Explore the vast range of titles available.

During the second Wind Forecast Improvement Project (WFIP2; October 2015–March 2017, held in the Columbia River Gorge and Basin area of eastern Washington and Oregon states), several improvements to the parameterizations used in the High Resolution Rapid Refresh (HRRR – 3 km horizontal grid spacing) and the High Resolution Rapid Refresh Nest (HRRRNEST – 750 m horizontal grid spacing) numerical weather prediction (NWP) models were tested during four 6-week reforecast periods (one for each season). For these tests the models were run in control (CNT) and experimental (EXP) configurations, with the EXP configuration including all the improved parameterizations. The impacts of the experimental parameterizations on the forecast of 80 m wind speeds (wind turbine hub height) from the HRRR and HRRRNEST models are assessed, using observations collected by 19 sodars and three profiling lidars for comparison. Improvements due to the experimental physics (EXP vs. CNT runs) and those due to finer horizontal grid spacing (HRRRNEST vs. HRRR) and the combination of the two are compared, using standard bulk statistics such as mean absolute error (MAE) and mean bias error (bias). On average, the HRRR 80 m wind speed MAE is reduced by 3 %–4 % due to the experimental physics. The impact of the finer horizontal grid spacing in the CNT runs also shows a positive improvement of 5 % on MAE, which is particularly large at nighttime and during the morning transition. Lastly, the combined impact of the experimental physics and finer horizontal grid spacing produces larger improvements in the 80 m wind speed MAE, up to 7 %–8 %. The improvements are evaluated as a function of the model's initialization time, forecast horizon, time of the day, season of the year, site elevation, and meteorological phenomena. Causes of model weaknesses are identified. Finally, bias correction methods are applied to the 80 m wind speed model outputs to measure their impact on the improvements due to the removal of the systematic component of the errors.

The traditional method for estimating weather forecast sensitivity to initial conditions uses adjoint models, which are limited to short lead times due to linearization around a control forecast. The advent of deep‐learning frameworks enables a new approach using backpropagation and gradient descent to iteratively optimize initial conditions, minimizing forecast errors. We apply this approach to the June 2021 Pacific Northwest heatwave using the GraphCast model, yielding over 90% reduction in 10‐day forecast errors over the Pacific Northwest. Similar improvements are found for Pangu‐Weather model forecasts initialized with the GraphCast‐derived optimal, suggesting that model error is an unimportant part of the perturbations. Eliminating small scales from the perturbations also yields similar forecast improvements. Extending the length of the optimization window, we find forecast improvement to about 23 days, suggesting atmospheric predictability at the upper end of recent estimates. Plain Language Summary This study examines a deep‐learning approach to understanding how small changes to initial conditions impact weather forecasts. Traditionally, a linear approach known as the adjoint method has been used to determine the sensitivity of forecasts to initial conditions. We leverage recent advancements in machine learning to find optimal initial conditions using the backpropagation method within deep‐learning frameworks. This approach iteratively searches for initial conditions that produce the best forecasts. We apply this method to GraphCast forecasts of the June 2021 Pacific Northwest extreme heatwave. We find that small changes to the initial conditions yield nearly perfect 10‐day weather forecasts of the heatwave in both the GraphCast and the Pangu‐Weather models. This research suggests that a significant increase in forecast skill may be possible from existing observations through better estimates of initial conditions. Key Points We use nonlinear gradient descent to optimize initial conditions for weather forecasting with machine learning models Application to the Pacific Northwest June 2021 heatwave reduces 10‐day forecast error by over 90 percent Forecast improvements are not sensitive to the forecast model, and derive mainly from analysis errors on synoptic and larger scales

The outer size of a tropical cyclone (TC) is important for predicting potential damage caused by the storm, yet improving the size forecast remains challenging. This study evaluates the TC intensity and outer size forecast performance of five numerical models from the National Oceanic and Atmospheric Administration (NOAA) based on 15 representative North Atlantic hurricanes from 2020 to 2022. We found there is little correlation between TC intensity and outer size forecast accuracy. Higher‐resolution models outperform coarser‐resolution models in intensity prediction but perform similarly for outer size regardless of resolution. Notably, the initial storm outer size may significantly influence the outer size forecast: initially larger storms may present greater challenges for accurate size prediction. Our results provide new insights into the complex relationship between TC size and intensity forecasting, highlighting the need to understand how environmental factors affect size forecasts and their connection to model resolution and configuration.

“Best tracks” are National Hurricane Center (NHC) poststorm analyses of the intensity, central pressure, position, and size of Atlantic and eastern North Pacific basin tropical and subtropical cyclones. This paper estimates the uncertainty (average error) for Atlantic basin best track parameters through a survey of the NHC Hurricane Specialists who maintain and update the Atlantic hurricane database. A comparison is then made with a survey conducted over a decade ago to qualitatively assess changes in the uncertainties. Finally, the implications of the uncertainty estimates for NHC analysis and forecast products as well as for the prediction goals of the Hurricane Forecast Improvement Program are discussed.

The primary goal of the Second Wind Forecast Improvement Project (WFIP2) is to advance the state-of-the-art of wind energy forecasting in complex terrain. To achieve this goal, a comprehensive 18-month field measurement campaign was conducted in the region of the Columbia River basin. The observations were used to diagnose and quantify systematic forecast errors in the operational High-Resolution Rapid Refresh (HRRR) model during weather events of particular concern to wind energy forecasting. Examples of such events are cold pools, gap flows, thermal troughs/marine pushes, mountain waves, and topographic wakes. WFIP2 model development has focused on the boundary layer and surface-layer schemes, cloud–radiation interaction, the representation of drag associated with subgrid-scale topography, and the representation of wind farms in the HRRR. Additionally, refinements to numerical methods have helped to improve some of the common forecast error modes, especially the high wind speed biases associated with early erosion of mountain–valley cold pools. This study describes the model development and testing undertaken during WFIP2 and demonstrates forecast improvements. Specifically, WFIP2 found that mean absolute errors in rotorlayer wind speed forecasts could be reduced by 5%–20% in winter by improving the turbulent mixing lengths, horizontal diffusion, and gravity wave drag. The model improvements made in WFIP2 are also shown to be applicable to regions outside of complex terrain. Ongoing and future challenges in model development will also be discussed.

Hydropower facilities represent a key electricity generating resource in the U.S. Western Interconnection. These facilities rely upon forecasts of inflow when scheduling releases to generate electricity. However, hydropower operations represented in bulk power systems models do not reflect uncertainty in inflow forecasts. This study aims to evaluate how inflow forecast uncertainties impact hydropower generation and revenues at the scale of an entire power grid at a spatial scale relevant to power system modeling. The question is critical and timely as more flexibility is called upon to integrate other technologies without understanding the flexibility already exercised. New advances are needed to represent hydropower contributions under operational uncertainty at the interconnection scale. Our contribution includes the development of consistent and coincident medium‐range (0–7 days) inflow forecasts and a generic hydropower scheduler, Forecast‐Informed Scheduler for Hydropower (FIScH), that captures non‐powered water management objectives and constraints and allows for varying electricity prices. This scheduler was applied at 242 hydropower facilities representing 86% of the conventional nameplate capacity in the Western Interconnection. Hydropower revenues were examined for schedules developed using three sets of inflow forecasts with differing levels of accuracy over a 20‐year period from 2000 to 2019. In aggregate, we find that annual hydropower revenue decreases 0.08% when using more skillful forecasts, and 0.11% when using baseline persistence forecasts as compared to revenue using perfect forecasts. Regional and interannual results were more varied and ranged between −1 and 4%. The translation of improved forecast skill into higher revenues is non‐linear and varies regionally, with larger revenue changes on the west coast and smaller responses across the interior western U.S. Overall, we demonstrate that scheduling mostly alleviates the impact of inflow forecast errors on hydropower revenue. The study motivates the need for a more detailed evaluation into which specific hydrologic events impact hydropower scheduling and revenue at the system scale.

Biased, incomplete numerical models are often used for forecasting states of complex dynamical systems by mapping an estimate of a “true” initial state into model phase space, making a forecast, and then mapping back to the “true” space. While advances have been made to reduce errors associated with model initialization and model forecasts, we lack a general framework for discovering optimal mappings between “true” dynamical systems and model phase spaces. Here, we propose using a data‐driven approach to infer these maps. Our approach consistently reduces errors in the Lorenz‐96 system with an imperfect model constructed to produce significant model errors compared to a reference configuration. Optimal pre‐ and post‐processing transforms leverage “shocks” and “drifts” in the imperfect model to make more skillful forecasts of the reference system. The implemented machine learning architecture using neural networks constructed with a custom analog‐adjoint layer makes the approach generalizable across applications. Plain Language Summary Modeling and forecasting natural systems, such as Earth's oceans and atmosphere, is difficult due to their inherent unpredictability, our incomplete understanding of their dynamics, and their vastness and complexity. One way to improve forecasts is by improving physical representations within numerical models. However, models will always have shortcomings. The alternative approach explored here is to maximize the utility of available imperfect or incomplete models by revising how the model is used and how its forecast is interpreted. Here, we employ machine learning to learn pre‐ and post‐processing operators, called cross‐attractor transforms (CATs), which reduce the overall forecast errors from imperfect models. We demonstrate the framework's efficacy by using a simplified dynamical model as an imperfect representation of a higher‐dimensional chaotic dynamical system, analogous to using a simple pendulum to forecast the behavior of a double pendulum. In addition to improving forecasts, CATs offer insights into how the two systems evolve in time. The approach is generalizable across dynamical systems and disciplines. Key Points Forecasts from imperfect models are improved using optimized pre‐ and post‐processing operators Neural networks trained on imperfect models and reference truth forecasts efficiently derive these operators Forecasts from this hybrid machine‐learning approach are more accurate than purely data‐driven methods when applied to an idealized system

Real estate markets are inherently dynamic, influenced by economic fluctuations, policy changes and socio-demographic shifts, often leading to emergence of anomalous—regions, where market behavior significantly deviates from expected trends. Traditional forecasting models struggle to handle such anomalies, resulting in higher errors and reduced prediction stability. In order to address this challenge, we propose EGCN, a novel cluster-specific forecasting framework that first detects and clusters anomalous regions separately from normal regions, and then applies forecasting models. This structured approach enables predictive models to treat normal and anomalous regions independently, leading to enhanced market insights and improved forecasting accuracy. Our evaluations on the UK, USA, and Australian real estate market datasets demonstrates that the EGCN achieves the lowest error both anomaly-free (baseline) methods and alternative anomaly detection methods, across all forecasting horizons (12, 24, and 48 months). In terms of anomalous region detection, our EGCN identifies 182 anomalous regions in Australia, 117 in the UK and 34 in the US, significantly more than the other competing methods, indicating superior sensitivity to market deviations. By clustering anomalies separately, forecasting errors are reduced across all tested forecasting models. For instance, when applying Neural Hierarchical Interpolation for Time Series Forecasting, the EGCN improves accuracy across forecasting horizons. In short-term forecasts (12 months), it reduces MSE from 1.3 to 1.0 in the US, 9.7 to 6.4 in the UK and 2.0 to 1.7 in Australia. For mid-term forecasts (24 months), EGCN achieves the lowest errors, lowering MSE from 3.1 to 2.3 (US), 14.2 to 9.0 (UK), and 4.5 to 4.0 (Australia). Even in long-term forecasts (48 months), where error accumulation is common, EGCN remains stable; decreasing MASE from 6.9 to 5.3 (US), 12.2 to 8.5 (UK), and 16.0 to 15.2 (Australia), highlighting its robustness over extended periods. These results highlight how separately clustering anomalies allows forecasting models to better capture distinct market behaviors, ensuring more precise and risk-adjusted predictions.

Accurate short‐to‐subseasonal streamflow forecasts are becoming crucial for effective water management in an increasingly variable climate. However, streamflow forecast remains challenging over extended lead times, uncertainty in meteorological inputs, and increased frequency and variability in extreme weather and climate events. We implemented a Future Time Series Transformer (FutureTST) model for streamflow forecasting that separately integrates past meteorological and streamflow data while incorporating future weather conditions. FutureTST achieves a mean Nash‐Sutcliffe Efficiency (NSE) of 0.82 to 0.67 for 1‐ to 30‐day streamflow forecasts. Incorporating upstream streamflow information improved forecast accuracy by up to 10%. During real‐time forecast, FutureTST maintains higher forecast skills of 9.03 for 1‐day and 5.74 for 14‐day forecasts. In contrast, calibrated process‐based hydrological model forecasts become unreliable beyond a 4‐day lead time. Our findings demonstrate the potential of FutureTST as a reliable streamflow forecasting tool that offers a valuable addition to operational flood monitoring systems and climate‐resilient decision‐making.

Structure and intensity forecasts of 19 tropical cyclones (TCs) during the 2020 Atlantic hurricane season are investigated using two NWP systems. An experimental 4-km global ECMWF model (EC4) with upgraded moist physics is compared with a 9-km version (EC9) to evaluate the influence of resolution. EC4 is then benchmarked against the 4-km regional COAMPS–Tropical Cyclones (COAMPS-TC) system (CO4) to compare systems with similar resolutions. EC4 produced stronger TCs than EC9, with a >30% reduction of the maximum wind speed bias in EC4, resulting in lower forecast errors. However, both ECMWF predictions struggled to intensify initially weak TCs, and the radius of maximum winds (RMW) was often too large. In contrast, CO4 had lower biases in central pressure, maximum wind speed, and RMW. Regardless, minimal statistical differences between CO4 and EC4 intensity errors were found for ≥36-h forecasts. Rapid intensification cases yielded especially large intensity errors. CO4 produced superior forecasts of RMW, together with an excellent pressure–wind relationship. Differences in the results are due to contrasting physics and initialization schemes. ECMWF uses global data assimilation with no special treatment of TCs, whereas COAMPS-TC constructs a vortex for TCs with initial intensity ≥55 kt (∼28 m s −1 ) based on data provided by forecasters. Two additional ECMWF experiments were conducted. The first yielded improvements when the drag coefficient was reduced at high wind speeds, thereby weakening the coupling between the low-level winds and the surface. The second produced overly intense TCs when explicit deep convection was used, due to unrealistic mid–upper-tropospheric heating.