Statistical and dynamical downscaling techniques are widely applied in the development of local climate change scenarios. This talk traces the conceptual development of downscaling as a decision-support tool for climate risk assessment, resilience and adaptation planning. Four epochs are identified: (1) early exploration of local changes in key climate variables, such as temperature and precipitation extremes; (2) application of downscaled scenarios to climate impacts modelling (such as for agriculture yield or water resource assessments); (3) advent of ensemble-based methods and more sophisticated handling of uncertainty in the downscaling-impacts workflow; and (4) use of downscaled scenarios to stress-test adaptation options under plausible ranges of climate and non-climatic conditions. Each phase is illustrated by and reflected in the development of the Statistical DownScaling Model – Decision Centric (SDSM-DC) over more than two decades. Questions around fitness for purpose and appropriate uses of the tool are explored. The talk concludes by considering: where next for downscaling?

How to cite: Wilby, R. and Dawson, C.: Conceptual development and use of downscaled climate model information, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-3343, https://doi.org/10.5194/egusphere-egu23-3343, 2023.

Statistical bias adjustment is now common practice when using climate models for impact studies, prior to or in conjunction with downscaling methods. Examples of widely used methodologies include CDFt (Vrac et al. 2016), ISIMIP3BASD (Lange 2019) or equidistant CDF matching (Li et al. 2010). Though common practice, recent papers (Maraun et al. 2017) have found fundamental issues with statistical bias adjustment. When multivariate aspects are not evaluated, improper use of bias adjustment is not detected. Fundamental misspecifications of the climate model, such as the displacement of large-scale circulation, cannot be corrected. Furthermore, results are sensitive to internal climate variability over the reference period (Bonnet et al 2022). If applied, bias adjustment methods should therefore be evaluated carefully in multivariate aspects and targeted to the use-case at hand.

However, good practice in the evaluation and application of bias adjustment methods is inhibited by what we frame as practical issues. If at all, published bias adjustment methods are often published as individual software packages across different programming languages (mostly R and Python) that do not allow users to adapt aspects of the method, such as the fit distribution, to their use-case. Existing open-source software packages, such as ISIMIP3BASD or CDFt, often do not offer an evaluation framework that covers multivariate (spatial, temporal, multi-variable) aspects necessary to detect misuse of methods, or user-specific impact metrics. Several of these issues apply to downscaling similarly.

To address some of these practical issues, we developed the open-source software package ibicus in collaboration with ECWMF (available on PyPi, extensive documentation https://ibicus.readthedocs.io/en/latest/index.html, published under Apache 2.0 licence). The package implements eight peer-reviewed bias adjustment methods in a common framework. It also includes an extensive evaluation framework covering multivariate aspects as well as the ETCCDI climate indices. The package thereby contributes to enhanced flexibility and ease-of-use of better evaluation practises in bias adjustment.

Our contribution presents three case studies using ibicus, highlighting a number of pitfalls in the usage of bias adjustment for climate impact modelling, and shows possible ways to address these issues. We investigate extreme indices of precipitation and compound extreme temperature-precipitation indices, modification of the climate change trend, and dry spell length as an example of a temporal index, over northern Spain and Turkey.

We evaluate how bias adjustment adds to the ‘cascade of uncertainty’ and how this can be made transparent in the different use-cases. We also demonstrate how some of the fundamental issues that can arise when applying bias adjustment can be detected and how evaluation of spatial and temporal aspects such as dry spell length can be made specific to the use-case at hand to detect improper use of bias adjustment. Lastly, we demonstrate how the ‘best’ bias adjustment method may depend on the metric of interest, and therefore a user-centric design of comparison and evaluation methods is necessary.

How to cite: Spuler, F., Wessel, J., Cagnazzo, C., and Comyn-Platt, E.: Case studies in bias adjustment: addressing potential pitfalls through model comparison and evaluation using a new open-source python package, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14254, https://doi.org/10.5194/egusphere-egu23-14254, 2023.

Two main approaches to downscale global climate projections are possible: dynamical and statistical downscaling. Both families have been widely evaluated, but intercomparison studies between the two families are scarce, and usually limited to temperature and precipitation. In this work, we present a comparison between a Statistical Downscaling Model (SDM) based on Machine Learning and six Regional Climate Models (RCMs) from EURO-CORDEX, for five variables of interest: temperature, precipitation, wind, humidity and solar radiation. The study expands at a continental scale over Europe, with a spatial resolution of 0.11^o and daily data. Both the SDM and the RCMs are driven by the ERA-Interim reanalysis, and observations are taken from the gridded dataset E-OBS. Several aspects have been evaluated: daily series, mean values and extremes, spatial patterns and also temporal aspects. Additionally, in order to analyze the intervariable consistency, a multivariable index (Fire Weather Index) derived from the fundamental variables has been included. The SDM has reached better scores than the RCMs for all the evaluated aspects with only a few exceptions, mainly related to an underestimation of the variance. After bias correction, both the SDM and the six RCMs present similar results, with no significant differences among them. Results presented here, combined with the low computational expense of SDMs and the limited availability of RCMs over some CORDEX domains, should motivate the consideration of statistical downscaling at the same level as RCMs by official providers of regional information, and its inclusion in reference sites. Nonetheless, further analysis on crucial aspects such as the impact on long-term trends or the sensitivity of different methods to being driven by Global Climate Models instead of by a reanalysis, is needed.

How to cite: Hernanz, A., Correa, C., Domínguez, M., Rodríguez-Guisado, E., and Rodríguez-Camino, E.: Can Statistical Downscaling based on Machine Learning compete with Regional Climate Models? A comparison for temperature, precipitation, wind, humidity and radiation over Europe under present conditions, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1296, https://doi.org/10.5194/egusphere-egu23-1296, 2023.

The long-existing double-ITCZ problem in GCMs affects not only the models' ability in simulating the current climate, but also implies limitations regarding the assessment of climate sensitivity and global climate change. Using a regional climate model (RCM) with explicit convection at a horizontal grid spacing of 12 km in a large computational domain covering the tropical and sub-tropical Atlantic, we develop a bias-correction downscaling methodology to remove the biases of a driving GCM. The methodology is related to the pseudo-global warming (PGW) approach. Normally this method is used to impress the climate-change signal to a reanalysis-driven RCM simulation, but it can also be used to modulate the lateral-boundary conditions of a GCM, such as to remove the large-scale biases. We show that the double ITCZ problem persists with classical dynamical downscaling (i.e. when driving the RCM directly by the GCM output), but with our bias-corrected downscaling the double ITCZ problem can be removed. Detailed analysis reveals that the main cause of the double ITCZ problem can be attributed to the GCMs' SST bias. Compared to the GCMs' AMIP simulations, RCMs with higher resolution allow explicit deep convection and enable a better simulation of tropical convection and clouds. By improving the corresponding radiative forcing, vertical motion is better simulated. Subsidence stronger to the south of the ITCZ pushes the ITCZ more north in the boreal spring, which is consistent with the observation of the ITCZ. The developed methodology provides an opportunity for better constraining climate sensitivity by removing double-ITCZ biases.

How to cite: Liu, S., Zeman, C., and Schär, C.: Understanding the double-ITCZ problem over the Atlantic with bias-corrected downscaling, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11595, https://doi.org/10.5194/egusphere-egu23-11595, 2023.

Regional climate models, due to their systematic biases, are not usable for impact assessment and policy-relevant applications. It is common to post-process the regional model outputs with appropriate bias correction methodologies to provide reliable climate change information. We apply a distribution-based, trend-preserving quantile mapping procedure to bias correct the projections of daily precipitation and temperature from an ensemble of 5 RCMs driven by 5 GCMs, each at a resolution of 0.11°, chosen from the EURO-CORDEX initiative. The gridded observations from the German Weather Service, DWD-HYRAS, has been used as a reference for the bias correction. The impact of the bias correction is found to be more pronounced on precipitation than on temperature, as the precipitation biases are larger. The models are wetter and underestimate (overestimate) the daily maximum (minimum) temperature. The correction method eliminates large parts of these biases and maps the distributions of both the variables well with that of observations. The bias adjustment also leads to the narrowing down of the uncertainties in the projected changes of both the variables. The decomposition of total variance into model uncertainty and internal variability suggests that the bias correction acts mostly on the former component. The internal variability component does not seem, however, to undergo considerable changes following the bias correction. Due to the reduction of the uncertainty, we find a slight improvement in the signal-to-noise ratio in the projections. 

How to cite: Adakudlu, M., Xoplaki, E., and Luther, N.: Implications of statistical bias adjustment for uncertainties in regional model projections, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4495, https://doi.org/10.5194/egusphere-egu23-4495, 2023.

Dynamical downscaling is a widely-used approach to generate regional projections of future climate extremes at a finer scale. Previous studies indicated that the global climate model (GCM) bias correction method prior to dynamical downscaling can improve the simulation of the climate extreme to a certain extent. Recently, a new bias correction method termed MVT was developed. Note that this method did not correct the GCM biases of the climate extreme event explicitly. In this study, we evaluate the MVT method in terms of various climate extreme events through three dynamical downscaling simulations over Asia-western North Pacific with 25 km grid spacing throughout 1980–2014, and further investigate to what extent and how this bias correction method can improve the simulation of downscaled climate extreme events. The dynamical downscaling simulations driven by the original GCM dataset derived from the MPI-ESM1-2-HR (hereafter WRF_GCM), the bias-corrected GCM (hereafter WRF_GCMbc) are validated against that driven by the European Centre for Medium-Range Weather Forecasts Reanalysis 5 dataset, respectively. The results suggest that compared with the WRF_GCM, the WRF_GCMbc shows more than 26% decrease in root mean square errors of the precipitation and temperature extreme indices, and even 61% out of seasonal extreme indices show more than 50% reduction. Such improvements in the WRF_GCMbc are primarily caused by the correct simulation of the large-scale circulation due to the GCM bias correction. The large-scale circulation in turn improves the simulation of the precipitation and cloud by the water vapor transport and further improves the simulation of the 2m temperature by the radiation process and the surface energy balance, which contribute to the better simulation of the precipitation and temperature extreme indices.

How to cite: Zhang, M.-Z., Han, Y., Xu, Z., and Guo, W.: Validation of MVT bias correction in dynamical downscaling simulations for climate extreme, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-3741, https://doi.org/10.5194/egusphere-egu23-3741, 2023.

The nature and severity of climate change impacts varies significantly from region to region. Consequently, high-resolution climate information is needed for meaningful impact assessments and the design of mitigation strategies. This demand has lead to an increase in the coupling of Empirical Statistical Downscaling (ESD) models to General Circulation Model (GCM) simulations of future climate. Here, we present a new open-source Python package (pyESD; github.com/Dan-Boat/PyESD) that implements several Perfect Prognosis ESD (PP-ESD) methods and the whole downscaling cycle. The latter includes routines for data preparation, predictor selection and construction, model selection and training, evaluation, utility tools for relevant statistical tests, visualisation, and more. The package includes a collection of well-established Machine Learning algorithms and allows the user to choose a variety of estimators, cross-validation schemes, objective function measures, hyperparameter optimization, etc., in relatively few lines of codes. The package is highly modular and flexible, and allows quick and reproducible downscaling of any climate information, such as precipitation, temperature, wind speed or even glacial retreat. We demonstrate the effectiveness of the new PP-ESD framework by generating station-based downscaling products of precipitation and temperature for complex mountainous terrain in Southwest Germany.

How to cite: Mutz, S. G. and Boateng, D.: pyESD: An open-source Python framework for empirical-statistical downscaling of climate information, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7470, https://doi.org/10.5194/egusphere-egu23-7470, 2023.

Bias adjustment of climate models is today normally performed with quantile mapping methods that account for the whole distribution of the parameter. The bulk of the distribution is well described as long as sufficient data records are used (Berg et al., 2012), however, the extreme tails will always suffer from large uncertainties. These uncertainties stem from both the climate model and the reference data set, which prevents a robust and detailed identification of bias in the extreme tail. Empirical quantile mapping methods are therefore prone to overfitting, and may introduce substantial bias when applied outside the calibration period. Commonly, a constant adjustment is applied for values outside the range of the calibration period, but there is room for improvements of the extrapolation method.

While working with a climate service for Sweden, a clear offset was identified between data adjusted within and outside the calibration period for an extreme indicator of daily maximum precipitation. This study explores different extrapolation methods for the extreme tail of the distribution in the spline-based empirical quantile mapping method of the MIdAS bias adjustment method (Berg et al., 2022). By limiting the bias adjustment to the first 95% of the distribution, and thereafter applying a constant or a linear fit to the remaining 5% of data in the tail, the offset is strongly reduced and the adjusted extremes become more robust and plausible.

Berg, P., Feldmann, H., & Panitz, H. J. (2012). Bias correction of high resolution regional climate model data. Journal of Hydrology, 448, 80-92.

Berg, P., Bosshard, T., Yang, W., & Zimmermann, K. (2022). MIdASv0. 2.1–MultI-scale bias AdjuStment. Geoscientific Model Development, 15(15), 6165-6180.

How to cite: Berg, P., Bosshard, T., Bärring, L., Södling, J., Wilcke, R., Yang, W., and Zimmermann, K.: Reducing negative impacts of bias adjustment on the distribution tail and extreme climate indicators in MIdAS, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11124, https://doi.org/10.5194/egusphere-egu23-11124, 2023.

Extreme temperatures, especially long-lasting heat and cold waves in urban areas, lead to thermal stress of the population and increase the number of weather-related health risks and deaths. The observed climate trend and the associated increase of extreme weather events are expected to continue in the future. Thus, the evaluation of urban thermal stress and the associated health effects becomes an important issue for urban planning and risk management. For Austrian cities, an information system for temperature warnings already exists (Weather warnings, ZAMG), which is based on the information of regional weather forecast models. However, this information does not have the required spatial resolution needed to resolve urban structure and thus to account for the urban heat island effect or cold stress situations in winter.

The aim of this project is to provide the basis for the improvement of extreme weather/thermal (dis)comfort warning systems in Austrian major cities by using high-resolution weather predictions (100 m). Therefore, the soil model SURFEX (developed by Météo France) coupled with the AROME numerical weather forecast model is applied to selected cities in Austria and used to determine the best model configuration to compute short-term forecasts (+60 hours). This method provides not a full dynamical model, but a way of pyhsical downscaling with height corrections and a high-resolution surface model.

In this project, land use parameterization will be updated and improved based on Pan-European High Resolution Layers (e.g. Urban Atlas) of the Copernicus Land Monitoring service in ECOCLIMAP (predefined land use classes for SURFEX). The model output will be verified with in-situ operational and crowd-sourced observations. Furthermore, the results will be compared to the micro-scale urban climate model MUKLIMO_3 from the German Weather Service (100 m) and various thermal infrared (TIR with 150 to 250 m) datasets. The novel modeling approach for simulating thermal stress in urban areas serves as the basis for improving the operational prediction system of extreme temperatures, for optimizing the future extreme weather warning system at the ZAMG, and for decision-making for the involved cities and their stakeholders.

How to cite: Oswald, S., Schneider, S., Zuvela-Aloise, M., Hahn, C., and Wastl, C.: Improvement and verification of urban extreme temperature predictions with satellite and ground observations in Austria (VERITAS-AT), EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-164, https://doi.org/10.5194/egusphere-egu23-164, 2023.

The reliable prediction of flash flood relevant heavy precipitation events under climate change conditions remains a challenging task for the downscaling community. Therefore, a huge variety of downscaling approaches have been proposed and successfully applied, however, there is still potential for improvements. The conducted study aims to investigate potential improvements by circulation pattern (CP) trends conservation and their utilization for CP conditional statistical downscaling of daily summer precipitation in the (pre-)alpine region of Bavaria. The CPs have been created taking only atmospheric variables into consideration and the link to precipitation is established via CP conditional cumulative distribution functions (CDF) of the observed precipitation at selected measurement sites across the region. The derived CDFs allow for the sampling of CP conditional precipitation values at the station scale which are subsequently bias corrected by quantile mapping (QM) and parametric transfer functions (PTFs) as tested methods. The predicted precipitation values have been evaluated against obervations using different performance measures such as Kling-Gupta Efficiency (KGE), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). In order to properly account for extreme events the evaluation has been conducted for the complete precipitation distribution and for the distribution above the 95th percentile seperately. The results show that the described CP conditional downscaling approach is capable of yielding more accurate daily precipitation values especially in the extremes compartment in which an average gain in prediction skill of + 0.24 and a maximum gain of + 0.6 in terms of KGE has been observed. This shows that the conservation of trends and atmospheric information through CPs and their utilization for downscaling can lead to improved precipitation downscaling results.

How to cite: Böker, B., Laux, P., Olschewski, P., and Kunstmann, H.: Accurate heavy precipitation prediction in an (pre-)alpine area: The benefit of trend conservation in circulation pattern conditional statistical downscaling., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2260, https://doi.org/10.5194/egusphere-egu23-2260, 2023.

The lack of long-term and consistent meteorological observations limits the application of land-surface simulators (e.g., of phenomena in hydrology, the cryosphere, ecology) at remote locations. For example, most permafrost areas are remote and lacking consistent meteorological time series, models that describe permafrost change over time cannot be driven for comparison with observations or for impact studies. Reanalysis-derived time series are valuable because they are available with global coverage, for a long time period, and for a broad set of physically consistent variables. Multiple reanalyses can be used to provide estimates of uncertainty. Practically, however, this data is difficult to use for several reasons: grid-scale reanalyses must be downscaled and interpolated horizontally (and vertically within the atmospheric column for mountains regions) to the site‑scale, differences in variables, units, and delivery between reanalyses must be reconciled, and large volumes of data need to be handled. Globsim is an open-source python library (available via GitHub) that was developed to handle these challenges and to facilitate a simulation workflow that takes advantage of the multiple reanalysis products available today. It outputs sub-daily meteorological time series that resemble meteorological stations for any location on the planet. Since the release of the first version of Globsim, we have improved usability, refactored code for maintainability and speed, and fixed a number of bugs. We also added support for ERA5 ensemble data, and added more sophisticated heuristic downscaling algorithms, including TOPOscale for elevation-adjusted radiative fluxes. We use Globsim as a core tool in a multi-model permafrost simulation workflow and, as a future step, we intend to use it as part of a debiasing routine to make predictions of permafrost using climate scenarios. We expect this tool to be broadly applicable to climate change impact modelers and other scientists using climate driven simulations working in (remote) locations that lack meteorological data of sufficient quality and duration for their application.

How to cite: Brown, N., Gruber, S., and Cao, B.: Globsim v.3 – Improvements to an open-source software library for utilizing atmospheric reanalyses in point-scale land surface simulation, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10595, https://doi.org/10.5194/egusphere-egu23-10595, 2023.

Climate simulations often need to be adjusted before carrying out climate impact studies at regional scale in order to reduce the biases often present in climate models. To do that, bias adjustment methods are usually applied to climate output simulations and are calibrated over a reference period. This period ideally includes good observational coverage and is often defined as the 2 or 3 more recent decades. However, on these timescales, the climate state may be influenced by the low-frequency internal climate variability. There is therefore a risk of introducing a bias to the climate projections by bias-adjusting simulations with low-frequency variability in a different phase to that of the observations. We proposed here a new pseudo-reality framework using an ensemble of simulations performed with the IPSL-CM6A-LR climate model in order to assess the impact of the low-frequency internal climate variability of the North Atlantic sea surface temperatures on bias-adjusted projections of mean and extreme surface temperature over Europe. We show that adjusting a simulation in a similar phase of the Atlantic Multidecadal Variability to that of the pseudo-observations reduces the pseudo-biases in temperature projections. Therefore, for models and regions where low frequency internal variability matters, it is recommended to sample relevant climate simulations to be bias adjusted in a model ensemble or alternatively to use a very long reference period when possible.

How to cite: Bonnet, R., Vrac, M., Boucher, O., and Jin, X.: Sensitivity of bias adjustment methods to low-frequency internal climate variability over the reference period: an ideal model study, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6529, https://doi.org/10.5194/egusphere-egu23-6529, 2023.

In this study, the predictability of the coastal ocean is assessed in a downscaling exercise for the Adriatic Sea using NEMO 3.6 over a 19 years’ time window (2001-2019). Inspired by the perfect model approach (Denis et al. 2002, De Elia et al. 2002) using a dynamical downscaling setup, a high resolution (2 km) experiment (Big Brother – BB) for the entire Adriatic Sea is used as the “true” reference for a smaller domain, downscaling experiment (Little Brother – LB) in the Northern Adriatic subbasin. The LB experiment has the same horizontal resolution as the BB (2 km) and is downscaled from a low resolution parent model (6 km), in a ratio of 1/3 resolution jump. The 2 km horizontal resolution fits the purpose of reaching an eddy-permitting grid spacing in the Adriatic basin (Masina and Pinardi, 1994; Cushman-Roisin et al. 2002).

Power spectral density analysis is used to evaluate the kinetic energy variance on the frequency domain among the experiments and compare them with the BB experiment. Overall, the LB is more energetic than the parent model, and the timing of the peaks of energy coincides with the ones of the BB. The energy on the 1 year signal is higher in the LB than the BB. The LB can recover a significant amount of energy for all peaks, with special attention to the 6 months period, which is poorly captured by the parent model. The 4 months signal is equally represented in BB and LB, while there is an underestimation of the 6 months signal of LB with respect to BB. Energy in the LB does not deviate from BB more than ~20% in the low frequencies and ~10% in the high frequencies, while the parent model presents in a whole lower energy than the BB, with higher differences on the low frequencies.

The Northern Adriatic circulation is largely influenced by the surface buoyancy flux and the wind forcing (Cessi et al., 2014), which play a significant role in the energy budget and the anti-estuarine overturning circulation of the Adriatic basin. Differences between LB and parent model results may be associated with the energy cascade due to interactions of internal dynamic processes which are differently represented at different resolutions. Differences between LB and BB results are the effect of the downscaling method and the horizontal resolution ratio between the parent model and the nested LB.

Moreover, the analysis of the wavenumber spectra allows a clear overview of the energy distribution in the space domain among the experiments and the representation of small-scale features in the LB. Small scale features less than twice the grid spacing (~12 km) are absent in the low-resolution parent model outputs. Therefore, the comparison with the true reference, BB, reveals the energy spectrum of the parent model solves only the larger scales, while the downscaling LB can recover the smaller scales absent in the initial and lateral boundary conditions.

How to cite: Tatsch Eidt, R., Verri, G., Santos da Costa, V., Gunduz, M., and Navarra, A.: A downscaling exercise for the Adriatic Sea in a perfect model approach, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12899, https://doi.org/10.5194/egusphere-egu23-12899, 2023.

Copernicus Climate Change Service (C3S) integrates multiple seasonal forecast models of climate variables with multiple ensemble realizations. Assessing the risks of natural hazards with high impacts on human and natural systems and providing actionable services at the local scale require high-resolution predictions. We implement the AI-based approach proposed by Heidari et al. (2023) to address such needs and reach a kilometer scale. While downscaling seasonal forecasts, it is crucial to transfer the full range of the uncertainties given by the ensembles.

This study assesses how uncertainty is transferred by an AI-based downscaling approach. Quantile-based metrics are here used to measure the ensemble variability between seasonal forecasts and their downscaled products. On the other side, quantile-based metrics can also give an alternative description of the ensemble variabilities, which could replace the raw ensemble members in the downscaling process. In this study, the AI-downscaling system is tested by inputting (a) raw ensemble members and (b) quantile-based metrics. Transferred uncertainty and downscaling accuracy are then evaluated to develop and implement an optimal downscaling approach with hazard-dependent inputs being selected at  regional and local scales.

Heidari F., Lin Q., Espitia Sarmiento E.F., Toreti A., and Xoplaki E. (2023): A deep learning technique to realistically bias correct and downscale seasonal forecast ensembles of climate variables towards the development of an AI-based early warning system, EGU 2023 abstract

How to cite: Lin, Q., Heidari, F., Espitia Sarmiento, E. F., Adakudlu, M., Toreti, A., and Xoplaki, E.: Analysis of Ensemble Uncertainty Transfer in AI-Based Downscaling of C3S Seasonal Forecast, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9644, https://doi.org/10.5194/egusphere-egu23-9644, 2023.

Enel, as most of the Energy Players, has an important exposure on weather risk due to the indirect effect of the power demand and to the direct effects on renewable production. A large component of such risk comes from the hydroelectric production, this is especially true in Southern America where, in some countries, it can represent up to 70% of the total production. We present a practical development of an operational chain to extract information from the seasonal forecasts produced by SEAS5. It works on some catchments in Colombia and Peru with the aim to provide an ensemble forecast of monthly precipitations at a high resolution from the fields at low resolution provided by Copernicus. To produce the high-resolution fields of precipitations we developed a procedure based on Lorenz et al. (2021); for our scope, the biases of the SEAS5 forecasts are corrected following a reference climatology obtained from the SEAS5 hindcasts that is calibrated over the cumulative distribution function calculated be mean of historical measurements of the IDEAM weather stations. The method and preliminary results as well as the validation will be shown in this work.

How to cite: Rea, G., Galuzzo, D., and Formenton, M.: Development of an operational seasonal forecast in Colombia and Peru by mean of statistical downscaling of the SEAS5-Copernicus data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12395, https://doi.org/10.5194/egusphere-egu23-12395, 2023.