Ensemble Forecast System Research Articles

Using Stochastically Perturbed Parameterizations to Represent Model Uncertainty. Part I: Implementation and Parameter Sensitivity

Abstract Accurately representing model-based sources of uncertainty is essential for the development of reliable ensemble prediction systems for NWP applications. Uncertainties in discretizations, algorithmic approximations, and diabatic and unresolved processes combine to influence forecast skill in a flow-dependent way. An emerging approach designed to provide a process-level representation of these potential error sources, stochastically perturbed parameterizations (SPP), is introduced into the Canadian operational Global Ensemble Prediction System. This implementation extends the SPP technique beyond its typical application to free parameters in the physics suite by sampling uncertainty both within the dynamical core and at the formulation level using “error models” when multiple physical closures are available. Because SPP perturbs components within the model, internal consistency is ensured and conservation properties are not affected. The full SPP scheme is shown to increase ensemble spread to keep pace with error growth on a global scale. The sensitivity of the ensemble to each independently perturbed “element” is then assessed, with those responsible for the bulk of the response analyzed in more detail. Perturbations to surface exchange coefficients and the turbulent mixing length have a leading impact on near-surface statistics. Aloft, a tropically focused error model representing uncertainty in the advection scheme is found to initiate growing perturbations on the subtropical jet that lead to forecast improvements at higher latitudes. The results of Part I suggest that SPP has the potential to serve as a reliable representation of model uncertainty for ensemble NWP applications. Significance Statement Ensemble systems account for the negative impact that uncertainties in prediction models have on forecasts. Here, uncertain model parameters and algorithms are subjected to perturbations representing impact on forecast errors. By initiating error growth within the model calculations, the equally skillful members of the ensemble remain physically realistic and self-consistent, which is not guaranteed by other depictions of model error. This “stochastically perturbed parameterization” technique (SPP) comprises many small error sources, each analyzed in isolation. Each source is related to a limited set of processes, making it possible to determine how the individual perturbations affect the forecast. We conclude that SPP in the Canadian Global Ensemble Forecasting System produces realistic estimates of the impact of model uncertainties on forecast skill.

An Ensemble Forecasting Method for Dealing with the Combined Effects of the Initial and Model Errors and a Potential Deep Learning Implementation

Abstract In this paper, a new nonlinear forcing singular vector (NFSV) approach is proposed to provide mutually independent optimally combined modes of initial perturbations and model perturbations (C-NFSVs) in ensemble forecasts. The C-NFSVs are a group of optimally growing structures that take into account the impact of the interaction between the initial errors and the model errors effectively, generalizing the original NFSV for simulations of the impact of the model errors. The C-NFSVs method is tested in the context of the Lorenz-96 model to demonstrate its potential to improve ensemble forecast skills. This method is compared with the orthogonal conditional nonlinear optimal perturbations (O-CNOPs) method for estimating only the initial uncertainties and the orthogonal NFSVs (O-NFSVs) for estimating only the model uncertainties. The results demonstrate that when both the initial perturbations and model perturbations are introduced in the forecasting system, the C-NFSVs are much more capable of achieving higher ensemble forecasting skills. The use of a deep learning approach as a remedy for the expensive computational costs of the C-NFSVs is evaluated. The results show that learning the impact of the C-NFSVs on the ensemble provides a useful and efficient alternative for the operational implementation of C-NFSVs in forecasting suites dealing with the combined effects of the initial errors and the model errors. Significance Statement A new ensemble forecasting method for dealing with combined effects of initial errors and model errors, i.e., the C-NFSVs, is proposed, which is an extension of the NFSV approach for simulating the model error effects in ensemble forecasts. The C-NFSVs provide mutually independent optimally combined modes of initial perturbations and model perturbations. This new method is tested for generating ensemble forecasts in the context of the Lorenz-96 model, and there are indications that the optimally growing structures may provide reliable ensemble forecasts. Furthermore, it is found that a hybrid dynamical–deep learning approach could be a potential avenue for real-time ensemble forecasting systems when perturbations combine the impact of the initial and the model errors.

Forecast Dropouts in the NAVGEM Model: Characterization with Respect to Other Models, Large-Scale Indices, and Ensemble Forecasts

Abstract A forecast “bust” or “dropout” can be defined as an intermittent but significant loss of model forecast performance. Deterministic forecast dropouts are typically defined in terms of the 500-hPa geopotential height (Φ500) anomaly correlation coefficient (ACC) in the Northern Hemisphere (NH) dropping below a predefined threshold. This study first presents a multimodel comparison of dropouts in the Navy Global Environmental Model (NAVGEM) deterministic forecast with the ensemble control members from the Environment and Climate Change Canada (ECCC) Global Ensemble Prediction System (GEPS) and the National Centers for Environmental Prediction (NCEP) Global Ensemble Forecast System (GEFS). Then, the relationship between dropouts and large-scale pattern variability is investigated, focusing on the temporal variability and correlation of flow indices surrounding dropout events. Finally, three severe dropout events are examined from an ensemble perspective. The main findings of this work are the following: 1) forecast dropouts exhibit some relation between models; 2) although forecast dropouts do not have a single cause, the most severe dropouts in NAVGEM can be linked to specific behavior of the large-scale flow indices, that is, they tend to follow periods of rapidly escalating volatility of the flow indices, and they tend to occur during intervals where the AO and Pacific North American (PNA) indices are exhibiting unusually strong interdependence; and 3) for the dropout events examined from an ensemble perspective, the NAVGEM ensemble spread does not provide a strong signal of elevated potential for very large forecast errors.

Evaluating GFS and ECMWF Ensemble Forecasts of Integrated Water Vapor Transport along the U.S. West Coast

Abstract Atmospheric rivers (ARs) are long and narrow regions in the atmosphere of enhanced integrated water vapor transport (IVT) and can produce extreme precipitation and high societal impacts. Reliable and skillful forecasts of landfalling ARs in the western United States are critical to hazard preparation and aid in decision support activities, such as Forecast-Informed Reservoir Operations (FIRO). The purpose of this study is to compare the cool-season water year skill of the NCEP Global Ensemble Forecast System (GEFS) and ECMWF Ensemble Prediction System (EPS) forecasts of IVT along the U.S. West Coast for 2017–20. The skill is analyzed using probability-over-threshold forecasts of IVT magnitudes ≥ 250 kg m−1 s−1 (P250) using contingency table skill metrics in coastal Northern California and along the west coast of North America. Analysis of P250 with lead time (dProg/dt) found that the EPS provided ∼1 day of additional lead time for situational awareness over the GEFS at lead times of 6–10 days. Forecast skill analysis highlights that the EPS leads over the GEFS with success ratios 0.10–0.15 higher at lead times > 6 days for P250 thresholds of ≥25% and ≥50%, while event-based skill analysis using the probability of detection (POD) found that both models were largely similar with minor latitudinal variations favoring higher POD for each model in different locations along the coast. The relative skill of the EPS over the GEFS is largely attributed to overforecasting by the GEFS at longer lead times and an increase in the false alarm ratio. Significance Statement The purpose of this study is to evaluate the efficacy of the NCEP Global Ensemble Forecast System (GEFS) and the ECMWF Ensemble Prediction System (EPS) in forecasting enhanced water vapor transport along the U.S. West Coast commonly associated with landfalling atmospheric rivers and heavy precipitation. The ensemble models allow us to calculate the probability that enhanced water vapor transport will occur, thereby providing situational awareness for decision-making, such as in hazard mitigation and water resource management. The results of this study indicate that the EPS model is on average more skillful than the GEFS model at lead times of ∼6–10 days with a higher success ratio and lower false alarm ratio.

Developing and evaluating week 2 and weeks 3-4 outlook tools for extratropical storminess

Extratropical cyclones give rise to most of the high impact weather in the mid-to high-latitudes during the cool seasons, including heavy precipitation and strong winds. Thus it is important for stakeholders to be informed of approaching periods of increased or decreased cyclone activity. While individual cyclone tracks can be predicted out to about a week or so, from week 2 on, statistics summarizing cyclone activity, or storminess, are more useful. Storminess can be defined based on Lagrangian cyclone tracking or by Eulerian variance statistics. The outlook includes a combination of both methods. Lagrangian cyclone tracks provide information about where cyclones pass through and are more intuitive to users, while Eulerian variance statistics have been shown to be highly correlated with cyclone-related weather and are expected to be more predictable given that they are not as noisy. In this paper, we evaluate a storminess outlook tool developed based on dynamical model forecasts in the week-2 and weeks 3-4 time ranges. The outlook uses two 6-hourly subseasonal ensemble forecasts–the Global Ensemble Forecast System version 12 (GEFSv12), and the coupled Climate Forecast System version 2 (CFSv2). Hindcasts and operational forecasts from 1999–2016 are used to assess the prediction skill. Our results show that the GEFSv12 and CFSv2 combined ensemble has higher skill than either individual ensemble. The combined ensemble shows some skill in predicting cyclone amplitude and frequency out to weeks 3-4, with highest skill in winter, and lowest skill in summer. Models also show some skill in predicting the statistics of deep cyclones for week 2. The prediction skills for an Eulerian sea level pressure variance storminess metric is significantly higher than those for Lagrangian track statistics. Our results also show that GEFSv12 performs better than its predecessor GEFSv11. Correlations between the storminess indices and surface weather, including precipitation and high winds, are examined. A publicly accessible web page has been developed to display the subseasonal predictions in real time. The web page also contains information on climatology and forecast verification to enable users to make more informed use of the outlook.

An ensemble forecasting system for short-term power load based on multi-objective optimizer and fuzzy granulation

As an irreplaceable power source, electricity is responsible for sustaining the national economy and social development, and the precondition for the power system’s stable operation is to perform an accurate short-term load forecast (STLF). However, with the increasing forms of social power consumption and the emergence of large-scale sustainable resources on the grid, which make STLF increasingly challenging as the power load exhibits greater stochasticity and instability. Therefore, a novel STLF system is developed in this paper, which incorporates data fuzzy granulation, a high-performance optimizer for integrating forecasting sequences, point and interval forecasts. Moreover, the performance tests of the optimization algorithm verify that our proposed optimizer can obtain more approximate solution sets to the real Pareto front and outshines the traditional optimization algorithm concerning convergence and diversity. Load data from three regions of Australia demonstrate that our developed system can remarkably contribute to the accuracy and stability of the STLF, and also quantify the volatility and uncertainty of the power load, which allows power workers to better capture the fluctuation interval of future loads and effectively enhance the flexibility of grid operation.

How the Environmental Lifting Condensation Level Affects the Sensitivity of Simulated Convective Storm Cold Pools to the Microphysics Parameterization

Abstract Several studies have documented the sensitivity of convective storm simulations to the microphysics parameterization, but there is less research documenting how these sensitivities change with environmental conditions. In this study, the influence of the lifting condensation level (LCL) on the sensitivity of simulated ordinary convective storm cold pools to the microphysics parameterization is examined. To do this, seven perturbed-microphysics ensembles with nine members each are used, where each ensemble uses a different base state with a surface-based LCL between 500 and 2000 m. A comparison of ensemble standard deviations of cold-pool properties shows a clear trend of increasing sensitivity to the microphysics as the LCL is raised. In physical terms, this trend is the result of lower relative humidities in high-LCL environments that increase low-level rain evaporational cooling rates, which magnifies differences in evaporation already present among the members of a given ensemble owing to the microphysics variations. Omitting supersaturation from the calculation of rain evaporation so that only the raindrop size distribution influences evaporation leads to more evaporation in the low-LCL simulations (owing to more drops), as well as a slightly larger spread in evaporational cooling amounts between members in the low-LCL ensembles. Cold pools in the low-LCL environments are also found to develop earlier and are initially more sensitive to raindrop breakup owing to a larger warm-cloud depth. Altogether, these results suggest that convective storms may be more predictable in low-LCL environments, and forecasts of convection in high-LCL environments may benefit the most from microphysics perturbations within an ensemble forecasting system. Significance Statement Computer simulations of thunderstorms can have grid spacings ranging from tens to thousands of meters. Because individual precipitation particles form on scales smaller than these grid spacings, the bulk effects of precipitation processes in models must be approximated. Past studies have found that models are sensitive to these approximations. In this study, we test whether the sensitivity to these approximations changes with the relative humidity in the lowest 1–2 km of the atmosphere. We found that increasing the relative humidity decreases the sensitivity of simulations to the precipitation process approximations. These results can inform meteorologists about the uncertainties surrounding computer-generated thunderstorm forecasts and suggest environmental conditions where using several computer models with different precipitation process approximations may be beneficial.

Hybrid Neural Network Models for Postprocessing Medium-Range Forecasts of Tropical Cyclone Tracks over the Western North Pacific

Abstract Tropical cyclone (TC) track forecasts derived from dynamical models inherit their errors. In this study, a neural network (NN) algorithm was proposed for postprocessing TC tracks predicted by the Global Ensemble Forecast System (GEFS) for lead times of 2, 4, 5, and 6 days over the western North Pacific. The hybrid NN is a combination of three NN classes: 1) convolutional NN that extracts spatial features from GEFS fields; 2) multilayer perceptron, which processes TC positions predicted by GEFS; and 3) recurrent NN that handles information from previous time steps. A dataset of 204 TCs (6744 samples), which were formed from 1985 to 2019 (June–October) and survived for at least six days, was separated into various track patterns. TCs in each track pattern were distributed uniformly to validation and test dataset, in which each contained 10% TCs of the entire dataset, and the remaining 80% were allocated to the training dataset. Two NN architectures were developed, with and without a shortcut connection. Feature selection and hyperparameter tuning were performed to improve model performance. The results present that mean track error and dispersion could be reduced, particularly with the shortcut connection, which also corrected the systematic speed and direction bias of GEFS. Although a reduction in mean track error was not achieved by the NNs for every forecast lead time, improvement can be foreseen upon calibration for reducing overfitting, and the performance encourages further development in the present application.

Sensitivity of ECMWF coupled forecasts to improved initialization of the ocean mesoscale

AbstractIt has been hypothesized that initial conditions derived from eddy‐resolving ocean models could benefit eddy‐permitting forecast systems at subseasonal lead times (2 weeks to 2 months) by providing improved ocean initial conditions and reduced sea‐surface temperature (SST) errors associated with a more realistic representation of mesoscale features and ocean fronts such as the Gulf Stream. Here, we test this hypothesis by running a series of coupled 32‐day reforecasts with ocean initial conditions derived from native resolution (i.e., eddy‐permitting) and higher resolution (i.e., eddy‐resolving) ocean reanalyses. Initial conditions derived from the eddy‐resolving reanalysis reproduce observed temperature and sea‐level variability associated with mesoscale ocean eddies more accurately. Furthermore, forecasts of ocean variables are significantly improved in the extratropics, particularly in regions of observed intense mesoscale variability. This result is relevant for the marine forecasting community and demonstrates the potential for improved probabilistic forecasts of ocean variables in affordable ensemble forecast systems. However, the impacts on the mean state and forecast skill in the atmosphere are more limited. Based on an evaluation of model biases as a function of lead time, we conclude that there is room for further improvements to the initialization of the Gulf Stream that could lead to more significant impacts on atmospheric skill. Furthermore, we demonstrate that our reforecasts underestimate the strength of mesoscale air–sea interaction in the vicinity of strong SST fronts by 20–40%. We speculate that the impacts of improved ocean eddy initialization on atmospheric forecast skill may be larger in coupled models with higher ocean and atmosphere resolutions that can simulate air–sea interaction in the midlatitudes more realistically.

Operational Probabilistic Fog Prediction Based on Ensemble Forecast System: A Decision Support System for Fog

One of the well-known challenges of fog forecasting is the high spatio-temporal variability of fog. An ensemble forecast aims to capture this variability by representing the uncertainty in the initial/lateral boundary conditions (ICs/BCs) and model physics. The present study highlights a new operational Ensemble Forecast System (EFS) developed by the Indian Institute of Tropical Meteorology (IITM), Pune, to predict the fog over the Indo-Gangetic Plain (IGP) region using the visibility (Vis) diagnostic algorithm. The EFS framework comprises the WRF model with a 4 km horizontal resolution, initialized by 21 ICs/BCs. The advantages of probabilistic fog forecasting have been demonstrated by comparing control (CNTL) and ensemble-based fog forecasts. The forecast is verified using fog observations from the Indira Gandhi International (IGI) airport during the winter months of 2020–2021 and 2021–2022. The results show that with a probability threshold of 50%, the ensemble forecasts perform better than the CNTL forecasts. The skill scores of EFS are relatively promising, with a Hit Rate of 0.95 and a Critical Success Index of 0.55; additionally, the False Alarm Rate and Missing Rate are low, with values of 0.43 and 0.04, respectively. The EFS could correctly predict more fog events (37 out of 39) compared with the CNTL forecast (31 out of 39) and shows the potential skill. Furthermore, EFS has a substantially reduced error in predicting fog onset and dissipation (mean onset and dissipation error of 1 h each) compared to the CNTL forecasts.

Forecast calibrations of surface air temperature over Xinjiang based on U-net neural network

In this study, a deep learning method named U-net neural network is utilized to calibrate the gridded forecast of surface air temperature from the Global Ensemble Forecasting System (GEFS), with forecast lead times of 1–7 days in Xinjiang. The calibration performance of U-net is compared with three conventional postprocessing methods: unary linear regression (ULR), the decaying averaging method (DAM) and Quantile Mapping (QM). Results show that biases of the raw GEFS forecasts are mainly distributed in the Altai Mountains, the Junggar Basin, the Tarim Basin and the Kunlun Mountains. The four postprocessing methods effectively improve the forecast skills for all lead times, whereas U-net shows the best correction performance with the lowest mean absolute error (MAE) and the highest hit rate of 2°C (HR2) and pattern correlation coefficient (PCC). The U-net model considerably reduces the warm biases of the raw forecasts. The skill improvement magnitudes are greater in southern than northern Xinjiang, showing a higher mean absolute error skill score (MAESS). Furthermore, in order to distinguish the error sources of each forecasting scheme and to reveal their capabilities of calibrating errors of different sources, the error decomposition analysis is carried out based on the mean square errors. It shows that the bias term is the leading source of error in the raw forecasts, and barely changes as the lead time increases, which is mainly distributed in Tarim Basin and Kunlun Mountains. All four forecast calibrations effectively reduce the bias and distribution error of the raw forecasts, but only the U-net significantly reduces the sequence error.

Deep-learning-based post-processing for probabilistic precipitation forecasting

Ensemble prediction systems (EPSs) serve as a popular technique to provide probabilistic precipitation prediction in short- and medium-range forecasting. However, numerical models still suffer from imperfect configurations associated with data assimilation and physical parameterization, which can lead to systemic bias. Even state-of-the-art models often fail to provide high-quality precipitation forecasting, especially for extreme events. In this study, two deep-learning-based models—a shallow neural network (NN) and a deep NN with convolutional layers (CNN)—were used as alternative post-processing approaches to further improve the probabilistic forecasting of precipitation over China with 1–7 lead days. A popular conventional method—the censored and shifted gamma distribution-based ensemble model output statistics (CSG EMOS)—was used as the baseline. Re-forecasts run using a frozen EPS—Global Ensemble Forecast System version 12—were collected as the raw ensembles spanning from 2000 to 2019. The re-forecast data were generated once per day and consisted of one control run and four perturbed members. We used the calendar year 2018 as the validation period and 2019 as the testing period, and the remaining 18 years of data were used for training. According to the results, in terms of the continuous ranked probability score (CRPS) and the Brier score, the CNN model significantly outperforms the shallow NN model, as well as the CSG EMOS approach and the raw ensemble, especially for heavy or extreme precipitation events (those exceeding 50 mm/day). A remarkable degradation was seen when reducing the size of training samples from 18 years of data to two years. The spatial distribution of the CRPS shows that the stations in central China were better calibrated than those in other regions. With a lead time of 1 day, the CNN model was found to be superior to the other models (in terms of the CRPS) at 74.5% of the study stations. These results indicate that deep NNs can serve as a promising approach to the statistical post-processing of probabilistic precipitation forecasting.

Ensemble forecasting system based on decomposition-selection-optimization for point and interval carbon price prediction

In this study, a novel ensemble forecasting system based on data decomposition, feature selection, optimal sub-model determination and the improved multi-objective optimization is designed to conduct both point and interval carbon price predictions. To testify the performance of the developed system, three daily carbon price series are collected for experiments and verification. The results indicated that the ensemble system has achieved the optimal mean absolute percent errors of 0.9267, 1.6433 and 0.9303% for Sites 1–3 during point forecasting. In addition, the prediction interval coverage probabilities of three sites are 0.9800, 0.9600 and 0.9600 under the 95% confidence level for interval forecasting, which demonstrated that the developed system can provide scientific and sufficient reference for carbon trading markets.

Modelling point-of-consumption residual chlorine in humanitarian response: Can cost-sensitive learning improve probabilistic forecasts?

Ensuring sufficient free residual chlorine (FRC) up to the time and place water is consumed in refugee settlements is essential for preventing the spread of waterborne illnesses. Water system operators need accurate forecasts of FRC during the household storage period. However, factors that drive FRC decay after water leaves the piped distribution system vary substantially, introducing significant uncertainty when modelling point-of-consumption FRC. Artificial neural network (ANN) ensemble forecasting systems (EFS) can account for this uncertainty by generating probabilistic forecasts of point-of-consumption FRC. ANNs are typically trained using symmetrical error metrics like mean squared error (MSE), but this leads to forecast underdispersion forecasts (the spread of the forecast is smaller than the spread of the observations). This study proposes to solve forecast underdispersion by training an ANN-EFS using cost functions that combine alternative metrics (Nash-Sutcliffe efficiency, Kling Gupta Efficiency, Index of Agreement) with cost-sensitive learning (inverse FRC weighting, class-based FRC weighting, inverse frequency weighting). The ANN-EFS trained with each cost function was evaluated using water quality data from refugee settlements in Bangladesh and Tanzania by comparing the percent capture, confidence interval reliability diagrams, rank histograms, and the continuous ranked probability. Training the ANN-EFS using the cost functions developed in this study produced up to a 70% improvement in forecast reliability and dispersion compared to the baseline cost function (MSE), with the best performance typically obtained by training the model using Kling-Gupta Efficiency and inverse frequency weighting. Our findings demonstrate that training the ANN-EFS using alternative metrics and cost-sensitive learning can improve the quality of forecasts of point-of-consumption FRC and better account for uncertainty in post-distribution chlorine decay. These techniques can enable humanitarian responders to ensure sufficient FRC more reliably at the point-of-consumption, thereby preventing the spread of waterborne illnesses.

Leveraging Spatial Patterns in Precipitation Forecasts Using Deep Learning to Support Regional Water Management

AbstractShort‐term precipitation forecasts are critical to regional water management, particularly in the Western U.S. where atmospheric rivers can be predicted reliably days in advance. However, spatial error in these forecasts may reduce their utility when the costs of false positives and negatives differ greatly. Here we investigate whether deep learning methods can leverage spatial patterns in precipitation forecasts to (a) improve the skill of predicting the occurrence of precipitation events at lead times from 1 to 14 days, and (b) balance the tradeoff between the rate of false negatives and false positives by modifying the discrimination threshold of the classifiers. This approach is demonstrated for the Sacramento River Basin, California, using the Global Ensemble Forecast System (GEFS) v2 precipitation fields as input to convolutional neural network (CNN) and multi‐layer perceptron models. Results show that the deep learning models do not significantly improve the overall skill (F1 score) relative to the ensemble mean GEFS forecast with bias‐corrected threshold. However, additional analysis of the CNN models suggests they often correct missed predictions from GEFS by compensating for spatial error at longer lead times. Additionally, the deep learning models provide the ability to adjust the rate of false positives and negatives based on the ratio of costs. Finally, analysis of the network activations (saliency) indicates spatial patterns consistent with physical understanding of atmospheric river events in this region, lending additional confidence in the ability of the method to support water management applications.

Improving Probabilistic Quantitative Precipitation Forecasts Using Short Training Data through Artificial Neural Networks

Abstract Conventional statistical postprocessing techniques offer limited ability to improve the skills of probabilistic guidance for heavy precipitation. This paper introduces two artificial neural network (ANN)-based, geographically aware, and computationally efficient postprocessing schemes, namely, the ANN-multiclass (ANN-Mclass) and the ANN–censored, shifted gamma distribution (ANN-CSGD). Both schemes are implemented to postprocess Global Ensemble Forecast System (GEFS) forecasts to produce probabilistic quantitative precipitation forecasts (PQPFs) over the contiguous United States (CONUS) using a short (60 days), rolling training window. The performances of these schemes are assessed through a set of hindcast experiments, wherein postprocessed 24-h PQPFs from the two ANN schemes were compared against those produced using the benchmark quantile mapping algorithm for lead times ranging from 1 to 8 days. Outcomes of the hindcast experiments show that ANN schemes overall outperform the benchmark as well as the raw forecast over the CONUS in predicting probability of precipitation over a range of thresholds. The relative performance varies among geographic regions, with the two ANN schemes broadly improving upon quantile mapping over the central, south, and southeast, and slightly underperforming along the Pacific coast where skills of raw forecasts are the highest. Between the two schemes, the hybrid ANN-CSGD outperforms at higher rainfall thresholds (i.e., >50 mm day−1), though the outperformance comes at a slight expense of sharpness and spatial specificity. Collectively, these results confirm the ability of the ANN algorithms to produce skillful PQPFs with a limited training window and point to the prowess of the hybrid scheme for calibrating PQPFs for rare-to-extreme rainfall events.

Predicting the intensity of blocking events using machine and deep learning

Blocking events are high pressure systems that occur in the middle to upper latitudes, diverting the flow of the jet stream and preventing the regular progression of the weather. Blocking events can persist for days to weeks, potentially causing extreme weather events such as droughts or heatwaves. Due to the harmful effects associated with the weather conditions that blocking events can cause, there have been ongoing efforts to forecast them. One such effort includes the Global Ensemble Forecast System (GEFS), which has been reported to underestimate the intensity of stronger blocking events by approximately ten percent. In the current study, one deep learning and three machine learning models were developed to predict the intensity of newly formed blocking events at onset. It was hypothesized that the models would have comparable percent error to the GEFS while using less time and computational resources, that there would be a strong correlation between predicted and actual blocking intensity, and that the deep learning model would have lower error than the machine learning models. The results showed that the models did indeed have comparable error to the GEFS and that there was a statistically significant correlation between predicted and actual blocking intensity for all four models, thus supporting the first two parts of the hypothesis. The third part was not supported since there was no statistically significant difference in error between any of the models, however the deep learning model was noted for not overfitting unlike the machine learning models.

Predictability and skill of convection-permitting ensemble forecast systems in predicting the record-breaking “21·7” extreme rainfall event in Henan Province, China

Predictability and skill of convection-permitting ensemble forecast systems in predicting the record-breaking “21·7” extreme rainfall event in Henan Province, China

Comparison of different ensemble precipitation forecast system evaluation, integration and hydrological applications

Comparison of different ensemble precipitation forecast system evaluation, integration and hydrological applications

Ensemble wind speed forecasting system based on optimal model adaptive selection strategy: Case study in China

Under the vision of global carbon neutrality and peaking, the vigorous development of easily accessible renewable energy sources, such as wind energy, is extremely important. However, achieving accurate and reliable forecasts remains a challenging problem owing to the irregularity and volatility of wind speed data. Existing forecasting modules based on point prediction ignore the limitations of a single forecasting algorithm, fail to capture the wind speed uncertainty in data or achieve high forecasting accuracy. To overcome these shortcomings, an ensemble system of decomposition, adaptive selection, and forecast was proposed to simulate the actual wind speed data of a wind farm. Empirical analyses show that an excellent denoising effect is achieved by the data decomposition method. Moreover, the adaptability of forecasting modules to complex data is improved by the optimal submodel selection strategy. The reduction coefficients are effectively optimized by the multi-objective bald eagle search algorithm used for the first time. This system not only improves the point forecasting accuracy but also analyzes the uncertainty of wind speed, providing actual technical support for grid dispatching.