential height, 150‐hPa meridional and zonal wind, and polar vortex indices during cold winters. Our research quantifies the advancements of CMIP6 over CMIP5. Additionally, future changes in these variables under three different Shared Socioeconomic Pathways (SSPs), that is, SSP 1–2.6, SSP 2–4.5, and SSP 5–8.5, are projected based on 20 out of the 31 CMIP6 models. The results indicate that the multimodel ensemble means from both CMIP5 and CMIP6 effectively capture the main features of the observed Eurasian cold winters and their associated factors with good simulation agreement. The CMIP6 ensemble mean outperforms its CMIP5 counterpart, and both ensemble means (CMIP5 and CMIP6) perform better than individual CMIP6 models. Among CMIP6 models, 500‐hPa geopotential height achieves the highest simulation skill, whereas sea level pressure shows the lowest. Compared with same‐institute models from CMIP5, CMIP6 models show overall improvements with sea level pressure simulation being notably advanced. Under the three SSPs, the occurrence probability of cold winters is projected to decrease as the area and intensity indices of the polar vortex decline. Moreover, surface temperature anomalies are projected to exhibit a “warm Arctic and cold Eurasia” pattern, and the anticyclonic anomalies at 500 hPa and 150 hPa are projected to be centered at high latitudes.

Regional climate projections are essential for guiding local governments in developing effective mitigation strategies. A common technique for downscaling General Circulation Model (GCM) outputs is dynamical downscaling, but its high computational demands have motivated the search for alternative approaches, including statistical downscaling. This study presents a two-phase statistical downscaling framework to improve the spatial resolution and accuracy of precipitation and temperature projections. In the first phase, a Convolutional Neural Network (CNN), trained to learn spatial patterns from ERA5 reanalysis data, is employed to refine the coarse grid of GCMs. In the second phase, bias correction is performed using a quantile delta mapping technique, with ERA5 still serving as the reference dataset. The resulting downscaling framework is applied to outputs from five GCMs participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) under two Shared Socioeconomic Pathways (SSPs): SSP1-2.6 and SSP3-7.0. This work is part of the OurMED PRIMA project, which focuses on the Mediterranean region, a recognized climate change hotspot. Results indicate substantial improvements in the accuracy of temperature and precipitation projections compared to other downscaling methods. The proposed approach effectively captures fine-scale spatial variability, a crucial aspect for regional climate studies in complex regions like the Mediterranean region. The downscaled climate data are used to assess climate extremes by computing the indices defined by the Expert Team on Climate Change Detection and Indices (ETCCDI). These indices can offer valuable insights into evolving climate trends and extremes throughout the 21st century. The proposed methodology demonstrates significant potential for broader applications in regions requiring high-resolution climate data to support adaptation strategies and policy development.This work was supported by OurMED PRIMA Program project funded by the European Union’s Horizon 2020 research and innovation under grant agreement No. 2222. Valerio Todaro acknowledges financial support from the PNRR MUR project ECS_00000033_ECOSISTER.

General Circulation Models or Global Climate Models (GCMs) output consists of inevitable bias due to insufficient knowledge about parameterization schemes and other mathematical computations that involve thermodynamical and physical laws while designing climate models. Indian summer monsoon (southwest monsoon) accounts for 75%–90% of the annual rainfall over most climatic zones of India during the months, June, July, August, and September, which has a direct impact on the agricultural economy of India. The aim of this study is to bias correct the Coupled Model Intercomparison Project Phase – 6 (CMIP6) GCMs' precipitation data for the historical period from 1985 to 2014 and two Shared Socioeconomic Pathways (SSP) SSP1‐2.6 and SSP5‐8.5, from the period 2015 to 2100, with reference to the India Meteorological Department (IMD) observed rainfall gridded dataset. The datasets used are for the rain‐bearing Indian southwest monsoon season from the months, June to September. Monsoon Core Region is selected to carry out the bias correction using a couple of deep learning algorithms, namely one‐dimensional Convolutional Neural Network (CNN1D) and Long Short‐Term Memory Encoder‐Decoder (LSTM‐ED) Neural Network. The performance of both algorithms is evaluated with metrics. The LSTM‐ED algorithm yielded better results with least error output. The bias‐corrected data obtained using the LSTM‐ED algorithm is then compared with IMD observed rainfall data for the climatic events such as ENSO (El Niño and La Niña) and Positive and Negative IOD (Indian Ocean Dipole).

<strong class="journal-contentHeaderColor">Abstract.</strong> Deep learning (DL) has recently emerged as an innovative tool to downscale climate variables from large-scale atmospheric fields under the perfect-prognosis (PP) approach. Different convolutional neural networks (CNNs) have been applied under present-day conditions with promising results, but little is known about their suitability for extrapolating future climate change conditions. Here, we analyze this problem from a multi-model perspective, developing and evaluating an ensemble of CNN-based downscaled projections (hereafter DeepESD) for temperature and precipitation over the European EUR-44i (0.5<span class="inline-formula">^∘</span>) domain, based on eight global circulation models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5). To our knowledge, this is the first time that CNNs have been used to produce downscaled multi-model ensembles based on the perfect-prognosis approach, allowing us to quantify inter-model uncertainty in climate change signals. The results are compared with those corresponding to an EUR-44 ensemble of regional climate models (RCMs) showing that DeepESD reduces distributional biases in the historical period. Moreover, the resulting climate change signals are broadly comparable to those obtained with the RCMs, with similar spatial structures. As for the uncertainty of the climate change signal (measured on the basis of inter-model spread), DeepESD preserves the uncertainty for temperature and results in a reduced uncertainty for precipitation. To facilitate further studies of this downscaling approach, we follow FAIR principles and make publicly available the code (a Jupyter notebook) and the DeepESD dataset. In particular, DeepESD is published at the Earth System Grid Federation (ESGF), as the first continental-wide PP dataset contributing to CORDEX (EUR-44).

Abstract. Deep learning (DL) has recently emerged as an innovative tool to downscale climate variables from large-scale atmospheric fields under the perfect-prognosis (PP) approach. Different convolutional neural networks (CNNs) have been applied under present-day conditions with promising results, but little is known about their suitability for extrapolating future climate change conditions. Here, we analyze this problem from a multi-model perspective, developing and evaluating an ensemble of CNN-based downscaled projections (hereafter DeepESD) for temperature and precipitation over the European EUR-44i (0.5∘) domain, based on eight global circulation models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5). To our knowledge, this is the first time that CNNs have been used to produce downscaled multi-model ensembles based on the perfect-prognosis approach, allowing us to quantify inter-model uncertainty in climate change signals. The results are compared with those corresponding to an EUR-44 ensemble of regional climate models (RCMs) showing that DeepESD reduces distributional biases in the historical period. Moreover, the resulting climate change signals are broadly comparable to those obtained with the RCMs, with similar spatial structures. As for the uncertainty of the climate change signal (measured on the basis of inter-model spread), DeepESD preserves the uncertainty for temperature and results in a reduced uncertainty for precipitation. To facilitate further studies of this downscaling approach, we follow FAIR principles and make publicly available the code (a Jupyter notebook) and the DeepESD dataset. In particular, DeepESD is published at the Earth System Grid Federation (ESGF), as the first continental-wide PP dataset contributing to CORDEX (EUR-44).

<strong class="journal-contentHeaderColor">Abstract.</strong> Deep learning (DL) has recently emerged as an innovative tool to downscale climate variables from large-scale atmospheric fields under the perfect-prognosis (PP) approach. Different convolutional neural networks (CNNs) have been applied under present-day conditions with promising results, but little is known about their suitability for extrapolating future climate change conditions. Here, we analyze this problem from a multi-model perspective, developing and evaluating an ensemble of CNN-based downscaled projections (hereafter DeepESD) for temperature and precipitation over the European EUR-44i (0.5<span class="inline-formula">^∘</span>) domain, based on eight global circulation models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5). To our knowledge, this is the first time that CNNs have been used to produce downscaled multi-model ensembles based on the perfect-prognosis approach, allowing us to quantify inter-model uncertainty in climate change signals. The results are compared with those corresponding to an EUR-44 ensemble of regional climate models (RCMs) showing that DeepESD reduces distributional biases in the historical period. Moreover, the resulting climate change signals are broadly comparable to those obtained with the RCMs, with similar spatial structures. As for the uncertainty of the climate change signal (measured on the basis of inter-model spread), DeepESD preserves the uncertainty for temperature and results in a reduced uncertainty for precipitation. To facilitate further studies of this downscaling approach, we follow FAIR principles and make publicly available the code (a Jupyter notebook) and the DeepESD dataset. In particular, DeepESD is published at the Earth System Grid Federation (ESGF), as the first continental-wide PP dataset contributing to CORDEX (EUR-44).

Permafrost on the Tibetan Plateau is degrading: Historical and projected trends

The region of southern Africa (SAF) is among the most exposed climate change hotspots and is projected to experience severe impacts on multiple economical and societal sectors. For this reason, producing reliable projections of the expected impacts of climate change is key for local communities. In this work we use a set of 19 regional climate models (RCMs) performed in the context of the Coordinated Regional Climate Downscaling Experiment (CORDEX) – Africa and a set of 10 global climate models (GCMs) participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5), that were used as the driving GCMs in the RCM simulations. We are concerned about the degree to which RCM simulations are influenced by their driving GCMs, with regards to monthly precipitation climatologies, precipitation biases and precipitation change signal, according to the Representative Concentration Pathway (RCP) 8.5 for the end of the 21st century. We investigate the degree to which RCMs and GCMs are able to reproduce specific climatic features over SAF and over three sub-regions, namely the greater Angola region, the greater Mozambique region and the greater South Africa region. We identify that during the beginning of the rainy season, when regional processes are largely dependent on the coupling between the surface and the atmosphere, the impact of the driving GCMs on the RCMs is smaller, compared to the core of the rainy season, when precipitation is mainly controlled by the large-scale circulation. In addition, we show that RCMs are able to counteract the bias received by their driving GCMs, hence, we claim that the cascade of uncertainty over SAF is not additive, but indeed the RCMs do provide improved precipitation climatologies. The fact that certain bias patterns over the historical period (1985–2005) identified in GCMs are resolved in RCMs, provides evidence that RCMs are reliable tools for climate change impact studies over SAF.