High-Resolution Rapid Refresh Research Articles

Evaluating Machine Learning–Based Probabilistic Convective Hazard Forecasts Using The HRRR: Quantifying Hazard Predictability and Sensitivity to Training Choices

Abstract The High-Resolution Rapid Refresh (HRRR) model provides hourly updating forecasts of convective-scale phenomena, which can be used to infer the potential for convective hazards (e.g., tornadoes, hail, and wind gusts), across the United States. We used deterministic 2019–20 HRRR, version 4 (HRRRv4), forecasts to train neural networks (NNs) to generate 4-hourly probabilistic convective hazard forecasts [neural network probability forecasts (NNPFs)] for HRRRv4 initializations in 2021, using storm reports as ground truth. The NNPFs were compared to the skill of a smoothed updraft helicity (UH) baseline to quantify the benefit of the NNs. NNPF skill varied by initialization time and time of day but was all superior to the UH forecast. NNPFs valid at hours between 1800 and 0000 UTC were most skillful in aggregate, significantly exceeding the baseline forecast skill. Overnight NNPFs (i.e., valid 0600–1200 UTC) were least skillful, indicating a diurnal cycle in hazard predictability that was present across all HRRRv4 initializations. We explored the sensitivity of HRRRv4 NNPF skill to NN training choices. Including an additional year of 2021 HRRRv4 forecasts for training slightly improved skill for 2022 HRRRv4 NNPFs, while reducing the training dataset size by 40% using only forecasts with storm reports was not detrimental to forecast skill. Finally, NNs trained with 2018–20 HRRRv3 forecasts led to a reduction in NNPF skill when applied to 2021 HRRRv4 forecasts. In addition to documenting practical predictability challenges with convective hazard prediction, these findings reinforce the need for a consistent model configuration for optimal results when training NNs and provide best practices when constructing a training dataset with operational convection-allowing model forecasts. Significance Statement Convective hazards, such as hail and tornadoes, are often challenging to predict. To improve hazard predictions, we used machine learning (ML) to generate forecasts of convective hazards across the United States leveraging forecasts of prior events. The ML hazard forecasts were consistently better than a non-ML approach and varied in skill based on the time of day, with nighttime forecasts being particularly challenging. ML forecasts of wind gusts and hail were more skillful than tornadoes. Different strategies of constructing the dataset of prior events led to differences in forecast performance; thus, this work provides recommendations for how to assemble these datasets and train ML models to generate improved forecasts of severe weather events.

Retrieval of boundary layer precipitable water from GOES ABI using machine learning techniques

Abstract Low-level moisture is an important ingredient for forecasting severe storms, especially over the Great Plains where severe storms often develop along the dryline. Although ground-based observation systems such as lidar or radiosondes on weather balloons provide accurate information on low-level moisture, data are provided at limited locations, and the low temporal resolution of radiosondes makes it difficult to track a rapidly developing dryline. Geostationary satellites provide high spatial and temporal observation, but the channels of current geostationary satellites are mostly sensitive to water vapor at mid to upper levels. However, the split window difference (SWD) between the “clean” window channel (10.3 μm) and the “dirty” window channel (12.3 μm) is commonly used for estimating low-level water vapor. However, this estimation is complicated by surface temperature contributions, dependence on the lapse rate, and nonlinear relationships between SWD and moisture. This study applies machine learning techniques to infer boundary layer precipitable water (BLPW) from Geostationary Operational Environmental Satellite (GOES) Advanced Baseline Imager (ABI) data. Since there are few observations that cover wide regions for training convolutional neural networks, especially for the atmosphere above the surface, High-Resolution Rapid Refresh (HRRR) model outputs are used as the truth for training. Statistical results show that using ABI channel 13, 14, and 15 brightness temperatures, skin temperature, solar zenith angle, and GOES total precipitable water product as additional inputs show the lowest MSE. Case study results show that the model developed in this study is good at capturing strong moisture gradients and depicting a dryline.

Representation of Blowing Snow and Associated Visibility Reduction in an Operational High-Resolution Weather Model

Abstract Blowing snow is a hazard for motorists because it may rapidly reduce visibility. Numerical weather prediction models in the United States do not capture the movement of snow once it reaches the ground, but visibility reductions due to blowing snow can be diagnosed based on model-predicted land surface and environmental conditions that correlate with blowing snow occurrence. A recently developed diagnostic framework for forecasting blowing snow concentration and the associated visibility reduction is applied to High-Resolution Rapid Refresh (HRRR) and Rapid Refresh Forecast System (RRFS) model output including surface snow conditions to predict surface visibility reduction due to blowing snow. Twelve blowing snow events around Wyoming from 2018 to 2023 are examined. The analysis shows that visibility reductions due to blowing snow tend to be overpredicted, caused by the initial assumption of full driftability of the snowpack. This study refines the aging of the blowing snow reservoir with two methods. The first method estimates driftability based on time-varying snow density from the Rapid Update Cycle land surface model (RUC LSM) used in the HRRR and experimental RRFS models and is evaluated in a real-time context with the RRFS model. The second, complementary method diagnoses snowpack driftability using a process-based approach that requires data for recent snowfall, wind speed, and skin temperature. Compared to the full driftability assumption, this method shows limited improvements in forecasting skill. To improve model-based diagnosis of visibility reduction due to blowing snow, empirical work is needed to determine the relation between snowpack driftability and the recent history of snowfall and other weather conditions. Significance Statement Blowing snow presents a significant hazard to motorists and airport operations through sometimes very rapid and intense reductions in visibility, yet little predictive guidance exists for blowing snow. This study aims to improve the prediction of blowing snow occurrence and associated surface visibility reduction using diagnostics from an operational high-resolution weather model. One key challenge regards the question of driftability of the snowpack. This study evaluates two approaches to quantify driftability in terms of visibility reduction due to blowing snow and acknowledges that more measurements are needed to improve the representation of blowing snow physics in NWP models.

Validation of Cool-Season Snowfall Forecasts at a High-Elevation Site in Utah’s Little Cottonwood Canyon

Abstract Producing a quantitative snowfall forecast (QSF) typically requires a model quantitative precipitation forecast (QPF) and snow-to-liquid ratio (SLR) estimate. QPF and SLR can vary significantly in space and time over complex terrain, necessitating fine-scale or point-specific forecasts of each component. Little Cottonwood Canyon (LCC) in Utah’s Wasatch Range frequently experiences high-impact winter storms and avalanche closures that result in substantial transportation and economic disruptions, making it an excellent testbed for evaluating snowfall forecasts. In this study, we validate QPFs, SLR forecasts, and QSFs produced by or derived from the Global Forecast System (GFS) and High-Resolution Rapid Refresh (HRRR) using liquid precipitation equivalent (LPE) and snowfall observations collected during the 2019/20–2022/23 cool seasons at the Alta–Collins snow-study site (2945 m MSL) in upper LCC. The 12-h QPFs produced by the GFS and HRRR underpredict the total LPE during the four cool seasons by 33% and 29%, respectively, and underpredict 50th, 75th, and 90th percentile event frequencies. Current operational SLR methods exhibit mean absolute errors of 4.5–7.7. In contrast, a locally trained random forest algorithm reduces SLR mean absolute errors to 3.7. Despite the random forest producing more accurate SLR forecasts, QSFs derived from operational SLR methods produce higher critical success indices since they exhibit positive SLR biases that offset negative QPF biases. These results indicate an overall underprediction of LPE by operational models in upper LCC and illustrate the need to identify sources of QSF bias to enhance QSF performance. Significance Statement Winter storms in mountainous terrain can disrupt transportation and threaten life and property due to road snow and avalanche hazards. Snow-to-liquid ratio (SLR) is an important variable for snowfall and avalanche forecasts. Using high-quality historical snowfall observations and atmospheric analyses, we developed a machine learning technique for predicting SLR at a high mountain site in Utah’s Little Cottonwood Canyon that is prone to closure due to winter storms. This technique produces improved SLR forecasts for use by weather forecasters and snow-safety personnel. We also show that current operational models and SLR techniques underforecast liquid precipitation amounts and overforecast SLRs, respectively, which has implications for future model development.

Precipitation Proxies for Flash Flooding: A Seven-Year Analysis over the Contiguous United States

Abstract Flash flooding remains a challenging prediction problem, which is exacerbated by the lack of a universally accepted definition of the phenomenon. In this article, we extend prior analysis to examine the correspondence of various combinations of quantitative precipitation estimates (QPEs) and precipitation thresholds to observed occurrences of flash floods, additionally considering short-term quantitative precipitation forecasts from a convection-allowing model. Consistent with previous studies, there is large variability between QPE datasets in the frequency of “heavy” precipitation events. There is also large regional variability in the best thresholds for correspondence with reported flash floods. In general, flash flood guidance (FFG) exceedances provide the best correspondence with observed flash floods, although the best correspondence is often found for exceedances of ratios of FFG above or below unity. In the interior western United States, NOAA Atlas 14 derived recurrence interval thresholds (for the southwestern United States) and static thresholds (for the northern and central Rockies) provide better correspondence. The 6-h QPE provides better correspondence with observed flash floods than 1-h QPE in all regions except the West Coast and southwestern United States. Exceedances of precipitation thresholds in forecasts from the operational High-Resolution Rapid Refresh (HRRR) generally do not correspond with observed flash flood events as well as QPE datasets, but they outperform QPE datasets in some regions of complex terrain and sparse observational coverage such as the southwestern United States. These results can provide context for forecasters seeking to identify potential flash flood events based on QPE or forecast-based exceedances of precipitation thresholds. Significance Statement Flash floods result from heavy rainfall, but it is difficult to know exactly how much rain will cause a flash flood in a particular location. Furthermore, different precipitation datasets can show very different amounts of precipitation, even from the same storm. This study examines how well different precipitation datasets and model forecasts, used by forecasters to warn the public of flash flooding, represent heavy rainfall leading to flash flooding around the United States. We found that different datasets have dramatically different numbers of heavy rainfall events and that high-resolution model forecasts of heavy rain correspond with observed flash flood events about as well as precipitation datasets based on rain gauge and radar in some regions of the country with few observations.

An air quality and boundary layer dynamics analysis of the Los Angeles basin area during the Southwest Urban NOx and VOCs Experiment (SUNVEx)

Abstract. The NOAA Chemical Sciences Laboratory (CSL) conducted the Southwest Urban NOx and VOCs Experiment (SUNVEx) to study emissions and the role of boundary layer (BL) dynamics and sea-breeze (SB) transitions in the evolution of coastal air quality. The study presented utilizes remote sensing and in situ observations in Pasadena, California. Separate analyses are conducted on the synoptic conditions during ozone (O3) exceedance (>70 ppb) and non-exceedance (<70 ppb) days, as well as the fine-structure variability of in situ chemistry measurements during BL growth and SB transitions. Diurnal analyses spanning August 2021 revealed a markedly different wind direction during evenings preceding O3 exceedance (northerly) versus non-exceedance (easterly) days. Increased O3 occurred simultaneously with warmer and drier conditions, a reduction in winds, and an increase in volatile organic compounds (VOCs) and fine particulate matter (PM2.5). While the average BL height was lower and surface pressure was higher, the day-to-day variability of these quantities led to an overall weak statistical relationship. Investigations focused on the fine-structure variability of in situ chemistry measurements superimposed on background trends were conducted using a novel multivariate spectral coherence mapping (MSCM) technique that combined the spectral structure of two or more independent measurements through a wavelet analysis as reported by maximum-normalized scaleograms. A case study was chosen to illustrate the MSCM technique, where the dominant peaks in scaleograms were identified and compared to BL height during the growth phase. The temporal widths of peaks (τmax) derived from VOC and nitrogen oxide (NOx) scaleograms, as well as scaleograms combining VOCs, NOx, and variations in BL height, indicated a broadening with respect to time as the BL increased in depth. A separate section focused on comparisons between τmax and BL height during August 2021 revealed uncorrelated or weakly correlated scatter, except in the case of VOCs when really large τmax and relatively deep BL heights were ignored. Instances of large τmax and relatively deep BL heights occurred near sunrise and as onshore flow entered Pasadena, respectively. Wind transitions likely influenced both the dynamical evolution of the BL and tracer advection and thus offer additional challenges when separating factors contributing to the fine structure. Other insights gained from this work include observations of descending wind jets from the San Gabriel Mountains that were not resolved by the High-Resolution Rapid Refresh (HRRR) model and the derivation of intrinsic properties of oscillations observed in NOx and O3 during the interaction between an SB and enhanced winds above the BL that flowed in opposition to the SB.

On the Role of the Meridional Jet and Horizontal Potential Vorticity Dipole in the Iowa Derecho of 10 August 2020

Abstract On 10 August 2020, a derecho caused widespread damage across Iowa and Illinois. Des Moines station data show that the arrival of the gust front was characterized by an abrupt shift to northerly flow, exceeding 22 m s−1 for ∼20 min. To test the hypothesis that this northerly jet is associated with a horizontal potential vorticity (PV) dipole in the lower troposphere, we investigated the structure of PV in the University of Wisconsin Nonhydrostatic Modeling System (UWNMS) and of absolute vorticity in High-Resolution Rapid Refresh (HRRR) forecast analyses. This structure is described here for the first time. The negative PV member coincides with the downdraft, while the positive PV member coincides with the updraft, with a northerly jet between. The westerly inflow jet descends anticyclonically in the downdraft, joining with northerly flow from the surface anticyclone. The resulting northerly outflow jet creates the trailing comma-shaped radar echo. The speed of propagation of the derecho is similar to the westerly wind maximum in the 3–5-km layer associated with the approaching synoptic cyclone, which acts as a steering level for resonant amplification. Idealized diagrams and 3D isosurfaces illustrate the commonality of the PV dipole/northerly jet structure. Differences in this structure among the three model states are related to low-level wind shear theory. The PV dipole coincides with the pattern of diabatic stretching tendency, which shifts westward and downward relative to the updraft/downdraft with increasing tilt. The PV dipole can contribute toward dynamical stability in a derecho. Significance Statement The purpose of this work is to investigate the structure of potential vorticity (PV) in the lower troposphere in a derecho. It is found that a northerly outflow jet occurs between an east–west-oriented horizontal PV dipole, which is described here for the first time. The negative PV member coincides with the downdraft and is inertially unstable, while the positive PV member coincides with the updraft. This work contributes toward the theory of resonant structures and longevity. The 3–5-km westerly inflow layer constitutes a steering level, which controls propagation speed despite differences in structure. The degree of westward tilt with height is related to the pattern of forcing by diabatic stretching in producing the PV dipole.

Comparing Distributions of Overshooting Convection in HRRR Forecasts to Observations

AbstractOvershooting convection can significantly impact the chemical and radiative properties of the upper troposphere and lower stratosphere through the transport of various chemical species. These impacts include enhancements of water vapor and ozone‐depleting halocarbons, which both have important consequences for climate change. Therefore, accurate prediction of the Earth's climate system requires convective overshooting to be included. To better understand how convective transport is represented in current state‐of‐the‐art models, approximately 75,000 individual updrafts in the central and eastern United States are analyzed from High‐Resolution Rapid Refresh (HRRR) simulations and NEXRAD radar observations during May and July 2021. Distributions of echo top potential temperatures and heights, as well as diurnal cycles of overshooting frequency, are compared to observations. These distributions show mean, median, and maximum echo tops 2–3 km lower than observations, both in absolute and tropopause‐relative space, with evidence of updrafts losing momentum too rapidly above the tropopause. Diurnal cycles show accurate times of maximum and minimum overshooting, but significant errors at model initialization and evidence that some simulated overshoots continue too late into the overnight hours. Despite these deficiencies, distributions of simulated levels of maximum detrainment show decent agreement with observations. All results, including the severe underprediction of echo top heights, persist at shorter forecast lead times. This indicates a need to improve representation of overshooting storms in weather and climate models, even those that are convection‐permitting, or introduce a transport parameterization.

Evaluation of Hub‐Height Wind Forecasts Over the New York Bight

ABSTRACTAs offshore wind energy development accelerates in the United States, it is important to assess the accuracy of hub‐height wind forecasts from numerical weather prediction models over the ocean. Leveraging approximately 2 years of Doppler lidar observations from buoys in the New York Bight, we evaluate 80‐m wind speed forecasts from two weather models: the High‐Resolution Rapid Refresh (HRRR) atmospheric model and the Global Forecast System (GFS) coupled atmosphere‐ocean model. These models have different horizontal grid spacing, vertical layering, initialization methods, and parameterizations of boundary layer mixing and surface–atmosphere interactions. Despite these differences, the models demonstrate similar and highly skillful short‐term forecasts at three measurement sites. At the Hudson Southwest location that provides a full year of data, their performance is statistically indistinguishable: root mean square error (RMSE) = 2.1 m/s and the Pearson correlation coefficient for 24‐h forecasts of both models, and RMSE = 2.6 m/s and 0.83 for 48‐h forecasts. Twenty‐four‐hour forecasts also exhibit skill in predicting quiescent winds and winds associated with maximum turbine power. By Day 10, GFS forecasts on average have almost no skill. The short‐term forecast skill by the HRRR and GFS does not strongly depend on season or time of day, yet we find some dependence of the models' performance on near‐surface stability. Additionally, 4‐ to 14‐day forecasts by the GFS exhibit lower RMSE during summer relative to other seasons. The high skill of the HRRR and GFS short‐term forecasts establishes confidence in their utility for offshore wind energy maintenance and operation.

Sundowner Winds at Montecito during the Sundowner Winds Experiment

This study investigates the predictability of downslope windstorms located in Santa Barbara County, California, locally referred to as Sundowner winds, from both observed relationships and a high-resolution, operational numerical weather prediction model. We focus on April 2022, during which the Sundowner Winds Experiment (SWEX) was conducted. We further refine our study area to the Montecito region owing to some of the highest wind measurements occurring at or near surface station MTIC1, situated on the coast-facing slope overlooking the area. Fires are not uncommon in this area, and the difficulty of egress makes the population particularly vulnerable. Area forecasters often use the sea-level pressure difference (ΔSLP) between Santa Barbara Airport (KSBA) and locations to the north such as Bakersfield (KBFL) to predict Sundowner windstorm occurrence. Our analysis indicates that ΔSLP by itself is prone to high false alarm rates and offers little information regarding downslope wind onset, duration, or magnitude. Additionally, our analysis shows that the high-resolution rapid refresh (HRRR) model has limited predictive skill overall for forecasting winds in the Montecito area. The HRRR, however, skillfully predicts KSBA-KBFL ΔSLP, as does GraphCast, a machine learning weather prediction model. Using a logistic regression model we were able to predict the occurrence of winds exceeding 9 m s−1 with a high probability of detection while minimizing false alarm rates compared to other methods analyzed. This provides a refined and easily computed algorithm for operational applications.

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.

Probabilistic Forecasts of Atmospheric River events using the HRRR Ensemble

The nine-member High-Resolution Rapid Refresh Ensemble (HRRRE) is evaluated for its ability to forecast five Atmospheric River (AR) events that impacted California in February–March 2019. Two sets of retrospective HRRRE simulations are conducted, a control with the standard set of perturbations (initial and boundary conditions, stochastic parameters, and physics tendency), and an experiment with initial and boundary perturbations only. Reliability plots suggest the HRRRE control represents the observed Stage IV precipitation frequency well at 6-h to 24-h lead times, and rank histograms suggest the ensemble is slightly underdispersive. The HRRRE overpredicts precipitation frequency at the higher (25 mm) threshold. These results suggest the HRRRE is a useful tool to quantify probabilistic forecasts of AR events in this region. Removing stochastic physics perturbations did not substantially impact probabilistic forecasts, suggesting most of the ensemble spread is from initial and boundary condition perturbations. Spatially, ensemble precipitation coefficient of variance is lower (less forecast uncertainty) over the Sierra Nevada range than other regions, suggesting that these ensemble perturbations have a smaller impact on precipitation processes occurring over the Sierra Nevada range. More work should be conducted to understand the impacts of other model perturbations, such as microphysics, on ensemble performance, and to improve Stage IV accuracy with frozen precipitation in mountainous regions.

On the Importance of Regime-Specific Evaluations for Numerical Weather Prediction Models as Demonstrated Using the High-Resolution Rapid Refresh (HRRR) Model

Abstract The scientific literature has many studies evaluating numerical weather prediction (NWP) models. However, many of those studies averaged across a myriad of different atmospheric conditions and surface forcings that can obfuscate the atmospheric conditions when NWP models perform well versus when they perform inadequately. To help isolate these different weather conditions, we used observations from the U.S. Climate Reference Network (USCRN) obtained between 1 January and 31 December 2021 to distinguish among different near-surface atmospheric conditions [i.e., different near-surface heating rates (), incoming shortwave radiation (SWd) regimes, and 5-cm soil moisture (SM05)] to evaluate the High-Resolution Rapid Refresh (HRRR) Model, which is a 3-km model used for operational weather forecasting in the United States. On days with small (large) , we found afternoon T biases of about 2°C (−1°C) and afternoon SWd biases of up to 170 W m−2 (100 W m−2), but negligible impacts on SM05 biases. On days with small (large) SWd, we found daytime temperature biases of about 3°C (−2.5°C) and daytime SWd biases of up to 190 W m−2 (80 W m−2). Whereas different SM05 had little impact on T and SWd biases, dry (wet) conditions had positive (negative) SM05 biases. We argue that the proper evaluation of weather forecasting models requires careful consideration of different near-surface atmospheric conditions and is critical to better identify model deficiencies in order to support improvements to the parameterization schemes used therein. A similar, regime-specific verification approach may also be used to help evaluate other geophysical models. Significance Statement Improving weather forecasting models requires careful evaluations against high-quality observations. We used observations from the U.S. Climate Reference Network (USCRN) and found that the performance of the High-Resolution Rapid Refresh (HRRR) Model varies as a function of differences in near-surface heating and solar radiation. This finding indicates that model evaluations need to be conducted under varying near-surface weather conditions rather than averaging across multiple weather types. This new approach will allow for model developers to better identify model deficiencies and is a useful step to helping improve weather forecasts.

Forecasting High Wind Events in the HRRR Model over Wyoming and Colorado. Part II: Sensitivity of Surface Wind Speeds to Model Resolution and Physics

Abstract Strong wind events can cause severe economic loss and societal impacts. Peak winds over Wyoming and Colorado occur during the wintertime months, and in the first part of this two-part series, it was shown that the High-Resolution Rapid Refresh (HRRR) model displays large negative biases with respect to strong wind events. In this part of the study, we address two questions: 1) does increasing the horizontal resolution improve the representation of strong wind events over this region, and 2) are the biases in HRRR-forecasted winds related to the selected planetary boundary layer (PBL), surface layer (SL), and/or land surface model (LSM) parameterizations? We conduct Weather Research and Forecasting (WRF) Model simulations to address these two main questions. Increasing the horizontal resolution leads to a small improvement in the simulation of strong wind speeds over the complex terrain of Wyoming and Colorado. In general, changes in the PBL, SL, and LSM parameterizations show much larger changes in simulated wind speeds compared to increasing the model resolution alone. Specifically, changing from the Mellor–Yamada–Nakanishi–Niino scheme to the Mellor–Yamada–Janjić PBL and SL schemes results in nearly no change in the r2 values, but there is a decrease in the magnitude of the strong wind speed bias (from −12.52 to −10.16 m s−1). We attribute these differences to differences in the diagnosis of surface wind speeds and mixing in the boundary layer. Further analysis is conducted to determine the value of 1-km forecasts of strong winds compared with wind speed diagnostics commonly used by the National Weather Service. Significance Statement Motivated by prior studies showing low skill in the prediction of strong winds using state-of-the-art weather forecast models, in this study, we aim to investigate two questions: 1) does increasing the horizontal resolution improve the prediction of strong wind events over the complex terrain of Wyoming and Colorado, and 2) are the biases in High-Resolution Rapid Refresh (HRRR) forecasted winds related to the selected planetary boundary layer, surface layer, and/or land surface model parameterizations? We find that increasing the horizontal resolution provides a slight improvement in the prediction of strong winds. Further, considerable improvement in the prediction of strong winds is found for varying boundary layer, surface layer, and land surface parameterizations.

Forecasting High Wind Events in the HRRR Model over Wyoming and Colorado. Part I: Evaluation of Wind Speeds and Gusts

Abstract Strong wind events cause significant societal damage ranging from loss of property and disruption of commerce to loss of life. Over portions of the United States, the strongest winds occur in the cold season and may be driven by interactions with the terrain (downslope winds, gap flow, and mountain wave activity). In the first part of this two-part series, we evaluate the High-Resolution Rapid Refresh (HRRR) model wind speed and gust forecasts for the 2016–22 winter months over Wyoming and Colorado, an area prone to downslope windstorms and gap flows due to its complex topography. The HRRR model exhibits a positive bias for low wind speeds/gusts and a large negative bias for strong wind speeds/gusts. In general, the model misses many strong wind events, but when it does predict strong winds, there is a high false alarm probability. An analysis of proxies for surface winds is conducted. Specifically, 700- and 850-mb (1 mb = 1 hPa) geopotential height gradients are found to be good proxies for strong wind speeds and gusts at two wind-prone locations in Wyoming. Given the good agreement between low-level height gradients and surface wind speeds yet a strong negative bias for strong wind speeds and gusts, there is a potential shortcoming in the boundary layer physics in the HRRR model with regard to predicting strong winds over complex terrain, which is the focus of the second part of this two-part study. Last, the sites with the largest strong wind speed bias are found to mostly sit on the leeward side of high mountains, suggesting that the HRRR model performs poorly in the prediction of downslope windstorms. Significance Statement We investigate the performance of the High-Resolution Rapid Refresh (HRRR) model with respect to strong wintertime wind speeds and gusts over the complex terrain of Wyoming and Colorado. We show that the overall performance of the HRRR model is low with regard to strong wind speed and wind gust forecasts across the investigated winter seasons, with a large negative bias in predicted strong wind speeds and gusts and a small positive bias for weak wind speeds and gusts. The largest biases are found to be on the leeward side of high mountains, indicating poor prediction of downslope winds. This study also utilizes National Weather Service forecasting metrics to understand their performance with respect to strong wind forecasts, and we find that they provide skill in forecasting these events.

Analysis of bias correction of HRRR model outputs for offshore wind power ramp events

This study examines the impact of bias correction on the prediction of offshore wind power ramp events (WPREs) using NOAA's High-Resolution Rapid Refresh (HRRR) model. The analysis focuses on New England's offshore region, using Light Detection and Ranging (LiDAR) wind observations and the power curve derived from the nearby Block Island wind farm as a reference. The study characterizes bias in terms of amplitude and phase, finding a 30-min systematic lag error and a marginal underestimation of high wind speeds. Perfect bias correction treatments including both the amplitude and the phase are applied to HRRR wind speed, and their impact on WPREs is assessed using categorical performance measures. A large discrepancy in the number of cases with HRRR having much fewer cases than the reference is found, which is attributed to the model's temporal and spatial resolution. The results show that the amplitude bias correction has a marginal impact on reducing frequency bias, while the dynamic time warping (DTW) used in the phase bias correction shows promising results but limited performance. The study concludes that phase bias correction can be a feasible option to reduce bias in WPREs in offshore power systems.

Evaluation of Wintertime Precipitation Estimates and Forecasts in the Mountains of Colorado

Abstract Wintertime precipitation poses many observational and forecasting challenges, especially in the complex topography of the western United States where radar beam blockage and difficulty siting in situ observations yields more sparse observations than in the eastern United States. Uncertainty in western U.S. winter precipitation is known to be high, so much so that some studies have found model simulated precipitation to produce similar or better large-scale estimates of annual precipitation than gridded observational products during climatologically anomalous years. This study evaluates high-resolution gridded precipitation estimates from Multi-Radar Multi-Sensor (MRMS) and Stage IV as well as forecasts from NOAA’s High-Resolution Rapid Refresh (HRRR) model in the Colorado Rocky Mountains. Gridded precipitation estimates and forecasts are compared with in situ SNOTEL measurements for two seasons of wintertime precipitation. The influence of forecast length, lead time, and model elevation on seasonal precipitation predictions from the HRRR are investigated. Additional comparisons are made with the relatively dense network of observations deployed in Colorado’s East River Watershed during the Study of Precipitation, the Lower Atmosphere and Surface for Hydrometeorology (SPLASH) campaign. Gridded products and forecasts are found to underestimate cold-season precipitation by 25%–65% relative to in situ and aircraft measurements, with longer forecast periods and lead times (6–24 h) having smaller biases (25%–30%) than shorter forecast periods and lead times (55%–65%). The assessment of multiple years of observations indicates that these biases are related more to the data and methods used to create the gridded products and forecasts than to precipitation characteristics. Significance Statement In the mountainous western United States, it is very challenging to both observe and forecast wintertime precipitation, yet snowfall plays an important role in providing the region’s annual water supply. This study aims to increase our understanding of the biases in observations and forecasts of snowfall in the Colorado Rocky Mountains, which can in turn impact forecasts of water availability for the ensuing warm season. In this study we find high-resolution gridded precipitation estimates and forecasts to underestimate cold-season precipitation when compared with in situ observing stations, with longer-range forecasts (e.g., daily) being the least biased. These findings were consistent over two years of study and have broad implications for the hydrologic modeling and water management communities.

Generative Ensemble Deep Learning Severe Weather Prediction from a Deterministic Convection-Allowing Model

Abstract An ensemble postprocessing method is developed for the probabilistic prediction of severe weather (tornadoes, hail, and wind gusts) over the conterminous United States (CONUS). The method combines conditional generative adversarial networks (CGANs), a type of deep generative model, with a convolutional neural network (CNN) to postprocess convection-allowing model (CAM) forecasts. The CGANs are designed to create synthetic ensemble members from deterministic CAM forecasts, and their outputs are processed by the CNN to estimate the probability of severe weather. The method is tested using High-Resolution Rapid Refresh (HRRR) 1–24-h forecasts as inputs and Storm Prediction Center (SPC) severe weather reports as targets. The method produced skillful predictions with up to 20% Brier skill score (BSS) increases compared to other neural-network-based reference methods using a testing dataset of HRRR forecasts in 2021. For the evaluation of uncertainty quantification, the method is overconfident but produces meaningful ensemble spreads that can distinguish good and bad forecasts. The quality of CGAN outputs is also evaluated. Results show that the CGAN outputs behave similarly to a numerical ensemble; they preserved the intervariable correlations and the contribution of influential predictors as in the original HRRR forecasts. This work provides a novel approach to postprocess CAM output using neural networks that can be applied to severe weather prediction. Significance Statement We use a new machine learning (ML) technique to generate probabilistic forecasts of convective weather hazards, such as tornadoes and hailstorms, with the output from high-resolution numerical weather model forecasts. The new ML system generates an ensemble of synthetic forecast fields from a single forecast, which are then used to train ML models for convective hazard prediction. Using this ML-generated ensemble for training leads to improvements of 10%–20% in severe weather forecast skills compared to using other ML algorithms that use only output from the single forecast. This work is unique in that it explores the use of ML methods for producing synthetic forecasts of convective storm events and using these to train ML systems for high-impact convective weather prediction.

NYSolarCast: A solar power forecasting system for New York State

Solar energy generation capacity will need to be greatly increased to meet aggressive clean energy targets by New York State (NYS), which are 70% renewable energy (RE) generation by 2030, and 100% by 2040. Because solar energy is a variable, weather-driven resource, accurate forecasts of solar energy generation both in nowcast (intra-day) and day-ahead time horizons are necessary for electric utilities and independent system operators, as they maintain grid stability and maximize RE use.Meeting this need for NYS, a gridded, open-source solar power forecasting system called NYSolarCast was developed (https://github.com/ncar/NYSolarCast_delivery). NYSolarCast makes 15-min resolution predictions of global horizontal irradiance (GHI) on a 3-km grid covering all of NYS, which are then used to estimate solar power generation both for select utility-scale PV plants (15-min resolution) and for zone-aggregated distributed PV (1-h resolution). The statewide GHI forecasts are made by applying the StatCast statistical forecasting model, which is trained on over two years of GHI observations from pyranometers at all 126 NYS Mesonet stations and gridded numerical weather prediction forecasts from both the Weather Research and Forecasting model tuned for solar applications (WRF-Solar®) and NOAA’s operational High-Resolution Rapid Refresh (HRRR) model. Forecasts are made at each NYS Mesonet site and then blended outward into the rest of the grid. This paper gives an overview of NYSolarCast performance for intra-day (0–6-h) GHI and power forecasts during a one-year period.

An Evaluation of NOAA Modeled and In Situ Soil Moisture Values and Variability across the Continental United States

Abstract Estimates of soil moisture from two National Oceanic and Atmospheric Administration (NOAA) models are compared to in situ observations. The estimates are from a high-resolution atmospheric model with a land surface model [High-Resolution Rapid Refresh (HRRR) model] and a hydrologic model from the NOAA Climate Prediction Center (CPC). Both models produce wetter soils in dry regions and drier soils in wet regions, as compared to the in situ observations. These soil moisture differences occur at most soil depths but are larger at the deeper depths below the surface (100 cm). Comparisons of soil moisture variability are also assessed as a function of soil moisture regime. Both models have lower standard deviations as compared to the in situ observations for all soil moisture regimes. The HRRR model’s soil moisture is better correlated with in situ observations for drier soils as compared to wetter soils—a trend that was not present in the CPC model comparisons. In terms of seasonality, soil moisture comparisons vary depending on the metric, time of year, and soil moisture regime. Therefore, consideration of both the seasonality and soil moisture regime is needed to accurately determine model biases. These NOAA soil moisture estimates are used for a variety of forecasting and societal applications, and understanding their differences provides important context for their applications and can lead to model improvements. Significance Statement Soil moisture is an essential variable coupling the land surface to the atmosphere. Accurate estimates of soil moisture are important for forecasting near-surface temperature and moisture, predicting where clouds will form, and assessing drought and fire risks. There are multiple estimates of soil moisture available, and in this study, we compare soil moisture estimates from two different National Oceanic and Atmospheric Administration (NOAA) models to in situ observations. These comparisons include both soil moisture amount and variability and are conducted at several soil depths, in different soil moisture regimes, and for different seasons and years. This comprehensive assessment allows for an accurate assessment of biases within these models that would be missed when conducting analyses more broadly.