Model Output Statistics Research Articles

Intercomparison of prediction skills of ensemble methods using monthly mean temperature simulated by CMIP5 models

This study focuses on an objective comparison of eight ensemble methods using the same data, training period, training method, and validation period. The eight ensemble methods are: BMA (Bayesian Model Averaging), HMR (Homogeneous Multiple Regression), EMOS (Ensemble Model Output Statistics), HMR+ with positive coefficients, EMOS+ with positive coefficients, PEA_ROC (Performance-based Ensemble Averaging using ROot mean square error and temporal Correlation coefficient), WEA_Tay (Weighted Ensemble Averaging based on Taylor’s skill score), and MME (Multi-Model Ensemble). Forty-five years (1961-2005) of data from 14 CMIP5 models and APHRODITE (Asian Precipitation- Highly-Resolved Observational Data Integration Towards Evaluation of Water Resources) data were used to compare the performance of the eight ensemble methods. Although some models underestimated the variability of monthly mean temperature (MMT), most of the models effectively simulated the spatial distribution of MMT. Regardless of training periods and the number of ensemble members, the prediction skills of BMA and the four multiple linear regressions (MLR) were superior to the other ensemble methods (PEA_ROC, WEA_Tay, MME) in terms of deterministic prediction. In terms of probabilistic prediction, the four MLRs showed better prediction skills than BMA. However, the differences among the four MLRs and BMA were not significant. This resulted from the similarity of BMA weights and regression coefficients. Furthermore, prediction skills of the four MLRs were very similar. Overall, the four MLRs showed the best prediction skills among the eight ensemble methods. However, more comprehensive work is needed to select the best ensemble method among the numerous ensemble methods.

Comparison of hourly surface downwelling solar radiation estimated from MSG–SEVIRI and forecast by the RAMS model with pyranometers over Italy

Abstract. In this paper, we evaluate the performance of two global horizontal solar irradiance (GHI) estimates, one derived from Meteosat Second Generation (MSG) and another from the 1-day forecast of the Regional Atmospheric Modeling System (RAMS) mesoscale model. The horizontal resolution of the MSG-GHI is 3 × 5 km2 over Italy, which is the focus area of this study. For this paper, RAMS has the horizontal resolution of 4 km.The performances of the MSG-GHI estimate and RAMS-GHI 1-day forecast are evaluated for 1 year (1 June 2013–31 May 2014) against data of 12 ground-based pyranometers over Italy spanning a range of climatic conditions, i.e. from maritime Mediterranean to Alpine climate.Statistics for hourly GHI and daily integrated GHI are presented for the four seasons and the whole year for all the measurement sites. Different sky conditions are considered in the analysisResults for hourly data show an evident dependence on the sky conditions, with the root mean square error (RMSE) increasing from clear to cloudy conditions. The RMSE is substantially higher for Alpine stations in all the seasons, mainly because of the increase of the cloud coverage for these stations, which is not well represented at the satellite and model resolutions. Considering the yearly statistics computed from hourly data for the RAMS model, the RMSE ranges from 152 W m−2 (31 %) obtained for Cozzo Spadaro, a maritime station, to 287 W m−2 (82 %) for Aosta, an Alpine site. Considering the yearly statistics computed from hourly data for MSG-GHI, the minimum RMSE is for Cozzo Spadaro (71 W m−2, 14 %), while the maximum is for Aosta (181 W m−2, 51 %). The mean bias error (MBE) shows the tendency of RAMS to over-forecast the GHI, while no specific behaviour is found for MSG-GHI.Results for daily integrated GHI show a lower RMSE compared to hourly GHI evaluation for both RAMS-GHI 1-day forecast and MSG-GHI estimate. Considering the yearly evaluation, the RMSE of daily integrated GHI is at least 9 % lower (in percentage units, from 31 to 22 % for RAMS in Cozzo Spadaro) than the RMSE computed for hourly data for each station. A partial compensation of underestimation and overestimation of the GHI contributes to the RMSE reduction. Furthermore, a post-processing technique, namely model output statistics (MOS), is applied to improve the GHI forecast at hourly and daily temporal scales. The application of MOS shows an improvement of RAMS-GHI forecast, which depends on the site considered, while the impact of MOS on MSG-GHI RMSE is small.

Multimodel probabilistic prediction of 2 m-temperature anomalies on the monthly timescale

Abstract. The 2 m-temperature anomalies from the reforecasts of the CNR-ISAC and ECMWF monthly prediction systems have been combined in a multimodel super-ensemble. Tercile probability predictions obtained from the multimodel have been constructed using direct model outputs (DMO) and model output statistics (MOS), like logistic and nonhomogeneous Gaussian regression, for the 1990–2010 winter seasons. Verification with ERA-Interim reanalyses indicates that logistic regression gives the best results in terms of ranked probability skill scores (RPSS) and reliability diagrams for low–medium forecast probabilities. Also, it is argued that the logistic regression would not yield further improvements if a larger dataset was used.

How Suitable is Quantile Mapping For Postprocessing GCM Precipitation Forecasts?

GCMs are used by many national weather services to produce seasonal outlooks of atmospheric and oceanic conditions and fluxes. Postprocessing is often a necessary step before GCM forecasts can be applied in practice. Quantile mapping (QM) is rapidly becoming the method of choice by operational agencies to postprocess raw GCM outputs. The authors investigate whether QM is appropriate for this task. Ensemble forecast postprocessing methods should aim to 1) correct bias, 2) ensure forecasts are reliable in ensemble spread, and 3) guarantee forecasts are at least as skillful as climatology, a property called “coherence.” This study evaluates the effectiveness of QM in achieving these aims by applying it to precipitation forecasts from the POAMA model. It is shown that while QM is highly effective in correcting bias, it cannot ensure reliability in forecast ensemble spread or guarantee coherence. This is because QM ignores the correlation between raw ensemble forecasts and observations. When raw forecasts are not significantly positively correlated with observations, QM tends to produce negatively skillful forecasts. Even when there is significant positive correlation, QM cannot ensure reliability and coherence for postprocessed forecasts. Therefore, QM is not a fully satisfactory method for postprocessing forecasts where the issues of bias, reliability, and coherence pre-exist. Alternative postprocessing methods based on ensemble model output statistics (EMOS) are available that achieve not only unbiased but also reliable and coherent forecasts. This is shown with one such alternative, the Bayesian joint probability modeling approach.

Near-Surface Wind Observation Impact on Forecasts: Temporal Propagation of the Analysis Increment

AbstractThis study examines the assimilation of near-surface wind observations over land to improve wind nowcasting and short-term tropospheric forecasts. A new geostatistical operator based on geophysical model output statistics (GMOS) is compared with a bilinear interpolation scheme (Bilin). The multivariate impact on forecasts and the temporal evolution of the analysis increments produced are examined as well as the influence of background error covariances on different components of the prediction system. Results show that Bilin significantly degrades surface and upper-air fields when assimilating only wind data from 4942 SYNOP stations. GMOS on the other hand produces smaller increments that are in better agreement with the model state. It leads to better short-term near-surface wind forecasts and does not deteriorate the upper-air forecasts. The information persists longer in the system with GMOS, although the local improvements do not propagate beyond 6-h lead time. Initial model tendencies indicate that the mass field is not significantly altered when using static error covariances and the boundary layer parameterizations damp the poorly balanced increment locally. Conversely, most of the analysis increment is propagated when using flow-dependent error statistics. It results in better balanced wind and mass fields and provides a more persistent impact on the forecasts. Forecast accuracy results from observing system experiments (assimilating SYNOP winds with all observations used operationally) are generally neutral. Nevertheless, forecasts and analyses from GMOS are more self-consistent than those from both Bilin and a control experiment (not assimilating near-surface winds over land) and the information from the observations persists up to 12-h lead time.

Probabilistic Precipitation-Type Forecasting Based on GEFS Ensemble Forecasts of Vertical Temperature Profiles

Abstract A Bayesian classification method for probabilistic forecasts of precipitation type is presented. The method considers the vertical wet-bulb temperature profiles associated with each precipitation type, transforms them into their principal components, and models each of these principal components by a skew normal distribution. A variance inflation technique is used to de-emphasize the impact of principal components corresponding to smaller eigenvalues, and Bayes’s theorem finally yields probability forecasts for each precipitation type based on predicted wet-bulb temperature profiles. This approach is demonstrated with reforecast data from the Global Ensemble Forecast System (GEFS) and observations at 551 METAR sites, using either the full ensemble or the control run only. In both cases, reliable probability forecasts for precipitation type being either rain, snow, ice pellets, freezing rain, or freezing drizzle are obtained. Compared to the model output statistics (MOS) approach presently used by the National Weather Service, the skill of the proposed method is comparable for rain and snow and significantly better for the freezing precipitation types.

Bias correction and downscaling of future RCM precipitation projections using a MOS‐Analog technique

AbstractIn this study we assess the suitability of a recently introduced analog‐based Model Output Statistics (MOS) downscaling method (referred to as MOS‐Analog) for climate change studies and compare the results with a quantile mapping bias correction method. To this aim, we focus on Spain and consider daily precipitation output from an ensemble of Regional Climate Models provided by the ENSEMBLES project. The reanalysis‐driven Regional Climate Model (RCM) data provide the historical data (with day‐to‐day correspondence with observations induced by the forcing boundary conditions) to conduct the analog search of the control (20C3M) and future (A1B) global climate model (GCM)‐driven RCM values. First, we show that the MOS‐Analog method outperforms the raw RCM output in the control 20C3M scenario (period 1971–2000) for all considered regions and precipitation indices, although for the worst‐performing models the method is less effective. Second, we show that the MOS‐Analog method broadly preserves the original RCM climate change signal for different future periods (2011–2040, 2041–2070, 2071–2100), except for those indices related to extreme precipitation. This could be explained by the limitation of the analog method to extrapolate unobserved precipitation records. These results suggest that the MOS‐Analog is a spatially consistent alternative to standard bias correction methods, although the limitation for extreme values should be taken with caution in cases where this aspect is relevant for the problem.

Ensemble Postprocessing of Daily Precipitation Sums over Complex Terrain Using Censored High-Resolution Standardized Anomalies

Probabilistic forecasts provided by numerical ensemble prediction systems have systematic errors and are typically underdispersive. This is especially true over complex topography with extensive terrain-induced small-scale effects, which cannot be resolved by the ensemble system. To alleviate these errors, statistical postprocessing methods are often applied to calibrate the forecasts. This article presents a new full-distributional spatial postprocessing method for daily precipitation sums based on the standardized anomaly model output statistics (SAMOS) approach. Observations and forecasts are transformed into standardized anomalies by subtracting the long-term climatological mean and dividing by the climatological standard deviation. This removes all site-specific characteristics from the data and makes it possible to fit one single regression model for all stations at once. As the model does not depend on the station locations, it directly allows the creation of probabilistic forecasts for any arbitrary location. SAMOS uses a left-censored power-transformed logistic response distribution to account for the large fraction of zero observations (dry days), the limitation to nonnegative values, and the positive skewness of the data. ECMWF reforecasts are used for model training and to correct the ECMWF ensemble forecasts with the big advantage that SAMOS does not require an extensive archive of past ensemble forecasts as only the most recent four reforecasts are needed, and it automatically adapts to changes in the ECMWF ensemble model. The application of the new method to the central Alps shows that the new method is able to depict the small-scale properties and returns accurate fully probabilistic spatial forecasts.

Regional climate model and model output statistics method uncertainties and the effect of temperature and precipitation on future river discharges in Scandinavia

We analyze the importance of regional climate models (GCM-RCMs) and model output statistics (MOS) methods as uncertainty sources for future changes of various hydrological variables in Scandinavia. The Hydrological Predictions for the Environment (HYPE) model, driven with daily mean temperature and precipitation, is used to simulate changes in river discharges and other hydrological components from the present-day climate (1980–2009) to mid-21st century conditions (2041–2070). The results show that GCM-RCM differences explain most of the spread in the simulated changes in the annual mean cycle of river discharge. At seasonal level, MOS-method uncertainties are most important during the winter and spring, which is likely explained by the sensitivity of snow processes to the representation of daily variability in the MOS methods. To gain physical insights into the physical processes, the relative importance of changes to temperature or precipitation on changes in surface hydrology are also assessed. In most regions of Scandinavia, changes to temperature explain most of the changes in river discharge volumes and spring peaks. Precipitation changes only have a secondary role in modulating these changes. Again, these results are mostly explained by changes in snow processes in winter and increases in evapotranspiration in summer.

Seasonal precipitation prediction via data‐adaptive principal component regression

ABSTRACTThis article studies a problem of predicting seasonal precipitation over East Asia from real observations and multi‐model ensembles. Classical model output statistics approach based on principal component analysis (PCA) has been widely used for climate prediction. However, it may not be efficient in predicting precipitation since PCA assumes that information of data should be retained by the second moment of them, which is too stringent to climate data that can be skewed or asymmetric. This article presents a method based on data‐adaptive PCA (DPCA) by Lim and Oh (2016) that can adapt to non‐Gaussian distributed data. In addition to investigate the utility of DPCA for climate study, we propose a data‐adaptive principal component regression for seasonal precipitation prediction, which consists of DPCA and a regularized regression technique that is able to handle high‐dimensional data. We apply the proposed method to nine general circulation models for prediction of precipitations on the summer season (June, July, and August). The prediction ability of the proposed method is evaluated in comparison with observations and model outputs (prediction) of each constituent model.

Ensemble calibration with preserved correlations: unifying and comparing ensemble copula coupling and member‐by‐member postprocessing

Two seemingly unrelated postprocessing concepts retaining inter‐variable, spatial and temporal rank dependence structures from raw ensemble forecasts are compared and unified. The first concept combines ensemble model output statistics (EMOS), which fits and adjusts a parametric probability density function, with the ensemble copula coupling (ECC) reordering notion. The second concept is the member‐by‐member postprocessing (MBMP), based on a direct parametric adjustment of the raw ensemble members. It is shown that MBMP can be interpreted as a specific EMOS‐ECC variant. In an application to ensemble forecasts for temperature, EMOS‐ECC and MBMP succeed in outperforming the raw ensemble and conserving correlation structures.

The early rainy season in Central America: the role of the tropical North Atlantic SSTs

ABSTRACTWe explored the relationship between the precipitation anomalies during May to June as the first peak of the rainy season in the Pacific slope of Central America, and sea surface temperature (SST) fluctuations in the surrounding oceans, using canonical correlation analysis (CCA). With this approach, we studied variations in total precipitation, frequency of rainy days and the monthly occurrence of days with rainfall above (below) the 80th (20th) percentile, due to changes in the nearby SST. Composites of the sea‐level pressure (SLP), geopotential heights (200 hPa), relative humidity (700 hPa), horizontal moisture flux and wind at 850 hPa were estimated to provide a dynamical analysis. The composites are calculated using the information obtained with CCA. In addition, we used a general circulation model forced with fixed SST to explore the sensitivity of the model to the SST patterns found using CCA. The results show that the SST over the tropical North Atlantic controls the precipitation fluctuations at interannual scales, due to its connection with the tropical upper tropospheric trough. Warmer (colder) temperatures result in SLP below normal in the Caribbean region, associated with an increase in the heights at 200 hPa. This vertical configuration reduces the wind shear between 850 and 200 hPa and increases the input of humidity to mid‐levels, creating favourable conditions for deep convection, and favouring the generation of tropical cyclone activity. In the Pacific, a positive anomalous low‐level moisture flux is observed from the ocean to the continental parts of the region and may enhance the formation of mesoscale systems. The classic prediction schemes show a lead time of 1 or 2 months; this is an advantage for climate services operative work. The atmospheric model outcomes replicate the main results found in the composite analysis, reflecting its potential use for model output statistics predictive schemes.

Improved Gridded Wind Speed Forecasts by Statistical Postprocessing of Numerical Models with Block Regression

Abstract Numerical weather forecast errors are routinely corrected through statistical postprocessing by several national weather services. These statistical postprocessing methods build a regression function called model output statistics (MOS) between observations and forecasts that is based on an archive of past forecasts and associated observations. Because of limited spatial coverage of most near-surface parameter measurements, MOS have been historically produced only at meteorological station locations. Nevertheless, forecasters and forecast users increasingly ask for improved gridded forecasts. The present work aims at building improved hourly wind speed forecasts over the grid of a numerical weather prediction model. First, a new observational analysis, which performs better in terms of statistical scores than those operationally used at Météo-France, is described as gridded pseudo-observations. This analysis, which is obtained by using an interpolation strategy that was selected among other alternative strategies after an intercomparison study conducted internally at Météo-France, is very parsimonious since it requires only two additive components, and it requires little computational resources. Then, several scalar regression methods are built and compared, using the new analysis as the observation. The most efficient MOS is based on random forests trained on blocks of nearby grid points. This method greatly improves forecasts compared with raw output of numerical weather prediction models. Furthermore, building each random forest on blocks and limiting those forests to shallow trees does not impair performance compared with unpruned and pointwise random forests. This alleviates the storage burden of the objects and speeds up operations.

Application of model output statistics to the GEM-AQ high resolution air quality forecast

The aim of the presented work was to analyse the impact of data stratification on the efficiency of the Model Output Statistics (MOS) methodology as applied to a high-resolution deterministic air quality forecast carried out with the GEM-AQ model. The following parameters forecasted by the GEM-AQ model were selected as predictors for the MOS equation: pollutant concentration, air temperature in the lowest model layer, wind speed in the lowest model layer, temperature inversion and the precipitation rate. A representative 2-year series were used to construct regression functions. Data series were divided into two subsets. Approximately 75% of the data (first 3weeks of each month) were used to estimate the regression function parameters. Remaining 25% (last week of each month) were used to test the method (control period). The subsequent 12months were used for method verification (verification period). A linear model fitted the function based on forecasted parameters to the observations. We have assumed four different temperature-based data stratification methods (for each method, separate equations were constructed). For PM10 and PM2.5, SO2 and NO2 the best correction results were obtained with the application of temperature thresholds in the cold season and seasonal distribution combined with temperature thresholds in the warm season. For the PM10, PM2.5 and SO2 the best results were obtained using a combination of two stratification methods separately for cold and warm seasons. For CO, the systematic bias of the forecasted concentrations was partly corrected. For ozone more sophisticated methods of data stratification did not bring a significant improvement.

Clustering technique for interpretation of cloudburst over Uttarakhand

Data Mining has been used extensively in various business and scientific applications for last few years. Data mining has been found to be providing a deep insight into understanding the hidden facts in huge databases. Data mining is an interdisciplinary subfield of computer science that discovers patterns in large data sets by using methods at the intersection of artificial intelligence, machine learning, statistics, and database systems. In this paper, data mining technique for Interpretation of Weather Forecasts for one of the most disastrous weather phenomenon viz. cloudburst has been applied. Every year, cloudburst over hilly areas and coastal regions causes loss of lives and property. The forecasting and warning of these events is very difficult. There is no satisfactory technique for anticipating the occurrence of cloudbursts because of their small scale. A very fine network of radars is required to be able to detect the likelihood of a cloudburst and this would be prohibitively expensive. The warning of cloudburst could only be provided at a small lead time say a few hours in advance based on the interpretation of latest satellite imagery data, powerful radar (Doppler category), if available, or by using Model Output Statistics (MOS) models. Another dimension to forecasting this weather event has been identified by applying clustering technique on primary data forecasted by global and regional models of weather forecasting. A recent case of Cloudburst over Uttarakhand that caused a huge loss has been analyzed using k-means clustering technique of data mining. It has been observed that with the mining of Numerical Weather Prediction model forecast data, the signals of formation of cloudburst can be found3-4 days in advance.

Combining parametric low‐dimensional ensemble postprocessing with reordering methods

State‐of‐the‐art weather forecasts usually rely on ensemble prediction systems, accounting for the different sources of uncertainty. As ensembles are typically uncalibrated, they should get statistically postprocessed. Several multivariate ensemble postprocessing techniques that additionally consider spatial, inter‐variable and/or temporal dependences have been developed. These can be roughly divided into two groups. The first group comprises parametric, mainly low‐dimensional, approaches that are tailored to specific settings. The second group involves non‐parametric reordering methods that impose a specific dependence template on univariately postprocessed forecasts and are suitable in any dimension. In this article, these different strategies are combined, with the aim of exploiting the benefits of both concepts. Specifically, a high‐dimensional postprocessing problem is divided into multiple low‐dimensional instances, each of which is postprocessed via a suitable multivariate parametric method. From each postprocessed low‐dimensional distribution, a sample is drawn, which is then reordered according to the corresponding multidimensional rank structure of an appropriately chosen dependence template. In this context, different ranking concepts for multivariate settings are discussed. Finally, all reordered samples are aggregated to obtain the overall postprocessed ensemble. The new approach is applied to ensemble forecasts for temperature and wind speed at several locations from the European Centre for Medium‐Range Weather Forecasts, using a recent bivariate ensemble model output statistics postprocessing technique and a reordering based on the raw ensemble forecasts similar to the ensemble copula coupling method. It shows good predictive skill and outperforms reference ensembles.

Bivariate ensemble model output statistics approach for joint forecasting of wind speed and temperature

Forecast ensembles are typically employed to account for prediction uncertainties in numerical weather prediction models. However, ensembles often exhibit biases and dispersion errors, thus they require statistical post-processing to improve their predictive performance. Two popular univariate post-processing models are the Bayesian model averaging (BMA) and the ensemble model output statistics (EMOS). In the last few years increased interest has emerged in developing multivariate post-processing models, incorporating dependencies between weather quantities, such as for example a bivariate distribution for wind vectors or even a more general setting allowing to combine any types of weather variables. In line with a recently proposed approach to model temperature and wind speed jointly by a bivariate BMA model, this paper introduces a bivariate EMOS model for these weather quantities based on a truncated normal distribution. The bivariate EMOS model is applied to temperature and wind speed forecasts of the eight-member University of Washington mesoscale ensemble and of the eleven-member ALADIN-HUNEPS ensemble of the Hungarian Meteorological Service and its predictive performance is compared to the performance of the bivariate BMA model and a multivariate Gaussian copula approach, post-processing the margins with univariate EMOS. While the predictive skills of the compared methods are similar, the bivariate EMOS model requires considerably lower computation times than the bivariate BMA method.

Wind Pattern Recognition and Reference Wind Mast Data Correlations With NWP for Improved Wind-Electric Power Forecasts

A new statistical approach has been proposed for improved short-term wind-electric power forecasts of wind power plants (WPPs) based on a new wind-pattern-recognition technique, and reference wind mast (RWM) data correlations with numerical weather predictions (NWPs) to localize wind data to the given WPP site. For this purpose, first, NWP data are combined using adaptive boosting (AdaBoost) machine-learning algorithm to provide a proper combination of meteorological grid data from a set of grids around each WPP. Then, combined grid data are clustered, and for each cluster, an artificial neural network (ANN)/support vector machine (SVM) model is constructed to learn the relationship between the wind patterns of NWP data and RWM measurements. The outputs of this wind-to-wind modeling stage are used to obtain raw short-term 48-h-ahead wind-electric power forecasts via the consecutive statistical ANN/SVM models applied in the wind-to-power stage. The systematic errors are eliminated by applying model output statistics (MOS) and a weighted average combination method is used to obtain the final 48-h-ahead wind-electric power forecasts. The proposed model has been successfully applied to seven WPPs with installed capacities in the range from 10 to ${\sim}200\;\text{MW}$. The wind-electric power-forecast results obtained by using the proposed approach are compared with the reference benchmark models and other statistical models, which do not use any wind pattern recognition or NWP data correction via RWM measurements. It has been shown that the proposed wind-pattern-recognition technique and the wind-to-wind model developed by the use of an RWM for each WPP bring an improvement in the range from 2.3% to 5.1% on the normalized mean absolute error of wind-electric power forecasts, for an average training period of 2 years and a test period of 6 months for the given WPPs, as compared with conventional statistical methods.

Calibrated Ensemble Forecasts Using Quantile Regression Forests and Ensemble Model Output Statistics

Abstract Ensembles used for probabilistic weather forecasting tend to be biased and underdispersive. This paper proposes a statistical method for postprocessing ensembles based on quantile regression forests (QRF), a generalization of random forests for quantile regression. This method does not fit a parametric probability density function (PDF) like in ensemble model output statistics (EMOS) but provides an estimation of desired quantiles. This is a nonparametric approach that eliminates any assumption on the variable subject to calibration. This method can estimate quantiles using not only members of the ensemble but any predictor available including statistics on other variables. The method is applied to the Météo-France 35-member ensemble forecast (PEARP) for surface temperature and wind speed for available lead times from 3 up to 54 h and compared to EMOS. All postprocessed ensembles are much better calibrated than the PEARP raw ensemble and experiments on real data also show that QRF performs better than EMOS, and can bring a real gain for human forecasters compared to EMOS. QRF provides sharp and reliable probabilistic forecasts. At last, classical scoring rules to verify predictive forecasts are completed by the introduction of entropy as a general measure of reliability.

Regionalization of post-processed ensemble runoff forecasts

Abstract. For many years, meteorological models have been run with perturbated initial conditions or parameters to produce ensemble forecasts that are used as a proxy of the uncertainty of the forecasts. However, the ensembles are usually both biased (the mean is systematically too high or too low, compared with the observed weather), and has dispersion errors (the ensemble variance indicates a too low or too high confidence in the forecast, compared with the observed weather). The ensembles are therefore commonly post-processed to correct for these shortcomings. Here we look at one of these techniques, referred to as Ensemble Model Output Statistics (EMOS) (Gneiting et al., 2005). Originally, the post-processing parameters were identified as a fixed set of parameters for a region. The application of our work is the European Flood Awareness System (http://www.efas.eu), where a distributed model is run with meteorological ensembles as input. We are therefore dealing with a considerably larger data set than previous analyses. We also want to regionalize the parameters themselves for other locations than the calibration gauges. The post-processing parameters are therefore estimated for each calibration station, but with a spatial penalty for deviations from neighbouring stations, depending on the expected semivariance between the calibration catchment and these stations. The estimated post-processed parameters can then be used for regionalization of the postprocessing parameters also for uncalibrated locations using top-kriging in the rtop-package (Skøien et al., 2006, 2014). We will show results from cross-validation of the methodology and although our interest is mainly in identifying exceedance probabilities for certain return levels, we will also show how the rtop package can be used for creating a set of post-processed ensembles through simulations.