Search Results
You are looking at 1 - 10 of 36 items for
- Author or Editor: Luca Delle Monache x
- Refine by Access: All Content x
Abstract
An analog ensemble (AnEn) is constructed by first matching up the current forecast from a numerical weather prediction (NWP) model with similar past forecasts. The verifying observation from each match is then used as an ensemble member. For at least some applications, the advantages of AnEn over an NWP ensemble (multiple real-time model runs) may include higher efficiency, avoidance of initial condition and model perturbation challenges, and little or no need for postprocessing calibration. While AnEn can capture flow-dependent error growth, it may miss aspects of error growth that can be represented dynamically by the multiple real-time model runs of an NWP ensemble. To combine the strengths of the AnEn and NWP ensemble approaches, a hybrid ensemble (HyEn) is constructed by finding m analogs for each member of a small n-member NWP ensemble, to produce a total of m × n members.
Forecast skill is compared between the AnEn, HyEn, and an NWP ensemble calibrated using logistic regression. The HyEn outperforms the other approaches for probabilistic 2-m temperature forecasts yet underperforms for 10-m wind speed. The mixed results reveal a dependence on the intrinsic skill of the NWP members employed. In this study, the NWP ensemble is underspread for both 2-m temperature and 10-m winds, yet displays some ability to represent flow-dependent error for the former and not the latter. Thus, the HyEn is a promising approach for efficient generation of high-quality probabilistic forecasts, but requires use of a small, and at least partially functional, NWP ensemble.
Abstract
An analog ensemble (AnEn) is constructed by first matching up the current forecast from a numerical weather prediction (NWP) model with similar past forecasts. The verifying observation from each match is then used as an ensemble member. For at least some applications, the advantages of AnEn over an NWP ensemble (multiple real-time model runs) may include higher efficiency, avoidance of initial condition and model perturbation challenges, and little or no need for postprocessing calibration. While AnEn can capture flow-dependent error growth, it may miss aspects of error growth that can be represented dynamically by the multiple real-time model runs of an NWP ensemble. To combine the strengths of the AnEn and NWP ensemble approaches, a hybrid ensemble (HyEn) is constructed by finding m analogs for each member of a small n-member NWP ensemble, to produce a total of m × n members.
Forecast skill is compared between the AnEn, HyEn, and an NWP ensemble calibrated using logistic regression. The HyEn outperforms the other approaches for probabilistic 2-m temperature forecasts yet underperforms for 10-m wind speed. The mixed results reveal a dependence on the intrinsic skill of the NWP members employed. In this study, the NWP ensemble is underspread for both 2-m temperature and 10-m winds, yet displays some ability to represent flow-dependent error for the former and not the latter. Thus, the HyEn is a promising approach for efficient generation of high-quality probabilistic forecasts, but requires use of a small, and at least partially functional, NWP ensemble.
Abstract
An analog-based ensemble technique, the analog ensemble (AnEn), has been applied successfully to generate probabilistic predictions of meteorological variables, wind and solar power, energy demand, and the optimal bidding in the day-ahead energy market. The AnEn method uses a historical time series of past forecasts from a meteorological model or other prediction systems and observations of the quantity to be predicted. For each forecast lead time, the ensemble set of predictions is a set of observations from the past. These observations are those concurrent with the past forecasts at the same lead time, chosen across the past runs most similar to the current forecast. Recent applications have demonstrated that the AnEn introduces a conditional negative bias when predicting events in the right tail of the forecast distribution of wind speed, particularly when the training dataset is short. This underestimation increases when the predicted event occurs less frequently in the available historical data. A new bias correction for the AnEn using wind observations from more than 500 U.S. stations is tested to reduce the AnEn’s underestimation of rare events. It is shown that the conditional negative bias introduced by the AnEn in its standard application is significantly reduced by our novel approach. Also, the overall probabilistic AnEn performances improve when predicting wind speed higher than 10 m s−1 as demonstrated by lower values of the continuous ranked probability score. These improvements can be attributed to an increased reliability achieved by introducing the proposed bias correction algorithm.
Abstract
An analog-based ensemble technique, the analog ensemble (AnEn), has been applied successfully to generate probabilistic predictions of meteorological variables, wind and solar power, energy demand, and the optimal bidding in the day-ahead energy market. The AnEn method uses a historical time series of past forecasts from a meteorological model or other prediction systems and observations of the quantity to be predicted. For each forecast lead time, the ensemble set of predictions is a set of observations from the past. These observations are those concurrent with the past forecasts at the same lead time, chosen across the past runs most similar to the current forecast. Recent applications have demonstrated that the AnEn introduces a conditional negative bias when predicting events in the right tail of the forecast distribution of wind speed, particularly when the training dataset is short. This underestimation increases when the predicted event occurs less frequently in the available historical data. A new bias correction for the AnEn using wind observations from more than 500 U.S. stations is tested to reduce the AnEn’s underestimation of rare events. It is shown that the conditional negative bias introduced by the AnEn in its standard application is significantly reduced by our novel approach. Also, the overall probabilistic AnEn performances improve when predicting wind speed higher than 10 m s−1 as demonstrated by lower values of the continuous ranked probability score. These improvements can be attributed to an increased reliability achieved by introducing the proposed bias correction algorithm.
Abstract
An analog-based ensemble model output statistics (EMOS) is proposed to improve EMOS for the calibration of ensemble forecasts. Given a set of analog predictors and corresponding weights, which are optimized with a brute-force continuous ranked probability score (CRPS) minimization, forecasts similar to a current ensemble forecast (i.e., analogs) are searched. The best analogs and the corresponding observations form the training dataset for estimating the EMOS coefficients. To test the new approach for renewable energy applications, wind speed measurements at 100-m height from six measurement towers and wind ensemble forecasts at 100-m height from the European Centre for Medium-Range Weather Forecasts (ECMWF) Ensemble Prediction System (EPS) are used. The analog-based EMOS is compared against EMOS, an adaptive and recursive wind vector calibration (AUV), and an analog ensemble applied to ECMWF EPS. It is shown that the analog-based EMOS outperforms EMOS, AUV, and the analog ensemble at all measurement sites in terms of CRPS and Brier score for common and rare events. The CRPS improvements relative to EMOS reach up to 11% and are statistically significant at almost all sites. The reliability of the analog-based EMOS ensemble for rare events is better compared to EMOS and AUV and is similar compared to the analog ensemble.
Abstract
An analog-based ensemble model output statistics (EMOS) is proposed to improve EMOS for the calibration of ensemble forecasts. Given a set of analog predictors and corresponding weights, which are optimized with a brute-force continuous ranked probability score (CRPS) minimization, forecasts similar to a current ensemble forecast (i.e., analogs) are searched. The best analogs and the corresponding observations form the training dataset for estimating the EMOS coefficients. To test the new approach for renewable energy applications, wind speed measurements at 100-m height from six measurement towers and wind ensemble forecasts at 100-m height from the European Centre for Medium-Range Weather Forecasts (ECMWF) Ensemble Prediction System (EPS) are used. The analog-based EMOS is compared against EMOS, an adaptive and recursive wind vector calibration (AUV), and an analog ensemble applied to ECMWF EPS. It is shown that the analog-based EMOS outperforms EMOS, AUV, and the analog ensemble at all measurement sites in terms of CRPS and Brier score for common and rare events. The CRPS improvements relative to EMOS reach up to 11% and are statistically significant at almost all sites. The reliability of the analog-based EMOS ensemble for rare events is better compared to EMOS and AUV and is similar compared to the analog ensemble.
Abstract
This study presents a new implementation of the analog ensemble method (AnEn) to improve the prediction of wind speed for 146 storms that have impacted the northeast United States in the period 2005–16. The AnEn approach builds an ensemble by using a set of past observations that correspond to the best analogs of numerical weather prediction (NWP). Unlike previous studies, dual-predictor combinations are used to generate AnEn members, which include wind speed, wind direction, and 2-m temperature, simulated by two state-of-the-science atmospheric models [the Weather Research and Forecasting (WRF) Model and the Regional Atmospheric Modeling System–Integrated Community Limited Area Modeling System (RAMS–ICLAMS)]. Bias correction is also applied to each analog to gain additional benefits in predicting wind speed. Both AnEn and the bias-corrected analog ensemble (BCAnEn) are tested with a weighting strategy, which optimizes the predictor combination with root-mean-square error (RMSE) minimization. A leave-one-out cross validation is implemented, that is, each storm is predicted using the remaining 145 as the training dataset, with modeled and observed values over 80 stations in the northeast United States. The results show improvements of 9%–42% and 1%–29% with respect to original WRF and ICLAMS simulations, as measured by the RMSE of individual storms. Moreover, for two high-impact tropical storms (Irene and Sandy), BCAnEn significantly reduces the error of raw prediction (average RMSE reduction of 22% for Irene and 26% for Sandy). The AnEn and BCAnEn techniques demonstrate their potential to combine different NWP models to improve storm wind speed prediction, compared to the use of a single NWP.
Abstract
This study presents a new implementation of the analog ensemble method (AnEn) to improve the prediction of wind speed for 146 storms that have impacted the northeast United States in the period 2005–16. The AnEn approach builds an ensemble by using a set of past observations that correspond to the best analogs of numerical weather prediction (NWP). Unlike previous studies, dual-predictor combinations are used to generate AnEn members, which include wind speed, wind direction, and 2-m temperature, simulated by two state-of-the-science atmospheric models [the Weather Research and Forecasting (WRF) Model and the Regional Atmospheric Modeling System–Integrated Community Limited Area Modeling System (RAMS–ICLAMS)]. Bias correction is also applied to each analog to gain additional benefits in predicting wind speed. Both AnEn and the bias-corrected analog ensemble (BCAnEn) are tested with a weighting strategy, which optimizes the predictor combination with root-mean-square error (RMSE) minimization. A leave-one-out cross validation is implemented, that is, each storm is predicted using the remaining 145 as the training dataset, with modeled and observed values over 80 stations in the northeast United States. The results show improvements of 9%–42% and 1%–29% with respect to original WRF and ICLAMS simulations, as measured by the RMSE of individual storms. Moreover, for two high-impact tropical storms (Irene and Sandy), BCAnEn significantly reduces the error of raw prediction (average RMSE reduction of 22% for Irene and 26% for Sandy). The AnEn and BCAnEn techniques demonstrate their potential to combine different NWP models to improve storm wind speed prediction, compared to the use of a single NWP.
Abstract
This study explores the first application of an analog-based method to downscale precipitation estimates from a regional reanalysis. The utilized analog ensemble (AnEn) approach defines a metric with which a set of analogs (i.e., the ensemble) can be sampled from the observations in the training period. From the determined AnEn estimates, the uncertainty of the generated precipitation time series also can easily be assessed. The study investigates tuning parameters of AnEn, such as the choice of predictors or the ensemble size, to optimize the performance. The approach is implemented and tuned on the basis of a set of over 700 rain gauges with 6-hourly measurements for Germany and a 6.2-km regional reanalysis for Europe, which provides the predictors. The obtained AnEn estimates are evaluated against the observations over a 4-yr verification period. With respect to deterministic quality, the results show that AnEn is able to outperform the reanalysis itself depending on location and precipitation intensity. Further, AnEn produces superior results in probabilistic measures against a random-ensemble approach as well as a logistic regression. As a proof of concept, the described implementation allows for the estimation of synthetic probabilistic observation time series for periods for which measurements are not available.
Abstract
This study explores the first application of an analog-based method to downscale precipitation estimates from a regional reanalysis. The utilized analog ensemble (AnEn) approach defines a metric with which a set of analogs (i.e., the ensemble) can be sampled from the observations in the training period. From the determined AnEn estimates, the uncertainty of the generated precipitation time series also can easily be assessed. The study investigates tuning parameters of AnEn, such as the choice of predictors or the ensemble size, to optimize the performance. The approach is implemented and tuned on the basis of a set of over 700 rain gauges with 6-hourly measurements for Germany and a 6.2-km regional reanalysis for Europe, which provides the predictors. The obtained AnEn estimates are evaluated against the observations over a 4-yr verification period. With respect to deterministic quality, the results show that AnEn is able to outperform the reanalysis itself depending on location and precipitation intensity. Further, AnEn produces superior results in probabilistic measures against a random-ensemble approach as well as a logistic regression. As a proof of concept, the described implementation allows for the estimation of synthetic probabilistic observation time series for periods for which measurements are not available.
Abstract
The objective of this paper is to compare probabilistic 100-m wind speed forecasts, which are relevant for wind energy applications, from different regional and global ensemble prediction systems (EPSs) at six measurement towers in central Europe and to evaluate the benefits of combining single-model ensembles into multimodel ensembles. The global 51-member EPS from the European Centre for Medium-Range Weather Forecasts (ECMWF EPS) is compared against the Consortium for Small-Scale Modelling’s (COSMO) limited-area 16-member EPS (COSMO-LEPS) and a regional, high-resolution 20-member EPS centered over Germany (COSMO-DE EPS). The ensemble forecasts are calibrated with univariate (wind speed) ensemble model output statistics (EMOS) and bivariate (wind vector) recursive and adaptive calibration (AUV). The multimodel ensembles are constructed by pooling together raw or best-calibrated ensemble forecasts. An additional postprocessing of these multimodel ensembles with both EMOS and AUV is also tested. The best-performing calibration methodology for ECMWF EPS is AUV, while EMOS performs better than AUV for the calibration of COSMO-DE EPS. COSMO-LEPS has similar skill when calibrated with both EMOS and AUV. The AUV ECMWF EPS outperforms the EMOS COSMO-LEPS and COSMO-DE EPS for deterministic and probabilistic wind speed forecast skill. For most thresholds, ECMWF EPS has a comparable reliability and sharpness but higher discrimination ability. Multimodel ensembles, which are constructed by pooling together the best-calibrated EPSs, improve the skill relative to the AUV ECMWF EPS. An analysis of the error correlation among the EPSs indicates that multimodel ensemble skill can be considerably higher when the error correlation is low.
Abstract
The objective of this paper is to compare probabilistic 100-m wind speed forecasts, which are relevant for wind energy applications, from different regional and global ensemble prediction systems (EPSs) at six measurement towers in central Europe and to evaluate the benefits of combining single-model ensembles into multimodel ensembles. The global 51-member EPS from the European Centre for Medium-Range Weather Forecasts (ECMWF EPS) is compared against the Consortium for Small-Scale Modelling’s (COSMO) limited-area 16-member EPS (COSMO-LEPS) and a regional, high-resolution 20-member EPS centered over Germany (COSMO-DE EPS). The ensemble forecasts are calibrated with univariate (wind speed) ensemble model output statistics (EMOS) and bivariate (wind vector) recursive and adaptive calibration (AUV). The multimodel ensembles are constructed by pooling together raw or best-calibrated ensemble forecasts. An additional postprocessing of these multimodel ensembles with both EMOS and AUV is also tested. The best-performing calibration methodology for ECMWF EPS is AUV, while EMOS performs better than AUV for the calibration of COSMO-DE EPS. COSMO-LEPS has similar skill when calibrated with both EMOS and AUV. The AUV ECMWF EPS outperforms the EMOS COSMO-LEPS and COSMO-DE EPS for deterministic and probabilistic wind speed forecast skill. For most thresholds, ECMWF EPS has a comparable reliability and sharpness but higher discrimination ability. Multimodel ensembles, which are constructed by pooling together the best-calibrated EPSs, improve the skill relative to the AUV ECMWF EPS. An analysis of the error correlation among the EPSs indicates that multimodel ensemble skill can be considerably higher when the error correlation is low.
Abstract
The performance of analog-based and Kalman filter (KF) postprocessing methods is tested in climatologically and topographically different regions for point-based wind speed predictions at 10 m above the ground. The results are generated using several configurations of the mesoscale numerical weather prediction model ALADIN. This study shows that deterministic analog-based predictions (ABPs) improve the correlation between predictions and measurements while reducing the forecast error, with respect to both the starting model predictions and the KF-based correction. While the KF generally outperforms the ABPs in bias reduction, the combination of the KF and analog approach can be similarly successful. In the coastal complex area, characterized with a larger frequency of strong wind, the ABPs are more successful in reducing the dispersion error than the KF. The application of the KF algorithm to the analogs in the so-called analog space (KFAS) is the least prone to standard deviation underestimation among the ABPs. All ABPs improve the prediction of larger-than-diurnal motions, and KFAS is superior among all ABPs in predicting alternating wind regimes on time scales shorter than a day. The ABPs better distinguish different wind speed categories in the coastal complex terrain by using a higher-resolution model input. Differences among starting model and postprocessed forecasts in other types of terrain are less pronounced.
Abstract
The performance of analog-based and Kalman filter (KF) postprocessing methods is tested in climatologically and topographically different regions for point-based wind speed predictions at 10 m above the ground. The results are generated using several configurations of the mesoscale numerical weather prediction model ALADIN. This study shows that deterministic analog-based predictions (ABPs) improve the correlation between predictions and measurements while reducing the forecast error, with respect to both the starting model predictions and the KF-based correction. While the KF generally outperforms the ABPs in bias reduction, the combination of the KF and analog approach can be similarly successful. In the coastal complex area, characterized with a larger frequency of strong wind, the ABPs are more successful in reducing the dispersion error than the KF. The application of the KF algorithm to the analogs in the so-called analog space (KFAS) is the least prone to standard deviation underestimation among the ABPs. All ABPs improve the prediction of larger-than-diurnal motions, and KFAS is superior among all ABPs in predicting alternating wind regimes on time scales shorter than a day. The ABPs better distinguish different wind speed categories in the coastal complex terrain by using a higher-resolution model input. Differences among starting model and postprocessed forecasts in other types of terrain are less pronounced.
Abstract
The performance of the Hurricane Weather Research and Forecasting (HWRF) Model Rapid Intensification Analog Ensemble (RI-AnEn) is evaluated for real-time forecasts made during the National Oceanic and Atmospheric Administration (NOAA)’s 2018 Hurricane Forecast Improvement Program (HFIP) demonstration. Using a variety of assessment tools (Brier skill score, reliability diagrams, ROC curves, ROC skill scores), RI-AnEn is shown to perform competitively compared to both the deterministic HWRF and current operational probabilistic RI forecast aids. The assessment is extended to include forecasts from the 2017 HFIP demonstration and shows that RI-AnEn is the only model with significant RI forecast skill at all lead times in the Atlantic and eastern Pacific basins. Though RI-AnEn is overconfident in its RI forecasts, it is generally well calibrated for all lead times. Furthermore, significance testing indicates that for the 2017–18 Atlantic and eastern Pacific sample, RI-AnEn is more skillful than HWRF at all lead times and better than most of the other probabilistic guidance at 48 and 72 h. ROC curves reveal that RI-AnEn offers a good combination of sensitivity and specificity, performing comparably to SHIPS-RII at all lead times in both basins. With respect to specific high-impact cases from the 2018 Atlantic season, performance of RI-AnEn ranges from excellent (Hurricane Michael) to poor (Hurricane Florence). The multiyear assessment and results for two high-impact case studies from 2018 indicate that, while promising, RI-AnEn requires further work to refine its performance as well as to accurately situate its effectiveness relative to other RI forecasts aids.
Abstract
The performance of the Hurricane Weather Research and Forecasting (HWRF) Model Rapid Intensification Analog Ensemble (RI-AnEn) is evaluated for real-time forecasts made during the National Oceanic and Atmospheric Administration (NOAA)’s 2018 Hurricane Forecast Improvement Program (HFIP) demonstration. Using a variety of assessment tools (Brier skill score, reliability diagrams, ROC curves, ROC skill scores), RI-AnEn is shown to perform competitively compared to both the deterministic HWRF and current operational probabilistic RI forecast aids. The assessment is extended to include forecasts from the 2017 HFIP demonstration and shows that RI-AnEn is the only model with significant RI forecast skill at all lead times in the Atlantic and eastern Pacific basins. Though RI-AnEn is overconfident in its RI forecasts, it is generally well calibrated for all lead times. Furthermore, significance testing indicates that for the 2017–18 Atlantic and eastern Pacific sample, RI-AnEn is more skillful than HWRF at all lead times and better than most of the other probabilistic guidance at 48 and 72 h. ROC curves reveal that RI-AnEn offers a good combination of sensitivity and specificity, performing comparably to SHIPS-RII at all lead times in both basins. With respect to specific high-impact cases from the 2018 Atlantic season, performance of RI-AnEn ranges from excellent (Hurricane Michael) to poor (Hurricane Florence). The multiyear assessment and results for two high-impact case studies from 2018 indicate that, while promising, RI-AnEn requires further work to refine its performance as well as to accurately situate its effectiveness relative to other RI forecasts aids.
Abstract
This study explores an analog-based method to generate an ensemble [analog ensemble (AnEn)] in which the probability distribution of the future state of the atmosphere is estimated with a set of past observations that correspond to the best analogs of a deterministic numerical weather prediction (NWP). An analog for a given location and forecast lead time is defined as a past prediction, from the same model, that has similar values for selected features of the current model forecast. The AnEn is evaluated for 0–48-h probabilistic predictions of 10-m wind speed and 2-m temperature over the contiguous United States and against observations provided by 550 surface stations, over the 23 April–31 July 2011 period. The AnEn is generated from the Environment Canada (EC) deterministic Global Environmental Multiscale (GEM) model and a 12–15-month-long training period of forecasts and observations. The skill and value of AnEn predictions are compared with forecasts from a state-of-the-science NWP ensemble system, the 21-member Regional Ensemble Prediction System (REPS). The AnEn exhibits high statistical consistency and reliability and the ability to capture the flow-dependent behavior of errors, and it has equal or superior skill and value compared to forecasts generated via logistic regression (LR) applied to both the deterministic GEM (as in AnEn) and REPS [ensemble model output statistics (EMOS)]. The real-time computational cost of AnEn and LR is lower than EMOS.
Abstract
This study explores an analog-based method to generate an ensemble [analog ensemble (AnEn)] in which the probability distribution of the future state of the atmosphere is estimated with a set of past observations that correspond to the best analogs of a deterministic numerical weather prediction (NWP). An analog for a given location and forecast lead time is defined as a past prediction, from the same model, that has similar values for selected features of the current model forecast. The AnEn is evaluated for 0–48-h probabilistic predictions of 10-m wind speed and 2-m temperature over the contiguous United States and against observations provided by 550 surface stations, over the 23 April–31 July 2011 period. The AnEn is generated from the Environment Canada (EC) deterministic Global Environmental Multiscale (GEM) model and a 12–15-month-long training period of forecasts and observations. The skill and value of AnEn predictions are compared with forecasts from a state-of-the-science NWP ensemble system, the 21-member Regional Ensemble Prediction System (REPS). The AnEn exhibits high statistical consistency and reliability and the ability to capture the flow-dependent behavior of errors, and it has equal or superior skill and value compared to forecasts generated via logistic regression (LR) applied to both the deterministic GEM (as in AnEn) and REPS [ensemble model output statistics (EMOS)]. The real-time computational cost of AnEn and LR is lower than EMOS.
Abstract
Two new postprocessing methods are proposed to reduce numerical weather prediction’s systematic and random errors. The first method consists of running a postprocessing algorithm inspired by the Kalman filter (KF) through an ordered set of analog forecasts rather than a sequence of forecasts in time (ANKF). The analog of a forecast for a given location and time is defined as a past prediction that matches selected features of the current forecast. The second method is the weighted average of the observations that verified when the 10 best analogs were valid (AN). ANKF and AN are tested for 10-m wind speed predictions from the Weather Research and Forecasting (WRF) model, with observations from 400 surface stations over the western United States for a 6-month period. Both AN and ANKF predict drastic changes in forecast error (e.g., associated with rapid weather regime changes), a feature lacking in KF and a 7-day running-mean correction (7-Day). The AN almost eliminates the bias of the raw prediction (Raw), while ANKF drastically reduces it with values slightly worse than KF. Both analog-based methods are also able to reduce random errors, therefore improving the predictive skill of Raw. The AN is consistently the best, with average improvements of 10%, 20%, 25%, and 35% with respect to ANKF, KF, 7-Day, and Raw, as measured by centered root-mean-square error, and of 5%, 20%, 25%, and 40%, as measured by rank correlation. Moreover, being a prediction based solely on observations, AN results in an efficient downscaling procedure that eliminates representativeness discrepancies between observations and predictions.
Abstract
Two new postprocessing methods are proposed to reduce numerical weather prediction’s systematic and random errors. The first method consists of running a postprocessing algorithm inspired by the Kalman filter (KF) through an ordered set of analog forecasts rather than a sequence of forecasts in time (ANKF). The analog of a forecast for a given location and time is defined as a past prediction that matches selected features of the current forecast. The second method is the weighted average of the observations that verified when the 10 best analogs were valid (AN). ANKF and AN are tested for 10-m wind speed predictions from the Weather Research and Forecasting (WRF) model, with observations from 400 surface stations over the western United States for a 6-month period. Both AN and ANKF predict drastic changes in forecast error (e.g., associated with rapid weather regime changes), a feature lacking in KF and a 7-day running-mean correction (7-Day). The AN almost eliminates the bias of the raw prediction (Raw), while ANKF drastically reduces it with values slightly worse than KF. Both analog-based methods are also able to reduce random errors, therefore improving the predictive skill of Raw. The AN is consistently the best, with average improvements of 10%, 20%, 25%, and 35% with respect to ANKF, KF, 7-Day, and Raw, as measured by centered root-mean-square error, and of 5%, 20%, 25%, and 40%, as measured by rank correlation. Moreover, being a prediction based solely on observations, AN results in an efficient downscaling procedure that eliminates representativeness discrepancies between observations and predictions.
