Model Output Statistics Research Articles

Forecasting Summer Rainfall and Streamflow over the Yangtze River Valley Using Western Pacific Subtropical High Feature

The western Pacific subtropical high (WPSH) is one of the key systems affecting the summer rainfall over the Yangtze River Valley in China. In this study, the forecasting capacity of the WPSH for summer rainfall and streamflow is evaluated based on the WPSH index (WPSHI) derived from the NCEP/NCAR reanalysis dataset. It has been found that WPSHI can identify extreme flood years with a higher skill than normal wet years. Specifically, exceedance probability forecasting based on WPSHI has higher skills for higher thresholds of rainfall. For streamflow, adding WPSHI as a predictor only enhances the skill for higher thresholds of streamflow relative to models based on antecedent streamflow. Under the same framework, performances of two postprocessing approaches for dynamical forecasts, i.e., the model output statistics (MOS) approach and the reanalysis-based (RAN) approach are compared. Hindcasts from Climate Forecast System version 2 from the National Center for Environmental Prediction (CFSv2) are used to calculate WPSHI, which is used as the predictor for rainfall and streamflow. The result shows that the RAN approach performs better than the MOS approach. This study emphasizes the fact that the forecasting skill of exceedance probability would largely depend on the selected threshold of the predictand, and this fact should be noticed in future studies in the long-term forecasting field.

Calibrated ensemble forecasts of the height of new snow using quantile regression forests and ensemble model output statistics

Abstract. Height of new snow (HN) forecasts help to prevent critical failures of infrastructures in mountain areas, e.g. transport networks and ski resorts. The French national meteorological service, Météo-France, operates a probabilistic forecasting system based on ensemble meteorological forecasts and a detailed snowpack model to provide ensembles of HN forecasts. These forecasts are, however, biased and underdispersed. As for many weather variables, post-processing methods can be used to alleviate these drawbacks and obtain meaningful 1 to 4 d HN forecasts. In this paper, we compare the skill of two post-processing methods. The first approach is an ensemble model output statistics (EMOS) method, which can be described as a nonhomogeneous regression with a censored shifted Gamma distribution. The second approach is based on quantile regression forests, using different meteorological and snow predictors. Both approaches are evaluated using a 22 year reforecast. Thanks to a larger number of predictors, the quantile regression forest is shown to be a powerful alternative to EMOS for the post-processing of HN ensemble forecasts. The gain of performance is large in all situations but is particularly marked when raw forecasts completely miss the snow event. This type of situation happens when the rain–snow transition elevation is overestimated by the raw forecasts (rain instead of snow in the raw forecasts) or when there is no precipitation in the forecast. In that case, quantile regression forests improve the predictions using the other weather predictors (wind, temperature, and specific humidity).

India’s Maiden air quality forecasting framework for megacities of divergent environments: The SAFAR-project

Air quality is a strong health driver, its accurate assessment and forecast are important in densely populated megacities to take preventive steps. We describe the first Indian operational air quality framework, SAFAR (System of Air Quality and Weather Forecasting And Research), meant for decision-makers and a research tool with a capability of three days advance forecast in four Indian megacities of distinct environment and topography. The framework includes six different components from observations and modelling to outreach. To evaluate the performance of the forecast, we focus on particulate pollutants which largely define air quality of Indian metropolis. The model prediction skill is tested for the pilot year 2019-20 which is found to be reasonable. The Normalized Gross error of PM2.5 for Delhi is found to be highest (35%) whereas for other cities it is ∼13–20%. The Model Output Statistics (MOS) application enhanced operational forecast ability of numerical model which resulted in improving the accuracy for specific seasons (winter).

On the implementation of post-processing of runoff forecast ensembles

AbstractDifferent post-processing techniques are frequently employed to improve the outcome of ensemble forecasting models. The main reason is to compensate for biases caused by errors in model structure or initial conditions, and as a correction for under- or overdispersed ensembles. Here we use the Ensemble Model Output Statistics method to post-process the ensemble output from a continental scale hydrological model, LISFLOOD, as used in the European Flood Awareness System (EFAS). We develop a method for local calibration and interpolation of the post-processing parameters and compare it with a more traditional global calibration approach for 678 stations in Europe based on long term observations of runoff and meteorological variables. For the global calibration we also test a reduced model with only a variance inflation factor. Whereas the post-processing improved the results for the first 1-2 days lead time, the improvement was less for increasing lead times of the verification period. This was the case both for the local and global calibration methods. As the post-processing is based on assumptions about the distribution of forecast errors, we also present an analysis of the ensemble output that provides some indications of what to expect from the post-processing.

Calibration of ECMWF Seasonal Ensemble Precipitation Reforecasts in Java (Indonesia) Using Bias-Corrected Precipitation and Climate Indices

AbstractThe seasonal precipitation forecast is one of the essential inputs for economic and agricultural activities and has significant impact on decision-making. Large-scale modes of climate variability have strong relationships with seasonal rainfall in Java and are natural candidates for use as potential predictors in a statistical postprocessing application. We explore whether using climate indices as additional predictors in the statistical postprocessing of ECMWF Seasonal Forecast System 5 (SEAS5) precipitation can improve skill. We use parametric statistical postprocessing by applying a logistic distribution-based ensemble model output statistics (EMOS) technique. We add a variety of potential predictors in the analysis, namely SEAS5 raw and empirical quantile mapping (EQM) bias-corrected precipitation, Niño-3.4 index, dipole mode index (DMI), Madden–Julian oscillation (MJO) indices, sea surface temperature (SST) around Java, and several other predictors. We analyze the period of 1981–2010, focusing on July, August, September, and October. We use the continuous ranked probability skill score (CRPSS) and Brier skill score (BSS) in a comparative verification of raw, EQM, and EMOS seasonal precipitation forecasts. We have found that it is essential to use EQM-corrected precipitation as a predictor instead of raw precipitation in the latter. Besides, Niño-3.4 and DMI forecasts are not needed as extra predictors to improve monthly precipitation forecasts for the first lead month, except for September. However, for somewhat longer lead months, in September and October when there is more skill than climatology, the model that includes only Niño-3.4 and DMI forecasts as potential predictors performs about the same compared to the model that uses only EQM-corrected precipitation as a predictor.

Adjusting spatial dependence of climate model outputs with cycle-consistent adversarial networks

Climate model outputs are commonly corrected using statistical univariate bias correction methods. Most of the time, those 1d-corrections do not modify the ranks of the time series to be corrected. This implies that biases in the spatial or inter-variable dependences of the simulated variables are not adjusted. Hence, over the last few years, some multivariate bias correction (MBC) methods have been developed to account for inter-variable structures, inter-site ones, or both. As proof-of-concept, we propose to adapt a computer vision technique used for Image-to-Image translation tasks (CycleGAN) for the adjustment of spatial dependence structures of climate model projections. The proposed algorithm, named MBC-CycleGAN, aims to transfer simulated maps (seen as images) with inappropriate spatial dependence structure from climate model outputs to more realistic images with spatial properties similar to the observed ones. For evaluation purposes, the method is applied to adjust maps of temperature and precipitation from climate simulations through two cross-validation approaches. The first one is designed to assess two different post-processing schemes (Perfect Prognosis and Model Output Statistics). The second one assesses the influence of nonstationary properties of climate simulations on the performance of MBC-CycleGAN to adjust spatial dependences. Results are compared against a popular univariate bias correction method, a “quantile-mapping” method, which ignores inter-site dependencies in the correction procedure, and two state-of-the-art multivariate bias correction algorithms aiming to adjust spatial correlation structure. In comparison with these alternatives, the MBC-CycleGAN algorithm reasonably corrects spatial correlations of climate simulations for both temperature and precipitation, encouraging further research on the improvement of this approach for multivariate bias correction of climate model projections.

Multi-model Subseasonal Precipitation Forecasts over the Contiguous United States: Skill Assessment and Statistical Postprocessing

AbstractThis study assessed multi-model subseasonal precipitation forecasts (SPFs) from eight subseasonal experiment (SubX) models over the contiguous United States (CONUS) and explored the generalized extreme value distribution (GEV)-based ensemble model output statistics (EMOS) framework for postprocessing multi-model ensemble SPF. The results showed that the SubX SPF skill varied by location and season, and the skill were relatively high in the western coastal region, north-central region, and Florida peninsula. The forecast skill was higher during winter than summer seasons, especially for lead week 3 in the northwest region. While no individual model consistently outperformed the others, the simple multi-model ensemble (MME) demonstrated a higher skill than any individual model. The GEV-based EMOS approach dramatically improved the MME subseasonal precipitation forecast skill at long lead times. The continuous ranked probability score (CRPS) was improved by approximately 20% in week 3 and 43% in lead week 4; the 5-mm Brier skill score (BSS) was improved by 59.2% in lead week 3 and 50.9% in lead week 4, with the largest improvements occurring in the northwestern, north-central, and southeastern CONUS. Regarding the relative contributions of the individual SubX model to the predictive skill, the NCEP model was given the highest weight at the shortest lead time, but the weight decreased dramatically with the increase in lead time, while the CESM, EMC, NCEP, and GMAO models were given approximately equal weights for lead weeks 2-4. The presence of active MJO conditions notably increased the forecast skill in the north-central region during weeks 3-4, while the ENSO phases influenced the skill mostly in the southern regions.

Skewed and Mixture of Gaussian Distributions for Ensemble Postprocessing

The implementation of statistical postprocessing of ensemble forecasts is increasingly developed among national weather services. The so-called Ensemble Model Output Statistics (EMOS) method, which consists of generating a given distribution whose parameters depend on the raw ensemble, leads to significant improvements in forecast performance for a low computational cost, and so is particularly appealing for reduced performance computing architectures. However, the choice of a parametric distribution has to be sufficiently consistent so as not to lose information on predictability such as multimodalities or asymmetries. Different distributions are applied to the postprocessing of the European Centre for Medium-range Weather Forecast (ECMWF) ensemble forecast of surface temperature. More precisely, a mixture of Gaussian and skewed normal distributions are tried from 3- up to 360-h lead time forecasts, with different estimation methods. For this work, analytical formulas of the continuous ranked probability score have been derived and appropriate link functions are used to prevent overfitting. The mixture models outperform single parametric distributions, especially for the longest lead times. This statement is valid judging both overall performance and tolerance to misspecification.

Comparing probabilistic forecasts of the daily minimum and maximum temperature

Understanding changes in the frequency, severity, and seasonality of daily temperature extremes is important for public policy decisions regarding heat waves and cold snaps. A heat wave is sometimes defined in terms of both the daily minimum and maximum temperature, which necessitates the generation of forecasts of their joint distribution. In this paper, we develop time series models with the aim of providing insight and producing forecasts of the joint distribution that can challenge the accuracy of forecasts based on ensemble predictions from a numerical weather prediction model. We use ensemble model output statistics to recalibrate the raw ensemble predictions for the marginal distributions, with ensemble copula coupling used to capture the dependency between the marginal distributions. In terms of time series modelling, we consider a bivariate VARMA-MGARCH model. We use daily Spanish data recorded over a 65-year period, and find that, for the 5-year out-of-sample period, the recalibrated ensemble predictions outperform the time series models in terms of forecast accuracy.

Accounting for skew when post-processing MOGREPS-UK temperature forecast fields

AbstractWhen statistically post-processing temperature forecasts, it is almost always assumed that the future temperature follows a Gaussian distribution conditional on the output of an ensemble prediction system. Recent studies, however, have demonstrated that it can at times be beneficial to employ alternative parametric families when post-processing temperature forecasts, that are either asymmetric or heavier-tailed than the normal distribution. In this article, we compare choices of the parametric distribution used within the Ensemble Model Output Statistics (EMOS) framework to statistically post-process 2m temperature forecast fields generated by the Met Office’s regional, convection-permitting ensemble prediction system, MOGREPS-UK. Specifically, we study the normal, logistic and skew-logistic distributions. A flexible alternative is also introduced that first applies a Yeo-Johnson transformation to the temperature forecasts prior to post-processing, so that they more readily conform to the assumptions made by established post-processing methods. It is found that accounting for the skewness of temperature when post-processing can enhance the performance of the resulting forecast field, particularly during summer and winter and in mountainous regions.

Automatic Regional Interpretation and Forecasting System Supported by Machine Learning

The Model Output Statistics (MOS) model is a dynamic statistical weather forecast model based on multiple linear regression technology. It is greatly affected by the selection of parameters and predictors, especially when the weather changes drastically, or extreme weather occurs. We improved the traditional MOS model with the machine learning method to enhance the capabilities of self-learning and generalization. Simultaneously, multi-source meteorological data were used as the input to the model to improve the data quality. In the experiment, we selected the four areas of Nanjing, Beijing, Chengdu, and Guangzhou for verification, with the numerical weather prediction (NWP) products and observation data from automatic weather stations (AWSs) used to predict the temperature and wind speed in the next 24 h. From the experiment, it can be seen that the accuracy of the prediction values and speed of the method were improved by the ML-MOS. Finally, we compared the ML-MOS model with neural networks and support vector machine (SVM), the results show that the prediction result of the ML-MOS model is better than that of the above two models.

Postprocessing continental-scale, medium-range ensemble streamflow forecasts in South America using Ensemble Model Output Statistics and Ensemble Copula Coupling

Probabilistic hydrological forecasting and ensemble techniques have leveraged streamflow prediction at regional to continental scales up to several weeks in advance. However, ensembles that only account for meteorological forecast uncertainty are typically biased and subject to dispersion errors, thus limiting their use for rational decision-making and optimization systems. Statistical postprocessing is therefore necessary to convert ensemble forecasts into calibrated and sharp predictive distributions, and it should also account for dependencies between lead times to enable realistic forecast trajectories. This work provides a continental-scale assessment of the use of statistical postprocessing on medium-range, ensemble streamflow forecasts over South America (SA). These forecasts were produced through a large-scale hydrologic–hydrodynamic model forced with a global precipitation dataset and ECMWF reforecast data. The Ensemble Model Output Statistics (EMOS) technique was used to generate conditional predictive distributions in 488 locations at each forecast lead time, while the Ensemble Copula Coupling method with the transformation scheme (ECC-T) was applied to derive ensemble traces from EMOS distributions. Postprocessed streamflow forecasts were cross-validated for the period from 1996 to 2014 using a range of verification metrics. Results showed that the skill and reliability of EMOS forecasts substantially improve over the raw ensembles, and that EMOS leads to skillful predictions relative to discharge climatology and persistence forecasts up to 15 days in advance in most locations. Furthermore, EMOS results in predictive distributions that are generally sharper than the climatology. Limitations in depicting autocorrelations of forecast trajectories were observed in rivers for which the raw ensemble spread is very low and EMOS has to largely increase dispersion, especially at short lead times. The study’s findings suggest that combining a continental-scale hydrological model with EMOS and ECC-T methods can lead to skillful predictions and realistic ensemble traces in several locations in SA, even if in situ hydrometeorological observations are not available in real time.

Seamless Multimodel Postprocessing for Air Temperature Forecasts in Complex Topography

AbstractStatistical postprocessing is applied in operational forecasting to correct systematic errors of numerical weather prediction models (NWP) and to automatically produce calibrated local forecasts for end-users. Postprocessing is particularly relevant in complex terrain, where even state-of-the-art high-resolution NWP systems cannot resolve many of the small-scale processes shaping local weather conditions. In addition, statistical postprocessing can also be used to combine forecasts from multiple NWP systems. Here we assess an ensemble model output statistics (EMOS) approach to produce seamless temperature forecasts based on a combination of short-term ensemble forecasts from a convection-permitting limited-area ensemble and a medium-range global ensemble forecasting model. We quantify the benefit of this approach compared to only postprocessing the high-resolution NWP. The multimodel EMOS approach (“mixed EMOS”) is able to improve forecasts by 30% with respect to direct model output from the high-resolution NWP. A detailed evaluation of mixed EMOS reveals that it outperforms either one of the single-model EMOS versions by 8%–12%. Temperature forecasts at valley locations profit in particular from the model combination. All forecast variants perform worst in winter (DJF); however, calibration and model combination improves forecast quality substantially. In addition to increasing skill as compared to single-model postprocessing, it also enables us to seamlessly combine multiple forecast sources with different time horizons (and horizontal resolutions) and thereby consolidates short-term to medium-range forecasting time horizons in one product without any user-relevant discontinuity.

Improved multi‐model ensemble forecasts of Iran's precipitation and temperature using a hybrid dynamical‐statistical approach during fall and winter seasons

AbstractSkillful seasonal climate forecasts can support decision making in water resources management and agricultural planning. In arid and semi‐arid regions, tailoring reliable forecasts has the potential to improve water management by using key hydroclimate variables months in advance. This article analyses and compares the performance of two common approaches (empirical and hybrid dynamical‐statistical) in seasonal climate forecasting over a drought‐prone area located in Southwest Asia including Iran. Empirical models are framed as a baseline skill that hybrid models need to outperform. Both approaches provide probabilistic forecasts of precipitation and temperature using canonical correlation analysis to provide forecasts at 0.25° resolution. Empirical models are developed based on the large‐scale observed atmosphere–ocean patterns for forecasting using antecedent climate anomalies as predictors, while the hybrid approach makes use of model output statistics to correct systematic errors in dynamical climate model forecast outputs. Eight state‐of‐the‐art dynamical models from the North American Multi‐Model Ensemble project are analysed. Individual models with the highest goodness index are weighted to develop seven different hybrid dynamical‐statistical Multi‐model Ensembles. In this study, (October–December) and (January–February) are considered as target seasons which are the most important periods within the water year for water resource allocation to the agriculture sector. The results show that the hybrid approach has improved performance compared to the raw general circulation models and purely empirical models, and that the performance of the hybrid models is season‐dependent. Seasonal forecasts of precipitation (temperature) have a higher skill in OND (JFM). In addition, in most cases, Multi‐model Ensemble (MME) is more skillful than the empirical models and outperforms individual dynamical models. However, the best individual model might be as skillful as the MME given the target season and region of interest.

Truncated generalized extreme value distribution‐based ensemble model output statistics model for calibration of wind speed ensemble forecasts

AbstractIn recent years, ensemble weather forecasting has become a routine at all major weather prediction centers. These forecasts are obtained from multiple runs of numerical weather prediction models with different initial conditions or model parametrizations. However, ensemble forecasts can often be underdispersive and also biased, so some kind of postprocessing is needed to account for these deficiencies. One of the most popular state of the art statistical postprocessing techniques is the ensemble model output statistics (EMOS), which provides a full predictive distribution of the studied weather quantity. We propose a novel EMOS model for calibrating wind speed ensemble forecasts, where the predictive distribution is a generalized extreme value (GEV) distribution left truncated at zero (TGEV). The truncation corrects the disadvantage of the GEV distribution‐based EMOS models of occasionally predicting negative wind speed values, without affecting its favorable properties. The new model is tested on four datasets of wind speed ensemble forecasts provided by three different ensemble prediction systems, covering various geographical domains and time periods. The forecast skill of the TGEV EMOS model is compared with the predictive performance of the truncated normal, log‐normal and GEV methods and the raw and climatological forecasts as well. The results verify the advantageous properties of the novel TGEV EMOS approach.

Correcting solar radiation from reanalysis and analysis datasets with systematic and seasonal variations

Global Horizontal Irradiance (GHI), a component of solar radiation, is used to assess potential and estimate power output of solar thermal and PV applications. The reanalysis or analysis data are alternative to satellite data for solar resource assessment in places where latter is not available or costly, so there is need to evaluate their performance. The evaluation of estimated GHI (six-hourly data) from five global reanalysis and analysis datasets (NCEP-NCAR, NCEP-DOE, CFSR, NCEP-GFS and NCEP-FNL) is performed using annual, seasonal and monthly statistical analyses. The range of annual rMBE, rMAE, rRMSE and R is 13–41%, 23–41%, 44–66% and 0.938–0.965 respectively in the two inspected stations for all reanalysis and analysis datasets. Systematic errors were observed in estimated GHI which signifies need to apply bias correction to get data with acceptable uncertainty. Five bias correction methods based on Measure-Correlate-Predict, Linear-Adaptation (LA1 and LA2), CDF and Model Output Statistics (MOS) were applied to correct original GHI. The range of rMAE and rRMSE after bias-correction is 12–19% and 23–37% respectively for six-hourly data. The overall results of MOS method are best, followed by the linear-adaptation method. The corrected GHI from CFSR and NCEP-GFS datasets can be used for preliminary solar resource assessment.

Maximum Instantaneous Wind Speed Forecasting and Performance Evaluation by Using Numerical Weather Prediction and On-Site Measurement

A maximum instantaneous wind speed forecast methodology based on the autoregressive with exogenous inputs (ARX) model is proposed, in which numerical weather prediction and on-site measurement are used as inputs and the model parameters are estimated using non-parametric regression with forgetting factors. The accuracy of prediction using a proposed dynamic model is then evaluated and compared to the conventional static model output statistics (MOS) model. It is found that the prediction accuracy is improved by utilizing numerical weather prediction with a higher horizontal resolution. Finally, the predictability of the maximum instantaneous wind speed higher than 15m/s is evaluated using the receiver operating characteristic (ROC) curve and the area under the curve (AUC). The optimal quantile level of the maximum instantaneous wind speed is derived using a cost function.

Probabilistic forecast of low-level wind shear over Jeju international airport using non-homogeneous regression model

Ensemble verification of low-level wind shear (LLWS) is an important issue in airplane landing operation and management. However, there have been few studies on the probabilistic forecasts of LLWS obtained from ensemble prediction system. In this study, we analyzed a reliability analysis to verify LLWS ensemble member forecasts and observation based on the limited grid points around Jeju International Airport in Jeju. Homogeneous and non-homogeneous regression models were used to reduce the bias and dispersion existing ensemble prediction system and to provide probabilistic forecast. Prior to applying probabilistic forecast model, reliability analysis was conducted by using rank histogram to identify the statistical consistency of LLWS ensemble forecasts and corresponding observations. Based on the results of our study, we found that LLWS ensemble forecasts had a consistent positive bias, indicating over-forecasting, and were under-dispersed for all seasons. To correct such biases, homogeneous regression and non-homogeneous regressions as EMOS (Ensemble Model Output Statistics) and EMOS exchangeable model by assuming exchangeable ensemble members were applied. The prediction skills of the methods were compared by the mean absolute error and continuous ranked probability score. We found that the prediction skills of probabilistic forecasts of EMOS exchangeable model were superior to the bias-corrected forecasts in terms of deterministic prediction.

Probabilistic Forecasting of the 500 hPa Geopotential Height over the Northern Hemisphere Using TIGGE Multi-model Ensemble Forecasts

Bayesian model averaging (BMA) and ensemble model output statistics (EMOS) were used to improve the prediction skill of the 500 hPa geopotential height field over the northern hemisphere with lead times of 1–7 days based on ensemble forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP), and UK Met Office (UKMO) ensemble prediction systems. The performance of BMA and EMOS were compared with each other and with the raw ensembles and climatological forecasts from the perspective of both deterministic and probabilistic forecasting. The results show that the deterministic forecasts of the 500 hPa geopotential height distribution obtained from BMA and EMOS are more similar to the observed distribution than the raw ensembles, especially for the prediction of the western Pacific subtropical high. BMA and EMOS provide a better calibrated and sharper probability density function than the raw ensembles. They are also superior to the raw ensembles and climatological forecasts according to the Brier score and the Brier skill score. Comparisons between BMA and EMOS show that EMOS performs slightly better for lead times of 1–4 days, whereas BMA performs better for longer lead times. In general, BMA and EMOS both improve the prediction skill of the 500 hPa geopotential height field.

Novel multivariate quantile mapping methods for ensemble post-processing of medium-range forecasts

Statistical post-processing is an indispensable tool for providing accurate weather forecasts and early warnings for weather extremes. Most statistical post-processing is univariate, with dependencies introduced via use of an empirical copula. Standard empirical methods take a dependence template from either the raw ensemble output (ensemble copula coupling, ECC) or the observations (Schaake Shuffle, SSh). There are drawbacks to both methods. In ECC it is assumed that the raw ensemble simulates the dependence well, which is not always the case (e.g. 2-meter temperature in The Netherlands). The Schaake Shuffle is not able to capture flow dependent changes to the dependence and the choice of observations is key. Here we compare a reshuffled standard ensemble model output statistics (EMOS) approach with two multivariate bias adjustment approaches that have not been used before in a post-processing context: 1) the multivariate bias correction with N-dimensional probability density function transform (MBCn) and 2) multivariate ranks that are defined with optimal assignment (OA). These methods have the advantage that they are able to explicitly capture the dependence structure that is present in the observations. We apply ECC, the Schaake Shuffle, MBCn and OA to 2-m and dew point temperature forecasts at seven stations in The Netherlands. Forecasts are verified with both univariate and multivariate methods, and using a heat index derived from both variables, the wet-bulb globe temperature (WBGT). Our results demonstrate that the spatial and inter-variable dependence structure is more realistic in MBCn and OA compared to ECC or the Schaake Shuffle. The variogram score shows that while ECC is most skilful for the first two days, at moderate lead times MBCn is most skilful and at the longest lead times OA is more skilful than both ECC and MBCn. Overall, we highlight the importance of considering the dependence between variables and locations in the statistical post-processing of weather forecasts.