Project Accomplishments: Probable Maximum Precipitation Estimation Modernization

Abstract

A set of Surface Radiation Budget Network (SURFRAD) measurements across the lower 48 United States has allowed a closer inspection of weather model representations of downward shortwave radiation in the last several years. In this study, it is found that downward shortwave radiation (SW↓) is excessive for the NOAA 3-km HRRR model at each of the 14 SURFRAD stations distributed across the lower United States when averaged over 2-month periods. Possible causes for this station-consistent SW↓ bias error were hypothesized. Three were eliminated by this study and two were then evaluated in this study. We found that this error was not from clear-sky errors but from insufficient attenuation by clouds. It was also found that this cloud deficiency was partly caused by a dry bias in atmospheric water vapor initial conditions. New experiments using the hourly cycled HRRR model–assimilation system were designed and carried out for three seasons with modified data assimilation addressing the dry bias problem and reduction of effective radius for cloud water droplets for both explicit and subgrid-scale clouds. The assimilation and cloud optical parameter changes contributed similarly toward a combined reduced SW↓ radiation bias by 80% in the fall season and 84% in the winter season but by only 35% in the summer season. Even with the improved data assimilation, a dry bias contributing to deficient clouds continues, which is a topic to be explored in a following study.

Significance Statement

Weather forecasts of all durations are dependent on accurate forecasts of clouds. Even the well-known 3-km NOAA HRRR model was found to have errors in clouds, resulting in forecasts of too-warm near-surface temperatures and too little precipitation. In our study including model experiments in three different seasons, we found two key ideas that can improve future storm forecasts: better use of observations to start the weather models to avoid initial dryness and to brighten model cloud forecasts by assuming that cloud droplets are slightly smaller than previously prescribed. These changes can improve NOAA forecasts for aviation, energy, and severe weather in successors to the current HRRR weather model.

Abstract

Extreme precipitation is a highly impactful phenomena, the observation and prediction of which underpin many pressing societal planning and response needs. Rigorous, confident understanding of extreme precipitation and its risks, hazards, and potential impacts (e.g., floods, landslides) requires precipitation data with high resolution and a sufficiently long period of record. Many applications also require low-latency data with global coverage. We examine six precipitation datasets to characterize the representation of extreme precipitation over the continental United States. The Analysis of Record for Calibration (AORC), version 1.1, Stage IV, and CONUS404 provide 4-km hourly precipitation estimates over the CONUS only, while Integrated Multi-satellitE Retrievals for GPM (IMERG) V07 (both early and final) and Multisource Weighted-Ensemble Precipitation, version 2.8 (MSWEP), are 10-km resolution global datasets with 30-min and 3-h time steps, respectively. The global precipitation products struggle to capture extreme precipitation, especially that produced by warm-season convection; IMERG also exhibited spurious events with much higher precipitation than any other dataset, often during West Coast atmospheric rivers. Stage IV and AORC were generally similar to one another, although Stage IV also revealed artifacts in the western United States due to poor radar coverage and missing hourly data. CONUS404 is the only nonobservational dataset included in this study, and while it seems to capture the climatology of extreme precipitation over the CONUS fairly well, it does not reproduce the observed spatial representation of many individual events. This work provides an important foundation from which to further contextualize the representation of extreme precipitation in contemporary, frequently used datasets.

Significance Statement

Many U.S. agencies are producing or improving products related to extreme precipitation and its impacts, which require precipitation information with high spatiotemporal resolution and long periods of record. This study examines the representation of extreme precipitation in six datasets meeting these criteria, including the climatological spatial distribution of precipitation extremes, upper quantiles of the precipitation distribution, and case studies. This work provides a critical foundation for a better understanding of our confidence in extreme precipitation estimates as functions of region, season, and/or storm type. Improving how extreme precipitation estimates are understood and used by scientists and stakeholders translates to better forecasts, preparation, and resilience.

Abstract

The High-Resolution Ensemble Forecast (HREF) system is the first operational convection-allowing ensemble in the United States. Its membership consists of five deterministic convection-allowing models, several of which are used in operational forecasting, plus five time-lagged forecasts. HREF forecasts of summertime (JJA) extreme precipitation, defined as exceeding average recurrence interval (ARI) thresholds at 2, 10, and 50 years at 1-, 6-, and 24-h accumulation periods, are evaluated over 3 years (2021–23) with a consistent model configuration. This study evaluates the representation of precipitation extremes by the ensemble, ensemble forecast performance, and the contribution of the individual members to forecasts of extreme precipitation both CONUS-wide as well as over the six different regions. Both observed and predicted ARI exceedances were found to most frequently occur over the western U.S. Forecasts of 6- and 24-h accumulations were found to be reliable for forecast probability < 30% within a 25-km radius of an observed ARI exceedance. Results for forecast probability> 50% were often noisy due to the small sample size of highly predictable extreme events. Forecast skill from both the ensemble and its individual members generally increased with increasing accumulation period and decreased with increasing ARI threshold. Time-lagged members were less likely to contribute to forecasts of extreme precipitation, having lower probability of detection and success ratio than the nonlagged members, though the lag–High-Resolution Rapid Refresh (HRRR) was often an exception. These results provide important, and heretofore lacking, contextual information to forecasters using the HREF and its individual members.

Significance Statement

The High-Resolution Ensemble Forecast (HREF) system has been running operationally in a constant configuration for 3 years. Despite this, little evaluation has been done on the ensemble and the contribution of its individual members, particularly with regard to the prediction of extremes. Here, we examine forecasts of summertime extreme precipitation from the HREF both CONUS-wide and regionally, including the model climatology of heavy precipitation, ensemble forecasts, and the contribution of individual members to forecasts of extremes. In this way, we provide contextual information that can be used by forecasters using both the HREF and the deterministic models that comprise it in cases of extreme precipitation and by developers working on the next generation of convection-allowing models.

Abstract

Probabilistic forecasts of excessive rainfall based on the fraction of High-Resolution Ensemble Forecast (HREF) members predicting precipitation above a given threshold are used widely in predicting excessive rainfall; however, there is not yet a published study evaluating the skill of these forecasts. In this study, we document the performance of these forecasts over a 3-yr period, including regional and seasonal variations in skill. We find that there is considerable sensitivity to how excessive rainfall events are defined, especially in regions with large differences in the number of excessive rainfall events between different datasets. When verifying against Stage IV exceedances of flash flood guidance (FFG), both the 0000 and 1200 UTC HREF probabilities of exceeding 6-h FFG thresholds exhibit a higher Brier skill score (BSS) than the operational 0900 UTC day-one excessive rainfall outlook (ERO) in five of eight regions in the contiguous United States (CONUS), while probabilities of exceeding fixed 6- or 12-h precipitation thresholds provide a higher BSS than the ERO in another two regions. There is regional variability in the thresholds providing the highest BSS, with FFG (or 75% of FFG) generally providing the best forecasts in the eastern United States, but fixed thresholds providing the best forecasts in the western United States. Only in the southeastern United States are threshold-based HREF forecasts unable to beat the ERO. The 1200 UTC HREF-based forecasts using regionally optimal thresholds beat the ERO by 25%–30% in terms of BSS. Our results suggest that HREF probabilities of exceeding precipitation thresholds have considerable value for excessive rainfall prediction.

Significance Statement

Predicting excessive rainfall and flash flooding is a challenging problem. Operational forecasters often use the fraction of high-resolution weather models predicting rainfall above a given threshold as one tool to guide their excessive rainfall outlooks, but it is unclear which threshold they should use or how much skill the resulting probabilities have. Here, we evaluate these probabilities against observed excessive rainfall events and compare them to operational forecasts. If we select the best-performing precipitation threshold in each region, we find that the model-based probabilities are more skillful than operational forecasts in seven of eight regions of the contiguous United States. These results inform forecasters about the best thresholds to use when developing their excessive rainfall outlooks.

Abstract

In this study, we implement the capabilities to assimilate GOES-R Geostationary Lightning Mapper flash extent density (FED) data within the Joint Effort for Data Assimilation Integration (JEDI) system, coupled with the Finite-Volume Cubed-Sphere (FV3) dynamical core for forecasting. We evaluate different data assimilation (DA) methods, including the local ensemble transform Kalman filter (LETKF), gain-form LETKF (LGETKF), and ensemble 3DVAR (En3DVar), for a test case with active convection over the United States. The convergence behavior of En3DVar is consistent with expectations. Sensitivity to the vertical localization strategies in the algorithms is examined. LGETKF applies gain-form vertical localization, which demands more computational resources than LETKF and En3DVar when using smaller vertical localization radii [e.g., 0.2 or 0.4, compared to larger radii like 1 or 4 in ln(p/p0) space]. While En3DVar achieves a better balance between accuracy and efficiency, it demands significantly more memory than LETKF and LGETKF, with the current JEDI implementation at least. Sensitivity experiments indicate that larger vertical localization radii [e.g., 4 in ln(p/p0) space] improve analysis and 6-h forecast after DA when verified against the observed reflectivity field. Overall, all three DA methods produce comparable results, outperforming the experiment that does not assimilate any data. This work serves to establish the credibility of the lightning DA implementation within the new JEDI system and to understand the effects of algorithm differences related to vertical covariance localization on the assimilation of FED data, whose observation operator involves column integration of a hydrometeor state variable.

Significance Statement

This work implemented capabilities to assimilate high-frequency lightning mapper data from the GOES-R geostationary weather satellites, within the next-generation data assimilation system called the Joint Effort for Data assimilation Integration, which will be used by all operational weather prediction systems of the U.S. National Weather Service. A lightning flash extent density (FED) observation operator with its tangent linear and adjoint components was implemented in the Joint Effort for Data Assimilation Integration’s (JEDI’s) Unified Forward Operator, making FED assimilation available to both ensemble Kalman filter and variational algorithms within JEDI. The new capability is tested with three data assimilation algorithms: the local ensemble transform Kalman filter (LETKF), gain-form LETKF, and ensemble 3DVAR to verify correctness across the algorithms and examine their different vertical localization treatments. Using a convective storm case, we confirm the correctness of the implementation and demonstrate potential positive impacts of assimilating the lightning data on convective storm forecasts, while noting that further multicase studies are needed to generalize the findings.

Abstract

The remnants of Hurricane Ida caused major damage and death in the United States on September 1st, 2021, and 11 people drowned in flooded basement apartments within New York City (NYC). It was catastrophic because the maximum hourly precipitation intensity, recorded as 3.47 inches (88.1 mm) per hour at Central Park, was unprecedentedly high for the NYC region. The stormwater infrastructure in NYC is built for 1.75 inches (44.5 mm) per hour, and so understanding the dynamic risk associated with Ida can inform city planning efforts for climate change’s impact on short duration extreme precipitation events. We contextualize this storm’s record-breaking hourly intensity within the historical record as well as project its risk in the near- to medium-term future using nonstationary stochastic models. These models are conditioned on average temperature (Tavg) and cooling degree day (CDD) projections from three climate models as a covariate, each with a SSP 126 and SSP 370 scenario. The likelihood of such a storm was slowly increasing even before Ida happened, but the projected aggregate reoccurrence risk of an event of Ida’s magnitude over time from the non-stationary models ranges from 4 to 52 times higher than the risk given by the stationary model. Using CDD as a covariate resulted in risks that were more than twice the magnitude than when using Tavg. Presenting both covariates provides a broader envelope of uncertainty, which highlights the importance and nuances in the choice of a regionally appropriate covariate for non-stationary risk analysis.

Abstract

Extreme precipitation events that occur during the warm season are commonly associated with high-intensity, short-duration rainfall that can lead to damaging flash-flood-related impacts. The NCEP High-Resolution Ensemble Forecast (HREF) system is an ensemble of convection-allowing models (CAMs) that provides both high-resolution precipitation guidance and forecast uncertainty information for anticipating these kinds of events. In this study, the probability-matched mean (PMM) and localized PMM (LPMM) from the HREF version 3, initialized at 0000 UTC, are evaluated for their ability to capture heavy and extreme precipitation in the warm-season months of June–August across the contiguous United States. Thresholds defined with the 2-, 10-, 50-, and 100-yr average recurrence interval (ARI) exceedances for 1-, 3-, and 6-h rainfall durations are assessed between forecast hours f = 12–36 to capture a complete convective cycle. The forecasts are evaluated against the Multi-Radar Multi-Sensor (MRMS) system quantitative precipitation estimation (QPE) pass 2 between 2021 and 2023 when the HREF membership remained unchanged. It was found that the HREF PMM is more skillful than the LPMM for almost all ARI thresholds and rainfall durations. The tendency for the PMM to overestimate ARI exceedances leads to a higher probability of detection (POD) and higher critical success index compared to the LPMM. In contrast, the HREF LPMM underestimates all ARI thresholds and durations in both the eastern and western United States. This is characterized by lower frequency bias and lower POD but a higher success ratio (i.e., lower false alarm rate) relative to the PMM.

Significance Statement

The High-Resolution Ensemble Forecast (HREF) system has been running operationally in a constant configuration for 3 years. Despite this, little evaluation has been done that quantifies how well the ensemble predicts extreme precipitation, particularly with regard to the HREF means. Here, we examine forecasts of summertime extreme precipitation from the HREF means, which include the arithmetic mean, probability-matched mean, and localized probability-matched mean. A climatology of each mean’s heavy and extreme precipitation and forecast performance are presented. The results from this study provide contextual information that can be used by forecasters using the HREF means for predicting extreme precipitation and by researchers for investigating how to produce the most skillful, ensemble-based QPF products.

Abstract

When extreme hydrometeorological events threaten the United States, quantitative precipitation forecasts (QPFs) from the National Centers for Environmental Prediction’s Weather Prediction Center (WPC) provide critical decision support to help mitigate risks to life and property. This study builds on previous work by providing an updated benchmark of WPC QPF skill for extreme precipitation events from 2012 to 2024, through evaluating the 6- and 24-h accumulations for the day 1–3 lead times. Extreme precipitation is defined using three thresholds: the 99th and 99.9th percentile precipitation values of all wet-site days from 2012 to 2024 for each National Oceanic and Atmospheric Administration (NOAA) River Forecast Center (RFC) region and the 2-yr average recurrence interval from the NOAA Atlas 14. WPC forecasts are verified against Stage IV quantitative precipitation estimates. Forecast skill is assessed seasonally, regionally, and temporally for the 1200 UTC forecast cycle. Results show that WPC forecasts of extreme precipitation have improved over time. The highest skill is observed along the West Coast, where events are dominated by atmospheric rivers from late fall through early spring. Forecasts in these regions often retain skill through day 3. The East Coast shows the second-highest skill, particularly in fall and winter, when extreme precipitation occurs from tropical and synoptic systems. Skill is lowest in inland areas, especially in the summer months when precipitation systems are weakly forced, smaller in scale, and often underforecasted by WPC. These events rarely show skill after a day 1 lead time.

Significance Statement

Extreme precipitation remains a primary driver of billion-dollar disasters in the United States, underscoring the growing need for accurate quantitative precipitation forecasts. By benchmarking the skill of National Oceanic and Atmospheric Administration’s (NOAA) Weather Prediction Center forecasts of extreme precipitation, we can understand where regional and seasonal weaknesses exist and evaluate the current boundaries of precipitation predictability. These insights are essential for guiding future advancements in operational forecasting and model development.

Abstract

Global warming is increasing the frequency of extreme precipitation events, including those referred to as having a return period of 1000 years. Yet frequency estimates of these high-end extremes are plagued by large uncertainties that result from the short record of weather station observations used to assess their historical context. To more robustly assess modern extreme precipitation events, we develop a novel approach that lengthens the data record by blending historical observations from NOAA’s Applied Climate Information System with Community Earth System Model simulations run from 850 to 2100 CE. We apply this method as a proof of concept to the record-breaking rainfall event in the central United States in July 2022. We find that the return period for this storm’s 24-h rainfall is ∼530 years (90% confidence interval: 370–700 years) for the greater St. Louis region and ∼280 years (90% confidence interval: 115–340 years) for eastern Kentucky. Moreover, the rainfall amount from this event is ∼2–4 times more likely to occur in the future relative to the preindustrial era. Compared to previous best practices, this study’s approach more precisely evaluates extreme precipitation magnitude and frequency, demonstrating potentially large reductions in the uncertainty associated with classifying a 1-in-1000-yr rainfall event. This blended data approach can be used to update previous methods used for assessing modern extreme precipitation events and to better prepare society and critical infrastructure for the present and future risks posed by precipitation extremes.

Significance Statement

We present a new method aimed at improving estimates of the magnitude and frequency of extreme precipitation. Our method lengthens the time coverage of precipitation data by combining weather station observations with climate model simulations spanning 850–2100 CE. We apply this methodology to a record-breaking rainfall event that occurred in July 2022 in the central United States. Our approach provides a more precise estimate of this event’s frequency and demonstrates that the rainfall amount from this event is ∼2–4 times more likely to occur in the future relative to the preceding millennium. Our findings can be used to better prepare society and infrastructure for the present and future risks posed by extreme precipitation.