Ensemble Prediction Research Articles

Ensemble hydrological predictions at an intraseasonal scale through a statistical–dynamical downscaling approach over southwestern Amazonia

ABSTRACT We developed and analyzed the performance of an ensemble forecasting system for the Madeira River basin, the largest sub-basin of the Amazon, with forecasts up to 30 days under different hydrometeorological conditions. We used outputs from the regional Eta model of precipitation and global climatological data as inputs to a large-scale hydrological model. Bias correction of precipitation through quantile mapping significantly improved the results, achieving a hit rate >70%. The system demonstrated the ability to discriminate between high, medium, and low flow conditions. Forecast performance is better for larger catchment areas. This system is expected to increase decision-making efficiency for flood and drought situations in the largest Amazon tributary.

Atomistic Prediction of Structures, Conformational Ensembles and Binding Energetics for the SARS-CoV-2 Spike JN.1, KP.2 and KP.3 Variants Using AlphaFold2 and Molecular Dynamics Simulations: Mutational Profiling and Binding Free Energy Analysis Reveal Epistatic Hotspots of the ACE2 Affinity and Immune Escape.

The most recent wave of SARS-CoV-2 Omicron variants descending from BA.2 and BA.2.86 exhibited improved viral growth and fitness due to convergent evolution of functional hotspots. These hotspots operate in tandem to optimize both receptor binding for effective infection and immune evasion efficiency, thereby maintaining overall viral fitness. The lack of molecular details on structure, dynamics and binding energetics of the latest FLiRT and FLuQE variants with the ACE2 receptor and antibodies provides a considerable challenge that is explored in this study. We combined AlphaFold2-based atomistic predictions of structures and conformational ensembles of the SARS-CoV-2 Spike complexes with the host receptor ACE2 for the most dominant Omicron variants JN.1, KP.1, KP.2 and KP.3 to examine the mechanisms underlying the role of convergent evolution hotspots in balancing ACE2 binding and antibody evasion. Using the ensemble-based mutational scanning of the spike protein residues and computations of binding affinities, we identified binding energy hotspots and characterized molecular basis underlying epistatic couplings between convergent mutational hotspots. The results suggested that the existence of epistatic interactions between convergent mutational sites at L455, F456, Q493 positions that enable to protect and restore ACE2 binding affinity while conferring beneficial immune escape. To examine immune escape mechanisms, we performed structure-based mutational profiling of the spike protein binding with several classes of antibodies that displayed impaired neutralization against BA.2.86, JN.1, KP.2 and KP.3. The results confirmed the experimental data that JN.1, KP.2 and KP.3 harboring the L455S and F456L mutations can significantly impair the neutralizing activity of class-1 monoclonal antibodies, while the epistatic effects mediated by F456L can facilitate the subsequent convergence of Q493E changes to rescue ACE2 binding. Structural and energetic analysis provided a rationale to the experimental results showing that BD55-5840 and BD55-5514 antibodies that bind to different binding epitopes can retain neutralizing efficacy against all examined variants BA.2.86, JN.1, KP.2 and KP.3. The results support the notion that evolution of Omicron variants may favor emergence of lineages with beneficial combinations of mutations involving mediators of epistatic couplings that control balance of high ACE2 affinity and immune evasion.

Investigating the medium‐range predictability of European heatwave onsets in relation to weather regimes using ensemble reforecasts

AbstractIn this study, the medium‐range predictability of heatwave (HW) onsets in four midlatitude European regions is investigated statistically with the help of ensemble reforecasts for the period 2001–2018. The concept of Euro‐Atlantic weather regimes is adopted to characterise HWs (about 50 in each region) and to study whether forecast skill may depend on the large‐scale dynamical setup. HW onsets over the British Isles and Scandinavia are mainly associated with Scandinavian and European blocking regimes, whereas the “no regime” case is observed more frequently for Central Europe. Stratified by weather regime, the predictability of heatwave onsets is then studied by means of a multiple metric‐based analysis of European Centre for Medium‐Range Weather Forecasts (ECMWF) and Global Ensemble Forecast System Version 12 (GEFSv12) ensemble reforecasts. For two of the regions considered, Central Europe and the British Isles, a conclusive picture is obtained: medium‐range predictive skill is significantly higher for HW onsets associated with Scandinavian or European blocking compared with cases with no pronounced regime. This skill advantage mostly concerns the large‐scale flow and, to some extent, 850‐hPa temperatures, but is generally not reflected in the correct prediction of near‐surface temperatures. Finally, we investigate for two regions how exceptionally good or poor forecasts are related to the atmospheric state during or shortly after forecast initialisation. At 10 days lead time, poor large‐scale flow predictive skill for Central European HW onsets is linked to anomalously high baroclinicity further upstream and an intensified North Atlantic jet stream, whereas good forecasts on average feature an initial state close to climatology. Forecast skill for near‐surface temperatures is not affected by such dynamical precursors, but rather by pre‐existing soil‐moisture anomalies. For the British region, exceptionally good forecasts of both large‐scale flow and near‐surface temperatures are associated with an already established continental blocking. In contrast to Central Europe, pre‐existing soil‐moisture anomalies play less of a role there.

Development of a monthly PM2.5 forecast model for Seoul, Korea, based on the dynamic climate forecast and a convolutional neural network algorithm

A convolutional neural network (CNN) model has been developed to predict monthly concentrations of particulate matter with a diameter ≤ 2.5 μm (PM2.5) in Seoul, Republic of Korea, during winter months (November through February). Seven meteorological variables influencing PM2.5 concentrations were selected as predictors: geopotential heights at 1000 and 500 hPa, zonal and meridional winds, relative humidity and temperature at 850 hPa, and boundary layer height. These predictors were obtained from the fifth generation of the European Centre for Medium-Range Weather Forecasts reanalysis (ERA5) for model construction and from the Korea Polar Prediction System (KPOPS) dynamic forecasts for operational predictions. From the 11-year ERA5 dataset for 2008–2020, one year was allocated for testing, while the remaining 10 years were randomly assigned for either training (80%) or validation (20%). This process was repeated until all 11 years were allocated for testing, resulting in 11 CNN models. The benchmark which used the ERA5 dataset as the predictors yielded a mean bias error (MBE) of 0.05 μg m−3, root mean square error (RMSE) of 2.41 μg m−3, Pearson correlation coefficient (R) of 0.85, and 71% hit rate in predicting the months of high PM2.5 concentrations (≥ 30 μg m−3) for the testing periods. The 11-CNN-model ensemble predictions using the KPOPS forecasts resulted in an MBE of 0.1 μg m−3, RMSE of 3.19 μg m−3, R of 0.74, and hit rate of 62.5% for the period 2008–2020. Considering that the average temporal correlation coefficients between the ERA5 and the KPOPS forecasts for the seven predictors ranged from 0.20 to 0.66, the CNN-based PM2.5 prediction model has demonstrated good performance in overcoming the adverse effects of the KPOPS forecast errors on predicting PM2.5 concentrations. The CNN model showed limitations for unusual period such as the COVID-19 period, November 2020 through February 2022, when the emissions in East Asia were substantially below their climatological normal levels. In conclusion, the CNN model developed in this study showed potential applicability to operational monthly PM2.5 concentration predictions for the years of normal emissions using solely meteorological data. This can provide professionals in various fields, such as health care and environmental science, with valuable insights for long-term planning.

Impact of Data Assimilation in Sensitive Features on the Predictability of the 2012 Great Arctic Cyclone

AbstractThe Great Arctic Cyclone 2012 (AC12) is used to understand the role of initial condition errors in the predictability at 2–3‐day forecast range of a high‐impact summer Arctic Cyclone (AC). Ensemble sensitivity analysis (ESA) is first performed to identify potentially sensitive regions of the cyclone evolution using an ensemble baseline forecast with conventional in situ observations assimilated. A pseudo‐observation method is then introduced to investigate impacts of hypothetical observations in these sensitive but unobserved regions. In the baseline experiments with in situ observations assimilated, the forecasted AC12 reaches its peak intensity 18 hr earlier than in the verifying Global Forecast System Analysis (GFS‐ANL) and the cyclone track is biased toward the southwest. Using ESA, the time of peak intensity and the cyclone track error are identified to be sensitive to the upstream trough, downstream ridge, and the tropopause polar vortex (TPV) to the northeast (NE TPV) of the AC12. These features were not observed by the in situ observation networks. To examine the impact of the observation gaps, pseudo‐observations drawn from GFS‐ANL are assimilated. Pseudo‐observations sample the three features separately to study the impact of the initial condition error on the predictability of AC12. The cyclone peak intensity timing error and track error are greatly reduced when the initial condition error is reduced near the NE TPV. A southward expansion of the NE TPV and the corresponding southward shifting low‐level front lead the forecasted AC12 to progress to the east, which better agrees with the verifying GFS‐ANL.

Objective method of predicting monsoon onset over Kerala in medium and extended range time scale using Numerical Weather Prediction models

The southwest monsoon onset over Kerala (MOK) is very crucial for an Agricultural country like India hence its prediction in the extended range time scale (about 3 weeks in advance) is very useful for the Kharif season in India. As the declaration of MOK involves subjective interpretation of the forecasters, an objective method of prediction of MOK based on dynamical models could successfully avoid bogus onsets. Two objective prediction methods for MOK are developed based on the real-time extended range forecast (ERF) of rainfall measured over Kerala coast 080-120N, 740-780E and the strength and depth of the westerly wind over the Arabian Sea (050-120N, 550-750E) for the period from 2003 to 2022 are used in the first method. In addition to these 3 indices, meridional pressure gradient along the west coast of India is used in the second method with 4 indices. The MOK date is defined objectively based on these variables exceeding the thresholds in both the methods. The results indicate that the MOK forecast with four indices performed well compared to that with three indices during the whole period from 2003 to 2022 with the mean deviation days of MOK found to be 0.75 and 3.05 days respectively. Overall, the dynamically defined onset date over Kerala based on the real-time ERF indicated useful skill at least about 2–3 weeks in advance. It is also demonstrated that the district-level Multi-Model Ensemble forecast in the medium range (5 days in advance) based on 5 global models can add value to the categorical ERF of MOK for the exact forecast date of MOK.

Investigating the seasonal SST Predictability in the Northern Tropical Atlantic Ocean in an ensemble prediction system

The present study comprehensively investigates the practical and intrinsic predictability of the sea surface temperature (SST) in the Northern Tropical Atlantic (NTA) based on the 138-year-long coupled hindcasts with a recently developed seasonal ensemble prediction system. This system can yield skillful deterministic predictions for the prominent warm and cold events at least 6 months ahead. Notably, it excels in providing probabilistic predictions for below- and above-normal events rather than for neutral events. The predictability of SST in the NTA undergoes remarkable seasonal variation with two peaks of predictability targeted at April and October regardless of the lead time. Various sources of predictability for these target months are revealed. For the target month of April, the preceding remote forcing from the El Niño–Southern Oscillation (ENSO) in the tropical Pacific Ocean combined with local signal results in the phase locking of the SST variation and seasonality of signal component over the NTA. This ultimately contributes to the high predictability targeted at April. However, From the perspective of potential predictability of the predictability targeted at October, which has been rarely mentioned in previous studies. It is also encouraging that, similar to the Indian Ocean Dipole, ENSO and the signal-to-noise ratio of the system mainly contribute to predictability beyond persistence at long lead times for the spring SST in the NTA. This indicates that potential future ENSO improvements may leave much room for improvement in the current SST prediction in the NTA.

Enhancing irrigation water productivity using short-range ensemble weather forecasts at basin scale: A novel framework for regions with high hydro-climatic variability

Integrating short-term weather forecasts with crop growth models has emerged as a valuable decision-support tool for enhancing irrigation water productivity in water-intensive crops. Our primary focus lies in formulating a simplified Quasi-Farmer Behavior Routine to utilize ensemble weather forecasts to provide irrigation guidance. The study proposes the concept of Rainfall Confidence Quotient (RCQ) that is optimized based on the forecast uncertainty levels for rainfall categories that affect the irrigation decision. We utilize two publicly available ensemble weather forecasts from NCEP-GEFS and ECM and a regionally fine-tuned high-resolution Weather Research and Forecasting (WRF) ensemble forecast system with a 3-day lead time. We have refined our methodology by conducting 800 simulation experiments across diverse scenarios, each reflecting distinct climate regimes, soil types, cropping seasons, and management practices. The proposed method consists of a sequential 3-checkpoint algorithm incorporating ensemble rainfall forecasts, antecedent soil moisture, and evapotranspiration. During the wet crop season of 2015–16, the enhanced methodology yielded a substantial increase in irrigation water productivity (IWP), averaging around 20–30 % increase across spatial locations, in contrast to the conventional irrigation practices that do not consider weather forecasts. However, during 2016–17, a dry year, the improvement in IWP only ranged between 2–10 %. On average, water savings (at field scale) of 200–500 mm were recorded in the wet year crop season of 2015–16, while 100–300 mm of water savings were achieved during the dry year crop season of 2016–17, without any considerable reductions in the crop yield. The robustness and adaptability of the developed approach have been established through comprehensive evaluation with field observation and remote sensing techniques, suggesting its potential scalability to similar hydro-climatic typologies across the world.

Future Changes of Extreme Precipitation and Related Atmospheric Conditions in East Asia under Global Warming Projected in Large Ensemble Climate Prediction Data

Abstract Extreme precipitation is expected to pose a more severe threat to human society in the future. This work assessed the historical performance and future changes of extreme precipitation and related atmospheric conditions in a large ensemble climate prediction dataset, the database for Policy Decision-making for Future climate change (d4PDF), over East Asia. Compared with the TRMM and ERA5 datasets, the historical climate in d4PDF represents favorably the precipitation characteristics and the atmospheric conditions, although some differences are notable in the moisture, vertical motion, and cloud water fields. The future climate projection indicates that both the frequency and intensity of heavy precipitation events over East Asia increase compared with those in the present climate. However, when comparing the atmospheric conditions in the historical and future climates for the same precipitation intensity range, the future climate indicates smaller relatively humidity, weaker ascent, less cloud water content, and smaller temperature lapse rate, which negatively affect generating extreme precipitation events. The comparison of the precipitation intensity at the same amount of precipitable water between the historical and future climates indicates that extreme precipitation is weaker in the future, because of the more stabilized troposphere in the future. The general increase in extreme precipitation under future climate is primarily due to the enhanced increase in precipitable water in the higher temperature ranges, which counteracts the negative conditions of the stabilized troposphere.

Quantifying and reducing flood forecast uncertainty by the CHUP-BMA method

Abstract. The Bayesian model averaging (BMA), hydrological uncertainty processor (HUP), and HUP-BMA methods have been widely used to quantify flood forecast uncertainty. This study proposes the copula-based hydrological uncertainty processor BMA (CHUP-BMA) method by introducing a copula-based HUP in the framework of BMA to bypass the need for a normal quantile transformation of the HUP-BMA method. The proposed ensemble forecast scheme consists of eight members (two forecast precipitation inputs; two advanced long short-term memory, LSTM, models; and two objective functions used to calibrate parameters) and is applied to the interval basin between the Xiangjiaba and Three Gorges Reservoir (TGR) dam sites. The ensemble forecast performance of the HUP-BMA and CHUP-BMA methods is explored in the 6–168 h forecast horizons. The TGR inflow forecasting results show that the two methods can improve the forecast accuracy over the selected member with the best forecast accuracy and that the CHUP-BMA performs much better than the HUP-BMA. Compared with the HUP-BMA method, the forecast interval width and continuous ranked probability score metrics of the CHUP-BMA method are reduced by a maximum of 28.42 % and 17.86 % within all forecast horizons, respectively. The probability forecast of the CHUP-BMA method has better reliability and sharpness and is more suitable for flood ensemble forecasts, providing reliable risk information for flood control decision-making.

Are Protein Conformational Ensembles in Agreement with Experimental Data? A Geometrical Interpretation of the Problem.

The conformational variability of biological macromolecules can play an important role in their biological function. Therefore, understanding conformational variability is expected to be key for predicting the behavior of a particular molecule in the context of organism-wide studies. Several experimental methods have been developed and deployed for accessing this information, and computational methods are continuously updated for the profitable integration of different experimental sources. The outcome of this endeavor is conformational ensembles, which may vary significantly in properties and composition when different ensemble reconstruction methods are used, and this raises the issue of comparing the predicted ensembles against experimental data. In this article, we discuss a geometrical formulation to provide a framework for understanding the agreement of an ensemble prediction to the experimental observations.

Chasing parts in quadrillion: applications of dynamical downscaling in atmospheric pollutant transport modelling during field campaigns

Atmospheric transport and dispersion models (ATDMs) are widely used to study and forecast pollution events. In the frame of the “Effect of Megacities on the transport and transformation of pollutants on the regional to global scales” (EMeRGe) project, ATDM forecasts were carried out to identify potential airborne sampling areas of perfluorocarbons (PFCs) emanating from controlled PFC releases. The forecasts involved short-distance transport over small-scale topographic maxima (Manilla; Philippines), short-distance transport over large-scale topographic maxima (Taipei, Taiwan) and long-distance transport over mixed topography (Nanjing, China, sampled over Taiwan). In situ aircraft measurements of PFC mixing ratios down to a few parts per quadrillion (ppqv) provide us with a unique dataset to explore the added benefits of dynamical downscaling. Transport simulations were repeated using FLEXPART driven by ERA5 and IFS meteorological data and FLEXPART-WRF with dynamically downscaled IFS data down to 1.1 km and four PBL parametrisations. Of the three cases studied, dynamical downscaling led to significant differences for the Manilla and Taipei releases that can be interpreted through changes in the modelled orographic flow regimes. The choice of PBL scheme also significantly impacted accuracy, but there was no systematically better-performing option, highlighting the benefits of ensemble forecasting. Results show how convergence and divergence between ensemble members can be utilised to help decision-making during field campaigns. This study highlights the role that dynamical downscaling can play as an important component in campaign planning when dealing with observations over orographically complex areas.

Definition of the effector landscape across 13 phytoplasma proteomes with LEAPH and EffectorComb.

'Candidatus Phytoplasma' genus, a group of fastidious phloem-restricted bacteria, can infect a wide variety of both ornamental and agro-economically important plants. Phytoplasmas secrete effector proteins responsible for the symptoms associated with the disease. Identifying and characterizing these proteins is of prime importance for expanding our knowledge of the molecular bases of the disease. We faced the challenge of identifying phytoplasma's effectors by developing LEAPH, a machine learning ensemble predictor composed of four models. LEAPH was trained on 479 proteins from 53 phytoplasma species, described by 30 features. LEAPH achieved 97.49% accuracy, 95.26% precision and 98.37% recall, ensuring a low false-positive rate and outperforming available state-of-the-art methods. The application of LEAPH to 13 phytoplasma proteomes yields a comprehensive landscape of 2089 putative pathogenicity proteins. We identified three classes according to different secretion models: 'classical', 'classical-like' and 'non-classical'. Importantly, LEAPH identified 15 out of 17 known experimentally validated effectors belonging to the three classes. Furthermore, to help the selection of novel candidates for biological validation, we applied the Self-Organizing Maps algorithm and developed a Shiny app called EffectorComb. LEAPH and the EffectorComb app can be used to boost the characterization of putative effectors at both computational and experimental levels, and can be employed in other phytopathological models.

Grand ensemble forecasts verification based on two high resolution (∼12 km) global ensemble prediction systems

The primary objective of this study is to quantitatively assess the skill of grand-ensemble forecasts generated from two high-resolution global ensemble prediction systems (EPSs). We have used the National Centre for Medium Range Weather Forecasting (NCMRWF), India global EPS, denoted as NEPS, and India Meteorological Department (IMD) EPS, known as GEFS. Both of these operational EPSs operate at ∼12 km horizontal grid size and produce ensemble forecast outputs. In this study, these two high-resolution EPSs based grand-ensemble products have been evaluated.Statistical analysis is conducted using seven-day forecasts from NEPS's 22 perturbed members and GEFS's 20 perturbed members. The performance of the global grand ensemble, consisting of 42 members, is evaluated for summer monsoon season, 1st July to 30th September 2019 in comparison to the individual EPS forecasts. The verification metrics are prepared using variables such as Zonal Wind and Temperature at 850 hPa, Geopotential at 500 hPa and Precipitation. These metrics are applied across both the Northern Hemisphere (20°N-90°N) and over the domain located in the South Asia encompassing India and its neighbouring areas (0–40 oN & 50–100°E). The verification metrics employed include ensemble spread, root-mean-square error (RMSE) of ensemble mean, Brier score, relative operating characteristic, area skill score, reliability, rank histogram, ranked probability score, continuous ranked probability score, and relative economic value.We noted that the grand ensemble forecasts in this study carries a higher skill compared to the two constituent models for lead time >3 days and for higher forecast probability for the longer lead times. Lower RMSE in the grand ensemble for longer lead-time forecasts (day 4–7) indicates a better representation when compared to forecasts from individual EPSs. However, it is worth noting that shorter lead-time forecasts (day 1–3) in the grand ensemble could potentially result in over-dispersion. Brier scores further indicate an enhancement in predictability of nearly one day in grand ensemble compared to the individual EPSs for T850, U850 and Z500. RPS of rainfall in a smaller domain centring on India indicates a better score for grand ensemble compared to both the EPS. Grand ensembles demonstrate its ability to provide better reliability and discrimination quality. The economic value of the forecasts based on grand ensemble, is higher with a wider range of cost/loss ratio compared to those in the individual EPSs.

Skilful probabilistic medium‐range precipitation and temperature forecasts over Vietnam for the development of a future dengue early warning system

AbstractDengue fever is a source of substantial health burden in Vietnam. Given the well‐established influence of temperature and precipitation on vector biology and disease transmission, predictions of meteorological variables, such as those issued by ECMWF as a world‐leading provider of global ensemble forecasts, are likely to be valuable model inputs to a future dengue early warning system. In the absence of established verification at municipal and regional scales, this study assesses the skill of rainy season (May–October) ensemble precipitation and 2‐m temperature retrospective forecasts over North and South Vietnam initialized for dates during the period 2001–2020, evaluated against the ERA5 reanalysis for the same period. Forecasts are found to be significantly skilful compared with both climatology and persistence for lead times up to 10 days, including for cumulative precipitation values considered against independent rain gauge data. Rank histograms demonstrate that ensembles generally avoid excessive bias and consistently positive CRPSS values indicate substantial skill for temperature and cumulative precipitation forecasts for all spatial scales considered, despite differences in rainy season characteristics between North and South Vietnam. This forecast reliability demonstrates that meteorological input data based on ECMWF ensemble forecasts would add appreciably more value to the development of a future dengue early warning system compared to reference forecasts like climatology or persistence. These results raise hope for further exploration of predictive skill for relevant meteorological variables, particularly focused on their downscaling to produce district‐level epidemiological forecasts for urban areas where dengue is most prevalent.

Using Machine Learning to Predict Convection-Allowing Ensemble Forecast Skill: Evaluation with the NSSL Warn-on-Forecast System

Abstract Forecasters routinely calibrate their confidence in model forecasts. Ensembles inherently estimate forecast confidence but are often underdispersive, and ensemble spread does not strongly correlate with ensemble-mean error. The misalignment between ensemble spread and skill motivates new methods for “forecasting forecast skill” so that forecasters can better utilize ensemble guidance. We have trained logistic regression and random forest models to predict the skill of composite reflectivity forecasts from the NSSL Warn-on-Forecast System (WoFS), a 3-km ensemble that generates rapidly updating forecast guidance for 0–6-h lead times. The forecast skill predictions are valid at 1-, 2-, or 3-h lead times within localized regions determined by the observed storm locations at analysis time. We use WoFS analysis and forecast output and NSSL Multi-Radar/Multi-Sensor composite reflectivity for 106 cases from the 2017 to 2021 NOAA Hazardous Weather Testbed Spring Forecasting Experiments. We frame the prediction task as a multiclassification problem, where the forecast skill labels are determined by averaging the extended fraction skill scores (eFSSs) for several reflectivity thresholds and verification neighborhoods and then converting to one of three classes based on where the average eFSS ranks within the entire dataset: POOR (bottom 20%), FAIR (middle 60%), or GOOD (top 20%). Initial machine learning (ML) models are trained on 323 predictors; reducing to 10 or 15 predictors in the final models only modestly reduces skill. The final models substantially outperform carefully developed persistence- and spread-based models and are reasonably explainable. The results suggest that ML can be a valuable tool for guiding user confidence in convection-allowing (and larger-scale) ensemble forecasts. Significance Statement Some numerical weather prediction (NWP) forecasts are more likely to verify than others. Forecasters often recognize situations where NWP output should be trusted more or less than usual, but objective methods for “forecasting forecast skill” are notably lacking for thunderstorm-scale models. Better estimates of forecast skill can benefit society through more accurate forecasts of high-impact weather. Machine learning (ML) provides a powerful framework for relating forecast skill to the characteristics of model forecasts and available observations over many previous cases. ML models can leverage these relationships to predict forecast skill for new cases in real time. We demonstrate the effectiveness of this approach to forecasting forecast skill using a cutting-edge thunderstorm prediction system and logistic regression and random forest models. Based on this success, we recommend the adoption of similar ML-based methods for other prediction models.

A novel ensemble deep reinforcement learning model for short‐term load forecasting based on Q‐learning dynamic model selection

AbstractShort‐term load forecasting is critical for power system planning and operations, and ensemble forecasting methods for electricity loads have been shown to be effective in obtaining accurate forecasts. However, the weights in ensemble prediction models are usually preset based on the overall performance after training, which prevents the model from adapting in the face of different scenarios, limiting the improvement of prediction performance. In order to improve the accurateness and validity of the ensemble prediction method further, this paper proposes an ensemble deep reinforcement learning approach using Q‐learning dynamic weight assignment to consider local behaviours caused by changes in the external environment. Firstly, the variational mode decomposition is used to reduce the non‐stationarity of the original data by decomposing the load sequence. Then, the recurrent neural network (RNN), long short‐term memory (LSTM), and gated recurrent unit (GRU) are selected as the basic power load predictors. Finally, the optimal weights are ensembled for the three sub‐predictors by the optimal weights generated using the Q‐learning algorithm, and the final results are obtained by combining their respective predictions. The results show that the forecasting capability of the proposed method outperforms all sub‐models and several baseline ensemble forecasting methods.

Using comparative extinction risk analysis to prioritize the IUCN Red List reassessments of amphibians

AbstractAssessing the extinction risk of species based on the International Union for Conservation of Nature (IUCN) Red List (RL) is key to guiding conservation policies and reducing biodiversity loss. This process is resource demanding, however, and requires continuous updating, which becomes increasingly difficult as new species are added to the RL. Automatic methods, such as comparative analyses used to predict species RL category, can be an efficient alternative to keep assessments up to date. Using amphibians as a study group, we predicted which species are more likely to change their RL category and thus should be prioritized for reassessment. We used species biological traits, environmental variables, and proxies of climate and land‐use change as predictors of RL category. We produced an ensemble prediction of IUCN RL category for each species by combining 4 different model algorithms: cumulative link models, phylogenetic generalized least squares, random forests, and neural networks. By comparing RL categories with the ensemble prediction and accounting for uncertainty among model algorithms, we identified species that should be prioritized for future reassessment based on the mismatch between predicted and observed values. The most important predicting variables across models were species’ range size and spatial configuration of the range, biological traits, climate change, and land‐use change. We compared our proposed prioritization index and the predicted RL changes with independent IUCN RL reassessments and found high performance of both the prioritization and the predicted directionality of changes in RL categories. Ensemble modeling of RL category is a promising tool for prioritizing species for reassessment while accounting for models’ uncertainty. This approach is broadly applicable to all taxa on the IUCN RL and to regional and national assessments and may improve allocation of the limited human and economic resources available to maintain an up‐to‐date IUCN RL.

Analysis and Prediction of Summer Rainfall over Southwestern Utah

Abstract While many studies have examined intense rainfall and flash flooding during the North American monsoon (NAM) in Arizona, Nevada, and New Mexico, less attention has focused on the NAM’s extension into southwestern Utah. This study relates flash flood reports and Multi-Radar Multi-Sensor (MRMS) precipitation across southwestern Utah to atmospheric moisture content and instability analyses and forecasts from the High-Resolution Rapid Refresh (HRRR) model during the 2021–23 monsoon seasons. MRMS quantitative precipitation estimates over southwestern Utah during the summer depend largely on the areal coverage from the KICX WSR-88D radar near Cedar City, Utah. Those estimates are generally consistent with the limited number of precipitation gauge reports in the region except at extended distances from the radar. A strong relationship is evident between days with widespread precipitation and afternoons with above-average precipitable water (PWAT) and convective available potential energy (CAPE) estimated from HRRR analyses across the region. Time-lagged ensembles of HRRR forecasts (initialization times from 0300 to 0600 UTC) that are 13–18 h prior to the afternoon period when convection is initiating (1800–2100 UTC) are useful for situational awareness of widespread precipitation events after adjusting for underprediction of afternoon CAPE. Improved skill is possible using random forest classification relying only on PWAT and CAPE to predict days experiencing excessive (upper quartile) precipitation. Such HRRR predictions may be useful for forecasters at the Salt Lake City National Weather Service Forecast Office to assist in issuing flash flood potential statements for visitors to national parks and other recreational areas in the region. Significance Statement Summer flash floods in southwestern Utah are a risk to area residents and millions of visitors annually to the region’s national parks, monuments, and recreational areas. The likelihood of flash floods within the region’s catchments depends on the intense afternoon and early evening convection initiated by lift and instability primarily due to terrain–flow interactions over elevated plateaus and mountains. Forecasts at lead times of 13–18 h of moisture and instability from the operational High-Resolution Rapid Refresh model have the potential to predict summer afternoons that are likely to have increased risks for higher rainfall amounts across southwestern Utah, although they are not expected to predict the likelihood of flash floods in any specific locale.

Reduced-order modeling of unsteady fluid flow using neural network ensembles

The use of deep learning has become increasingly popular in reduced-order models (ROMs) to obtain low-dimensional representations of full-order models. Convolutional autoencoders (CAEs) are often used to this end as they are adept at handling data that are spatially distributed, including solutions to partial differential equations. When applied to unsteady physics problems, ROMs also require a model for time-series prediction of the low-dimensional latent variables. Long short-term memory (LSTM) networks, a type of recurrent neural network useful for modeling sequential data, are frequently employed in data-driven ROMs for autoregressive time-series prediction. When making predictions at unseen design points over long time horizons, error propagation is a frequently encountered issue, where errors made early on can compound over time and lead to large inaccuracies. In this work, we propose using bagging, a commonly used ensemble learning technique, to develop a fully data-driven ROM framework referred to as the CAE-eLSTM ROM that uses CAEs for spatial reconstruction of the full-order model and LSTM ensembles for time-series prediction. When applied to two unsteady fluid dynamics problems, our results show that the presented framework effectively reduces error propagation and leads to more accurate time-series prediction of latent variables at unseen points.