1. Introduction
The summer North American monsoon (NAM) is responsible for frequent intense convection along the Mexican cordillera and over the mountain ranges of the southwestern United States (Douglas et al. 1993; Dunn and Horel 1994a,b; Maddox et al. 1995; Adams and Comrie 1997; Yang et al. 2019; Boos and Pascale 2021). Many studies have examined the conditions leading to extreme precipitation events associated with the NAM in this region and ways to predict their occurrence (Gutzler et al. 2009; Serra et al. 2016; Risanto et al. 2021). Not surprisingly, the fundamental building blocks for intense convection are necessary: lift, instability, and moisture (Doswell et al. 1996). Afternoon surface heating of elevated terrain provides the orographic lift that aids thunderstorm development (Maddox et al. 1979; Schumacher 2017). Mazon et al. (2016) classified extreme NAM weather events primarily in Arizona based on atmospheric instability, precipitable water vapor, and upper-level conditions.
Smith et al. (2019) provide a detailed climatological evaluation of the seasonality and locations of thunderstorms in northern Arizona and southern Utah. Their study was motivated by flash floods in Hildale, Utah, and within Zion National Park on 14 September 2015 that resulted in 20 fatalities, which account for nearly two-thirds of the total flash flood deaths during the past 25 NAM seasons in southwestern Utah. They note on the basis of lightning, radar, and streamflow records that the underlying terrain and synoptic/mesoscale setting modulate the occurrence and track of thunderstorms within the region. They also note that flash flood water volumes in the region’s watersheds are unrelated to basin scale, as is often the case in other regions of the United States where flash flooding may result from widespread heavy precipitation. Rather, the flash flood response is tied to the spatial scale of thunderstorms (1050 km2) and their proximity to the small catchments upstream of slot canyons and channel narrows.
Figure 1 illustrates the seven-county region of southwestern Utah that is the focus of this study. This 64 000-km2 region, larger than 10 U.S. states, encompasses low-lying deserts and scattered forests confined generally to the slopes and high plateaus of the western extent of the Colorado Plateau region. Prominent high-elevation features of the region include from west to east the Pine Valley Mountains and Markagunt, Paunsaugunt, Kaiparowits, and Aquarius Plateaus. The Cedar City, Utah, WSR-88D radar (KICX) is sited at an elevation of 3230 m on the Markagunt Plateau (Fig. 1). Although there are only ∼200 000 permanent residents in the region, over 10 million visits are recorded annually to the national parks, monuments, and recreation areas within the region. Zion National Park alone receives over five million visits, while the other major destinations (Bryce, Capitol Reef, and Glen Canyon) receive between them 13 million visitors annually.
Southwestern Utah with shaded terrain, brown 2000-m elevation contour, and black county outlines and labels. Light gray shading denotes national parks, monuments, and national recreation areas. Flash flood reports during the 2021–23 summer seasons are indicated by black cross symbols with locations of seven flash flood reports on 26 Jul 2021 in orange. The location of the Cedar City WSR-88D (KICX) is shown as well as general areas labeled in orange for the Pine Valley Mountains (Pin) and Markagunt (Mar), Paunsaugunt (Pau), Kaiparowits (Kai), and Aquarius (Aqu) Plateaus.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
Abrupt stepwise drops from the plateaus down to the desert floors are broken up by narrow riverine channels and slot canyons that contribute to the propensity for flash floods in the region. Quantifying the actual number of flash flood events in any region is difficult (Marjerison et al. 2016). The flash flood reports during the past three monsoon seasons (2021–23) in southwestern Utah cluster near the small number of towns and cities located near canyon exits and in the elevation band of the gray and white cliffs (∼1500–2500 m) within the parks, monuments, and recreation areas, hereafter referred to simply as Parks (Fig. 1). As summarized in Fig. 2, 745 flash floods in southwestern Utah have been reported to the National Center for Environmental Information (NCEI 2023) during the 1996–2023 period with the highest totals during 2013, 2021, and 2022. Of the 745 floods, 86% of them have been reported from May to September due to convective storms with 33 deaths and over $40 million in property damage.
Flash flood reports in the seven-county region of southwestern Utah during each year (NCEI 2023).
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
Forecasting flash floods at short lead times (<6 h) often leads National Weather Service (NWS) offices to issue flood warnings (i.e., flood conditions occurring or imminent) that rely on careful evaluation by forecasters of model guidance, interpretation of satellite and radar imagery of convective environments, and situational awareness of flood-prone locales (Gourley et al. 2012; Yussouf and Knopfmeier 2019; Martinaitis et al. 2023). The NWS Forecast Offices in Salt Lake City and Grand Junction issued for Utah the first and third highest numbers of flash flood warnings within the 1996–2023 period during 2021 and 2022, respectively (NCEI 2023).
Providing forecasts for the potential for organized convection and intense rainfall that might lead to flash floods within areas of complex terrain at lead times longer than 6 h requires greater reliance on numerical model output from operational convection-allowing models, such as the High-Resolution Rapid Refresh (HRRR; Sun et al. 2014; Blaylock and Horel 2020; Dowell et al. 2022; Grim et al. 2024). The NWS issues flood watches for such lead times. However, in order to avoid too many watches during the monsoon season, the Salt Lake City NWS Forecast Office (SLC WFO) issues flood watches in coordination with other WFOs typically on days when multiple basins in southern Utah are likely to be impacted. In addition, they issue forecasts of flash flood potential for the current and next 2 days to eleven government entities responsible for public safety across Utah. Flash flood potential forecasts are provided in southwestern Utah for the regions highlighted in Fig. 1: Bryce National Park, Capitol Reef National Park, Glen Canyon National Recreation Area, Grand Staircase-Escalante National Monument, and Zion National Park.
This study focuses on the meteorological conditions in southwestern Utah during the 2021–23 summer monsoon seasons associated with high-intensity precipitation episodes that might lead to flash flooding. Of course, the occurrence of flash floods in this region and elsewhere depends on the complex interplay between terrain, soil, and hydrologic factors (Smith et al. 2019; Hill and Schumacher 2021). We examine conditions between 15 June and 15 September when high risks exist in vulnerable locations for recreational injuries and fatalities, general public safety, and property damage. In addition, the availability of the Cedar City NWS radar within this region provides a better estimation of convective activity and rainfall than is available for other parts of the state influenced by the NAM.
Large-scale conditions analyzed by the HRRR are summarized in this study for widespread and intense rainfall events during the three summers in southwestern Utah. Day-to-day variability is contrasted during the three summers for precipitation, flash flood reports, moisture availability, and instability. Conditions during the afternoon of 26 July 2021 are presented when rainfall in excess of 4 cm fell in Cedar City, Utah, and flash floods were recorded at six other locations within the region (Fig. 1).
Motivated by the lead time needed by Salt Lake City NWS forecasters to consider numerical guidance prior to issuing their first flash flood potential forecasts for the current day, we evaluate the utility of HRRR time-lagged ensemble (TLE) forecasts of precipitable water (PWAT) and maximum convective available potential energy (CAPE) issued 13–18 h prior to the afternoon period when convection is initiated across the region [1800–2100 UTC; 1200–1500 mountain daylight time (MDT)]. Our approach is to test whether forecasts of high PWAT and CAPE averaged over the entirety of southwestern Utah may identify days likely to experience unusually high rainfall amounts across the region. We rely on hourly quantitative precipitation estimates (QPEs) from the Multi-Radar Multi-Sensor (MRMS) system (Martinaitis et al. 2021) to validate the HRRR model guidance. Of course, there has been no expectation from the outset of this study that HRRR forecasts at these lead times will provide an indication of the specific regions where intense rainfall may fall or flash floods may occur in specific catchments.
This research builds upon the study completed by Powell (2023) that examined the active summer monsoon seasons during 2021 and 2022. The results from that study have been expanded upon by examining conditions during the weaker 2023 monsoon season. Since prior studies and forecaster experience highlight that moisture availability and instability are dominant factors for intense NAM precipitation (Mazon et al. 2016; Smith et al. 2019; Yang et al. 2019; Yu et al. 2023), we hypothesized that HRRR TLE ensemble forecasts of high PWAT and CAPE would be related to widespread rainfall in southwestern Utah. That hypothesis is evaluated using statistical approaches including bias correction and random forest classification (Breiman 2001; McGovern et al. 2017; Speiser et al. 2019; Chase et al. 2022). The successful operational implementation of random forest approaches to predict excessive rainfall events on a national scale has been demonstrated by Schumacher et al. (2021), and such approaches also are considered as part of the excessive rainfall outlooks of the Weather Prediction Center (Burke et al. 2023). In addition, staff at the SLC WFO tested random forest methods for flash flood predictions across southern Utah (Seaman et al. 2024). We divide the 2021 and 2022 summers into training and validation samples for random forest predictions followed by independent forecasts for summer 2023.
2. Data
a. Precipitation, radar, and lightning observations
Although providing limited coverage across southwestern Utah, precipitation observations from 101 stations from the networks listed in Table 1 were available during all three summers. These networks were selected based on sensor quality and likelihood of routine maintenance. For individual case studies, all available stations measuring rainfall were considered, which typically provided reports from over 50 additional sites within southwestern Utah. Precipitation observations were obtained using accumulated precipitation calculation services developed by Synoptic Data Public Benefit Corporation (PBC).
Number of stations in southwestern Utah from selected networks reporting precipitation during all three summers.
Deep convection within southwestern Utah is generally detected by the Cedar City WSR-88D Doppler radar (KICX). This radar is located 3230 m above sea level near the southwestern edge of the Markagunt Plateau (Fig. 1). Due to its high elevation, 0.2° elevation scans are available to help identify conditions over the surrounding desert regions at elevations typically between 1000 and 2000 m.
Widespread coverage of vigorous convection-containing lightning is made possible by the ground-based National Lightning Detection Network (NLDN; Murphy et al. 2021). The flash energy density (FED) product provided by the MRMS system that is based on the NLDN is used in this study. FED is an estimate of the number of cloud-to-ground flashes per km2 during each 30-min period.
b. Radar multisensor quantitative precipitation estimates
The MRMS is an operational system of the National Centers for Environmental Prediction for estimating precipitation from radar, rain gauge, satellite, lightning, and numerical weather model data (Zhang et al. 2016; ElSaadani et al. 2018; Sharif et al. 2020; Zhang et al. 2020; Martinaitis et al. 2021). The MRMS QPE Pass II products for each hour and for each square kilometer along with FED and other diagnostic fields were accessed from Iowa State University’s Iowa Environmental Mesonet archive (Iowa State University 2023).
Radar precipitation estimates are the primary input for MRMS QPE in southwestern Utah. As described by Zhang et al. (2020) and Martinaitis et al. (2021), MRMS processing expands beyond single-radar dual-polarization QPE to include quality control (e.g., accounting for beam blockage and nonprecipitation echoes) and a radar quality index for each grid cell. If an area has adequate radar quality, radar estimates may be adjusted using weighted corrections based on precipitation gauge data to then yield the final QPE. Gauges from many of the networks listed in Table 1 are not incorporated into the MRMS QPE and hence are helpful to evaluate the accuracy of the QPE estimates. If the radar quality is poor during convective situations, precipitation estimates may be adjusted based on gridded estimates of precipitation from the HRRR model.
c. High-Resolution Rapid Refresh model
The HRRR is a convection-allowing, short-range forecast model with 3-km horizontal grid spacing run operationally every hour by the National Centers for Environmental Prediction for the CONUS region (Dowell et al. 2022; Grim et al. 2024). Forecasts at lead times out to 18 h are available every hour with lead times extended out to 48 h at 0000, 0600, 1200, and 1800 UTC. The HRRR benefits from advanced data assimilation techniques incorporating standard data observations (rawinsonde, aircraft, GPS precipitable water, etc.) and also includes 3D radar reflectivity data from the MRMS and lightning data from the NLDN (Hu et al. 2017; James and Benjamin 2017; Dowell et al. 2022). Access to the high-resolution forecast model output in grib2 format is currently available through Amazon Web Services and Google’s Cloud Platform (Dowell et al. 2022). This study relies on HRRR version 4 that was deployed on 2 December 2020. Model analysis (F00) and forecast (F01–F18) fields are retrieved in Zarr format from Amazon Web Services (Gowan et al. 2022). Supported by Amazon’s Sustainability Data Initiative, the Zarr files are created in order to split the CONUS into 96 compressed chunks, allowing more efficient downloading for smaller domains (Gowan et al. 2022).
HRRR model F00 analyses and F12–F18 forecasts of deep moisture (PWAT) and instability (surface-based CAPE) are relied upon extensively in this study as a means to assess conditions favorable for widespread convection across southwestern Utah. Mazon et al. (2016), Yang et al. (2019), and Yu et al. (2023) used similar metrics to study the NAM in Arizona and Nevada. Surface-based CAPE was found to be an adequate metric for estimating the potential for convective instability over elevated terrain due to the limited convective inhibition likely in these locations during summer afternoons.
HRRR QPE at 6–18-h lead times was compared to MRMS QPE analyses and NLDN FED as a means to assess the extent to which those forecasts might be useful for situational awareness for the likelihood of widespread precipitation across southwestern Utah. However, HRRR QPE forecasts substantially underestimated observed and analyzed precipitation in southwestern Utah. Hence, we do not rely on HRRR QPE forecasts in this study.
3. Results
a. 26 July 2021
Substantial property damage due to flash flooding occurred on 26 July 2021 near Cedar City, Utah. Seven flash flood reports were made on this day (Fig. 1). PWAT and CAPE analyzed by the HRRR during this afternoon from 1800 UTC 26 July to 0000 UTC 27 July (1200–1800 MDT) are shown in Fig. 3. The northerly moisture transport into Arizona, Nevada, and Utah was extensive (Fig. 3a) with local CAPE values in excess of 2000 J kg−1 across southwestern Utah (Fig. 3b). The areal- and daily averaged PWAT in this case was 2.7 cm, and the areal-averaged maximum CAPE was 900 J kg−1.
(a) Mean hourly HRRR PWAT (cm) during the period 1200–1800 MDT 26 Jul 2021. (b) Maximum hourly HRRR CAPE (J kg−1) during the same period.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
The prevailing flow during this afternoon was deep southeasterly winds from 700 to 250 hPa (not shown). Animations of GOES-West satellite imagery and KICX composite radar reflectivity during this afternoon confirmed the prevailing southeast-to-northwest progression of thunderstorm cells during the afternoon as shown by the composite radar reflectivity during the afternoon (Fig. 4a). Figure 4b shows the average hourly FED during 1200–1800 MDT with many areas exhibiting extensive lightning across southwestern Utah including frequent lightning near Cedar City and the other locations reporting flash flooding that afternoon.
(a) Maximum composite reflectivity (dBZ) from the KICX radar during 1300–1400 MDT 26 Jul 2021. (b) Average NLDN FED (strikes per square kilometer per hour) during the period 1200–1800 MDT 26 Jul 2021. Black dots show the locations of flash flood reports that day. Heavy solid lines denote terrain elevation at 2000 m (brown) and 3000 m (black). (c) Average hourly precipitation (cm h−1) during the period 1200–1800 MDT 26 Jul 2021. The locations of the flash flood reports are indicated by orange x values. (d) As in (c), but from MRMS analyses.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
Average hourly precipitation rates during the afternoon from 136 stations and MRMS QPE analyses during the afternoon are shown in Figs. 4c and 4d, respectively. While modest amounts of rainfall were observed across many areas of southwestern Utah during this afternoon, excessive amounts were highly localized, most notably reports of over 7 and 4 cm during the 6-h period at the Utah Climate Center stations located in the Wah Wah Range and Cedar City, respectively (denoted by the two red circles in Fig. 4c). While no reports of flooding are available from the remote Wah Wah site near the northwestern edge of the domain, reports of flooding across the interstate southwest of Cedar City were associated with a strong cell (maximum composite reflectivity > ∼65 dBZ) from 1400 to 1500 MDT (not shown). Much of this precipitation was channeled directly into the residential and commercial areas of Cedar City, a city with 37 000 residents. The MRMS precipitation analyses for this afternoon capture many of the cells associated with the flash flood reports (Fig. 4d) as well as near Wah Wah, even though beam blockage affected radar returns in that area. The latter was likely due to the use of lightning data by the MRMS since that Utah Climate Center station was not assimilated into the MRMS.
Seasonal summaries
Figure 5a highlights the average daily precipitation (mm) during the 2021–23 summer seasons (15 June–15 September) from 675 stations from the networks listed in Table 1 that are available in Utah and adjacent areas. While very helpful, these observations are clearly insufficient to capture the strong gradients in rainfall that take place across the region from desert lowlands to Utah’s high-elevation mountain ranges. High daily averages (2.5–3 mm) during these three summers among the 101 stations within southwestern Utah are found on the Markagunt Plateau near the KICX radar.
(a) Observed average daily precipitation (mm) shaded according to the scale at the bottom for stations in the networks listed in Table 1 during the 2021–23 summer seasons from 15 Jun to 15 Sep. KICX WSR-88D radar location is denoted by the black circle, and the brown contours denote 2000-m elevation. The rectangle encloses the southwest Utah region. (b) As in (a), but for MRMS-analyzed values of average daily precipitation.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
The spatial variability in the three-summer average of MRMS daily rainfall is shown in Fig. 5b. Mazon et al. (2016) highlighted the climatologically lower rainfall totals in the NAM’s extension into southern Utah compared to those in northern Arizona. As expected, all of the mountain ranges and plateaus of southwestern Utah exhibit higher MRMS QPE than their surrounding deserts (e.g., from west to east the Pine Valley Mountains and Markagunt, Paunsaugunt, Kairparowits, and Aquarius Plateaus). The 50th–75th MRMS QPE percentiles within 25 km of KICX on the Markagunt Plateau lie between 2.5 and 3 mm day−1, which is consistent with the observations in that region. However, some of the isolated ranges in central and eastern Utah (e.g., Henry and Abajo Mountains) far from KICX and the Grand Junction, WY, radar (KGJX) have higher gauge totals than MRMS QPE. Based on prior research (e.g., Herman and Schumacher 2018), feedback from NWS forecasters and comparing MRMS QPE to station observations, NLDN FED, and HRRR short-range forecasts during the 2021 and 2022 summer seasons, Powell (2023) concluded that the lack of radar coverage for such mountain ranges likely impacts the MRMS QPE estimates in the peripheries of the southwestern Utah domain.
As documented previously by Smith et al. (2019), deep convection in the region initiates around solar noon over the high plateaus and mountains and, depending on the prevailing flow, persists there or propagates away later during the afternoon and evening. Figure 6a illustrates the initial development between 1200 and 1500 MDT (1800–2100 UTC). For the region as a whole, the highest rainfall totals occur between 1500 and 1800 MDT (Fig. 6b) and then diminish during the evening (Fig. 6c). Maxima across the region of average rainfall totals during the other 3 h periods are below 0.4 mm day−1 (not shown).
MRMS 3-h QPE totals (mm) averaged over the three summers for (a) 1200–1500, (b) 1500–1800, and (c) 1800–2100 MDT. Brown contours denote 2000-m elevation.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
The accumulation of flash flood reports during the three summer seasons is presented in Fig. 7a. As mentioned earlier, the number of flash flood reports in the region depends on observers being present as well as favorable meteorological conditions leading to intense convection and hydrologic factors near specific vulnerable locales. The total number of flash flood reports was highest during 2021 and fewest in 2023 (see also Fig. 2). Intense convection leading to flash floods began in earnest during mid-July 2021, while the number of flash floods reported increased sharply during the mid-August 2022 monsoon period. While flash floods were reported on similar numbers of days during 2021 and 2022, they were reported on only 10 days during summer 2023.
(a) Accumulated flash flood reports during each summer season. (b) Accumulated daily precipitation (cm) averaged over southwestern Utah during each summer. (c) Accumulated mean daily flash flood potential rating forecasts issued by the SLC WFO for southwestern Utah.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
The accumulation during each season of daily MRMS QPE averaged over the entire southwestern Utah domain is shown in Fig. 7b. Although only half as many flash floods were observed during summer 2022 compared to 2021, the accumulated precipitation across the domain was higher during that summer. In addition, even though the 2023 NAM had a delayed start until mid-July, the accumulated seasonal rainfall was comparable to that observed in 2021. Less-intense precipitation likely fell in regularly monitored flood-prone regions during summer 2023 compared to that during the other 2 years.
As mentioned earlier, Salt Lake City NWS forecasters have extensive experience issuing watches and warnings for flash flood conditions across southwestern Utah on the basis of many data and numerical model resources. Based on their experience and recommendations from constituents, the flash flood potential ranking (FFPR) was developed over 15 years ago initially for Zion National Park and is now issued by the SLC WFO to benefit government agencies responsible for public safety for the five parks indicated in Fig. 1. FFPR values are first issued for the current and next 2 days typically in the early morning (0800–1000 UTC; 0200–0400 MDT) and then updated as necessary throughout the day. Figure 7c summarizes the issuance of the initial FFPRs. While park staff prefer that the FFPR be defined by text descriptors, Salt Lake City forecasters have assigned loose probabilities of flash flood occurrences to them as follows: Not Expected (12.5%), Possible (37.5%), Probable (62.5%), and Expected (87.5%). The daily values accumulated in Fig. 7c are the average of those probability values (0–1) computed from the five park forecast samples.
The FFPR trends during each summer follow the trends in areal-averaged rainfall (Fig. 7b) rather than flash flood reports (Fig. 7a); e.g., the highest risk overall for flash flooding was predicted to be during summer 2022. In addition, specific monsoonal periods with larger average rainfall also were periods when FFPR forecasts were higher (e.g., late July and early September 2023).
b. Forecasting widespread rainfall
As mentioned in the introduction, an objective of the study by Powell (2023) and this work was to test whether HRRR model guidance at lead times between 13 and 18 h would be useful for predicting afternoons most likely to have widespread rainfall across southwestern Utah. Our TLE approach shown schematically in Fig. 8 builds on the capability to use hourly updating HRRR forecasts and prior work that has relied on HRRR TLEs (e.g., Thompson et al. 2017; Xu et al. 2019; Gowan et al. 2022). The sample of initialization times we use (0300–0500 UTC) is motivated by the lead time needed by Salt Lake City NWS forecasters to consider numerical guidance prior to issuing overnight their first FFPR forecasts for that afternoon. As shown in Fig. 8, a sample of 12 PWAT forecasts and maximum CAPE are available 13–18 h prior to the afternoon period when convection is initiating across the region (1800–2100 UTC; 1200–1500 MDT). Hence, to gain additional lead time given that HRRR forecasts are available each hour only at lead times up to 18 h, we focus on the conditions when convection is starting (Fig. 6a), rather than when the convection and rainfall reach their peak (Fig. 6b). We test whether the ensemble average forecasts of high PWAT and CAPE averaged over the entirety of southwestern Utah may identify days likely to experience unusually high rainfall amounts across the region. As might be expected, the ensembles tend to be under-dispersive such that the domain averages for the individual ensemble members are not far from the ensemble mean.
Twelve-member TLE forecasts for the early afternoon period during which convection typically starts in southwestern Utah are generated from HRRR forecasts initialized from 2100 to 2300 UTC the previous evening.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
The seasonal trends in ensemble mean PWAT analyses averaged over southwestern Utah during early afternoon exhibit relatively monotonic increases of 0.75–1.0 cm day−1 with higher totals during summer 2022 than the other two summers (Fig. 9a). The average PWAT of the TLE forecasts valid during those same afternoon periods (dashed lines) is slightly lower than analyzed during all three summers, which reflects a very small dry bias of HRRR forecasts. Similar to the seasonal evolutions of rainfall (Fig. 7b), the seasonal accumulation of maximum CAPE based on early afternoon HRRR analyses exhibits periods when the NAM is more active in this region than in other periods (Fig. 9b). A higher CAPE is also evident during summer 2022 than during the other two summers. It should be noted that the analyzed CAPE values shown in this study over southwestern Utah may underestimate actual conditions. Prior work with earlier versions of the HRRR found the analyses had a stable bias compared to estimates from rawinsondes (Evans et al. 2018). Of greater importance is the clear bias of the CAPE forecasts to be more stable than analyzed (Evans et al. 2018; MacDonald and Nowotarski 2023). This deficiency also likely contributes to dry biases of HRRR model precipitation forecasts (Dougherty et al. 2021; Powell 2023).
(a) HRRR PWAT (cm) analyses during 1800–2100 UTC (1200–1500 MDT) averaged over southwestern Utah and accumulated during each summer season (solid lines). HRRR TLE PWAT forecasts valid at the same time (dashed lines). (b) As in (a), but for maximum hourly CAPE (J kg−1).
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
Following the approach used by Mazon et al. (2016), Yang et al. (2019), and Yu et al. (2023), we relate regional PWAT and CAPE to daily mean QPE in Fig. 10 during all three summers. The large-scale CAPE values in Fig. 10a are domain averages of the maximum CAPE at every grid point during the 1200–1500 MDT period, while the PWAT values are domain-averaged PWAT during that time period. The size and color of each dot denote the accumulated precipitation for that entire day, which predominantly falls between 1200 and 2100 MDT (Fig. 6).
(a) HRRR analyses of maximum PWAT (cm) vs CAPE (J kg−1) during 1800–2100 UTC (1200–1500 MDT) each day of the three summers averaged over southwestern Utah. The color and size of each circle denote daily MRMS precipitation (cm). Cross symbols denote days with at least one flash flood report. The solid lines highlight the mean PWAT and CAPE. (b) As in (a), but for HRRR TLE ensemble mean forecasts of PWAT vs CAPE initialized between 0300 and 0500 UTC and valid at 1800–2100 UTC.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
As shown in Fig. 10a, MRMS QPE amounts tend to be low if either PWAT averaged over southwestern Utah is less than the 3-yr mean in PWAT (1.7 cm) or maximum CAPE is less than its 3-yr mean (390 J kg−1); i.e., nearly all the high rainfall amounts are located in the upper-right quadrant of Fig. 10a. Cross symbols in Fig. 10a denote the 63 days when at least one NCEI flash flood report was recorded. Nearly all of those days lie as well in the upper quadrant of Fig. 10a (above-normal PWAT and CAPE). Smith et al. (2019) also found that lightning activity increased in the region when PWAT was greater than 1.5 cm and grew substantively after 2.0 cm. Hence, above-normal CAPE and PWAT are reasonable estimates for days likely to experience higher rainfall totals across the region and, to some extent, the number of days with flash flood reports.
Figure 10b relates the daily accumulated MRMS QPE to the ensemble mean forecasts of CAPE and PWAT. Comparing Figs. 10a and 10b, the downward shift in the days with high rainfall amounts and flash flood reports results from the large stable bias of HRRR CAPE forecasts evident as well in Fig. 9b without any substantive PWAT bias (Fig. 9a). Table 2 also illustrates the HRRR CAPE forecast bias. The counts of days and average rainfall along the diagonals summarize the cases when HRRR forecast conditions match those analyzed. HRRR CAPE forecasts underestimated analyzed CAPE on 37 days. Rainfall totals for the region on those days were often high.
Days and average daily precipitation (PPT; cm) for HRRR TLE forecasts (columns) relative to HRRR analyses (rows) during all three summers. Forecasts and analyses are subdivided into whether PWAT and CAPE are below (B) or above (A) three-season averages across southwestern Utah. Bold face denotes categorical hits, while italics highlight frequent underpredictions of CAPE.
Accounting for the large CAPE bias is a possible way to utilize HRRR predictions to identify those afternoons that might have enhanced convection. Compensating for both the PWAT and CAPE biases evident during the 2021 and 2022 summers (−0.03 cm and −135 J kg−1), predictions for conditions during summer 2023 are shown in Table 3. The number of underestimates of above-normal CAPE days drops from 34% (37 of 108) to 19% (6 of 31).
As in Table 2, but for the 2023 season only and after correcting for the small PWAT and large negative CAPE biases evident during the 2021 and 2022 seasons.
The predictions in Table 3 address identifying those afternoons during summer 2023 with above-normal CAPE and PWAT that often are associated with higher rainfall amounts across southwestern Utah. An alternative approach is to identify ways to predict those days when daily MRMS rainfall amounts are unusually high (≥0.14 cm), approximately the top 25% of those observed during the 2023 summer. Figure 11a shows the analyzed MRMS rainfall amounts during each day of that summer. Several distinct heavier rainfall periods are evident: 15–16 June; 30 July–3 August; 10–24 August; 31 August–4 September; and 10–13 September. Figures 11b and 11c show the HRRR-analyzed and TLE-forecasted values for afternoon PWAT and CAPE, respectively. The general relationships evident in earlier figures are encapsulated here: TLE PWAT forecasts are nearly identical to those analyzed; TLE CAPE forecasts underestimate CAPE analyses; and periods when PWAT and CAPE are both high often have higher rainfall totals.
(a) Daily MRMS rainfall (cm; bars) averaged over southwestern Utah during summer 2023. Symbols denote TLE forecast hits (blue) and false alarms (orange) for random forest (circles) and bias-corrected (triangles) methods to predict rainfall that exceeds the ∼75th percentile (solid line). (b) PWAT (cm) from HRRR analyses (blue) and TLE forecasts (gray) for the 1800–2100 UTC period. (c) As in (b), but for maximum CAPE (J kg−1). (d) Mean (circles) and spread (dashed lines) of FFPR forecasts issued by the SLC WFO for the five Parks in southwestern Utah. Categorical forecasts made for each park are Not Expected (No), Possible (Poss), Probable (Prob), and Expected (Exp). Days with flash floods reported are denoted by crosses.
Citation: Weather and Forecasting 39, 7; 10.1175/WAF-D-24-0018.1
A predictive scheme patterned loosely on the FFPR is to classify days into three categories based on TLE prior-summer bias-corrected forecasts of PWAT and CAPE: 1) If they are both below normal, then having rainfall in the upper 25th percentile is “Not Expected”; 2) if only one parameter is above normal, then having high rainfall is “Possible”; and 3) if both are above normal, then high rainfall is “Probable.” Table 4 summarizes the results of such a prediction scheme for the 2023 summer. Treating a Probable forecast as a hit and using standard forecast metrics (Wilks 2011; Chase et al. 2022), this approach would yield the following scores: 63% probability of detection (POD); 58% success ratio (SR); and 43% critical success index (CSI). Hits and false alarms are indicated in Fig. 11a by the blue and orange triangles, respectively. Misses are days when the MRMS rainfall is above the solid line (∼75th percentile), and there are no triangle symbols.
Bias-corrected HRRR PWAT and CAPE predictions of daily areal-mean MRMS precipitation exceeding ≥0.14 cm (∼75th percentile) during the 2023 season relative to MRMS daily areal-mean precipitation.
The numerous machine learning studies (e.g., Hill and Schumacher 2021) examining ways to predict extreme rainfall events led us to consider such approaches to overcome the large number of fall alarms and misses evident in Table 4 and Fig. 11a. However, rather than using a large number of predictor variables as is often used, we limited for clarity the predictors simply to PWAT and CAPE TLE mean forecast values. The arbitrary decision to target or label in terms of the relatively rare high rainfall days (MRMS ≥ 0.14 cm) lends itself to binary classification as opposed to predicting precipitation amount each day. The supervised random forest classification method is well suited for this application (Breiman 2001; Speiser et al. 2019; Chase et al. 2022; Hsieh 2023). We subdivided randomly the days during the 2021 and 2022 summers into training and validation datasets (67% and 33% of the samples, respectively), leaving the entire 2023 summer as the test dataset. The sklearn default threshold of 0.50 is used to discriminate between high and nonhigh rainfall days. Optimal hyperparameters were found to vary depending on the relative sizes of training and validation datasets, but the results applied to the 2023 test dataset tended to be invariant to these differences in maximum depth and number of estimators. The accuracy score was 0.93 with the relative feature importance of the two predictors being 53% and 47% for TLE PWAT and CAPE, respectively. It should be reiterated that this approach does not depend on bias correction; the predicted CAPE values are those underestimating what actually happened.
The results of the random forest classification are shown in Table 5 and Fig. 11a. The score metrics improve over the bias correction approach as follows: 83% POD; 77% SR; and 67% CSI. For the purposes of this study, it is unnecessary to attempt to maximize these scores from these relatively robust values. However, the selection of the 0.14 cm threshold was arbitrary and it is apparent in Fig. 11a that four of the six false alarms resulted from rainfall amounts slightly below that level. In addition, two of the four misses resulted from a midlatitude trough crossing the state during 15–16 June for which the PWAT is lower than typically found later during the monsoon season in this region.
As in Table 4, but for random forest classification predictions (pred) based on TLE PWAT and CAPE.
While the purposes for the SLC WFO FFPR metric are highly focused on the specific needs of the park staff to alert visitors to flash flood likelihood, it is instructive to compare the FFPR forecasts during the 2023 summer in Fig. 11d to the areal averages of MRMS QPE and HRRR analyses and TLE forecasts shown in the other panels of Fig. 11. It should come as no surprise that Probable or Expected predictions of flash floods at multiple parks are made when PWAT and CAPE tend to be high. The onset of the monsoon season on 18–19 July illustrates cases when intense localized thunderstorms led to flash floods in Capitol Reef National Park yet had less-intense rainfall across the entire region. These 2 days had the largest FFPR forecast spread across the five parks with the forecasters anticipating flash conditions to be Probable at Capitol Reef on both days and Possible or Not Expected at the other four parks. Note that the bias-corrected and random forest forecasts overestimated the areal coverage of rainfall to be expected on those days.
4. Summary
The NAM brings intense rainfall frequently during the summer to Mexico and the southwestern United States. Occasionally, flash floods in vulnerable locales within southwestern Utah lead to injuries, fatalities, and damage in population centers as well as heavily visited parks. The analysis of the 2021–23 seasons was focused on identifying the basic conditions conducive to intense rainfall over southwestern Utah that may lead to flash floods in the preferred areas that experience them. Rapid rainfall rates over the sparsely visited elevated plateaus or desert floors of the region will have less impact than those in the catchments upstream of the popular slot canyons where visitors congregate for recreational activities.
The 2021 and 2022 summers in southwestern Utah were two of the most active monsoon seasons on record leading to high numbers of flash flood reports. The 2023 monsoon season was delayed, and rainfall amounts across the region were somewhat lower than during the other two summers. However, the reduced number of flash flood reports during that season likely resulted more from the relative randomness of thunderstorms being less prevalent during the NAM within the catchments upstream of the most flood-prone canyons. Unfortunately, four deaths resulted from early season floods during March and May in the Buckskin Gulch canyon within the Grand Staircase-Escalante National Monument.
This research relied on the predominantly radar-based MRMS QPE analyses. While precipitation far from the KICX radar is likely underestimated, rainfall estimates from the MRMS across southwestern Utah appear reasonable. Estimates of rainfall and convection from station rain gauges, NLDN lightning, and MRMS QPE show the expected dependence on the underlying terrain with convection initiated at higher elevations during the afternoon. Also, as expected, widespread excessive summer rainfall in southwestern Utah generally occurs when the PWAT and CAPE are higher than typically observed across the region. The short time lag between early afternoon high CAPE and the onset of vigorous thunderstorms over the high terrain is often followed by subsequent downstream development that propagates above canyon headwaters by the prevailing flow.
The cumulative spatial and temporal variations of rainfall, moisture availability, and instability were examined during the three monsoon seasons. The prevailing conditions during 26 July 2021 were used to examine the factors contributing to seven flash floods, one in Cedar City, Utah, and the others scattered across several of the Parks in the region. PWAT and CAPE were unusually high relative to other days during the three summers with unidirectional flow from the southeast. The composite reflectivity above 60 dBZ near Cedar City and other locales would put these storms near the 90th percentile of flash flood-producing storms in the region (Smith et al. 2019).
During the three monsoon seasons, the HRRR generally provided accurate forecasts of the areal-averaged PWAT and under-forecasted CAPE at all lead times out to 18 h, similar to other findings of HRRR-forecasted CAPE and precipitation for convective events (Evans et al. 2018). The utility of using 12-member TLE means available 13–18 h prior to the typical onset of convection over the high terrain in the region was examined. Correcting for the negative CAPE forecast bias observed during the first two summers improved predictions of the conditions favorable for widespread rainfall during the 2023 summer. Using random forest classification with PWAT and CAPE as predictors for the upper quartile of areal-averaged rainfall showed promise at lead times relevant to operational predictions.
Repeating our analysis using TLE HRRR forecasts during summer 2024 will provide opportunities to test alternative machine learning strategies that might lead to improvements to the approach taken here. For example, each PWAT and CAPE forecast in the TLEs could be used as predictors albeit the lack of spread among the sample may limit that approach. Additional parameters could be considered (e.g., wind shear and HRRR QPE) or the domain could be confined to areas where flash floods are more likely or have a higher impact (e.g., limiting the areal extent to specific catchment basins within or near the Parks). The focus could also shift to predicting the extent within the domain of high rainfall rates (e.g., ≥1 cm h−1) rather than the daily rainfall total metric used here. Testing approaches for the entirety of southern Utah including areas with limited radar coverage would be possible by focusing on lightning as a proxy rather than MRMS QPE.
The operational transition of the HRRR to the Rapid Refresh Forecast System (RRFS) should be completed prior to summer 2025 (Dowell et al. 2022). Testing RRFS experimental output during summer 2024 may provide insight into how the approach developed here could take advantage of the RRFS six-member forecast ensembles to be available each hour as well as combining them over several initialization times into larger TLE ensembles. Loken et al. (2022) have explored random forecast approaches based on individual ensemble members as well as averages over the ensemble sample. In addition, it will be possible to extend the lead time for which forecasts of this type are available out to 36 h. Grim et al. (2024) have compared HRRR and RRFS forecasts of organized convective systems during summer 2022 in the eastern United States. There certainly will need to be similar studies undertaken to evaluate how well the RRFS handles the types of terrain-forced thunderstorms common to southwestern Utah and other regions within the western United States.
Meyer and Jin (2016, 2017) and Zhang (2023) among many other studies have examined the current and future trends of the NAM on the basis of downscaled global climate simulations. Since global and regional climate models have difficulty resolving precipitation over limited domains dominated by complex terrain such as southwestern Utah, downscaling proxy indicators of monsoonal strength (e.g., CAPE and PWAT as used in this study) may provide an approach to evaluate future changes in the NAM’s northern extent.
Acknowledgments.
We thank R. Chase, L. Dunn, A. Jacques, and W. J. Steenburgh for their comments related to this research. We also wish to express our appreciation to staff members (D. Church, G. Merrill, M. Seaman, and D. Van Cleave) of the Salt Lake City National Weather Service Forecast Office for their explanations of operational procedures used to forecast flash floods, feedback on this research, and access to data collected by their office. This work was possible thanks to the University of Utah Center for High Performance Computing (CHPC) for the computational hardware and the NOAA/National Weather Service Collaborative Science, Technology, and Applied Research (CSTAR) Program, Award NA20NWS4680046, through which this research has been supported. The creation of the HRRR-Zarr archive would not have been possible without an Amazon Sustainability Data Initiative Promotional Credits Award. We appreciate very much the work within our research group by T. Gowan, A. Jacques, and A. Kovac that made it possible to use and interpret the HRRR data. The first author acknowledges his role as a member of the Board of Directors of Synoptic Data PBC. The use of API services provided by Synoptic Data does not represent a conflict of interest.
Data availability statement.
The HRRR-Zarr archive is publicly available in the AWS Open Data Registry, made possible by credits awarded from the Amazon Sustainability Data Initiative: https://registry.opendata.aws/noaa-hrrr-pds/. Precipitation observations are available from Synoptic Data PBC. MRMS QPE files are available from the Iowa State University archives. Processed data files of PWAT, CAPE, and MRMS QPE used in the random forecast classification are available from the lead author upon request.
REFERENCES
Adams, D. K., and A. C. Comrie, 1997: The North American monsoon. Bull. Amer. Meteor. Soc., 78, 2197–2214, https://doi.org/10.1175/1520-0477(1997)078<2197:TNAM>2.0.CO;2.
Blaylock, B. K., and J. D. Horel, 2020: Comparison of lightning forecasts from the High-Resolution Rapid Refresh model to geostationary lightning mapper observations. Wea. Forecasting, 35, 401–416, https://doi.org/10.1175/WAF-D-19-0141.1.
Boos, W. R., and S. Pascale, 2021: Mechanical forcing of the North American monsoon by orography. Nature, 599, 611–615, https://doi.org/10.1038/s41586-021-03978-2.
Breiman, L., 2001: Random forests. Mach. Learn., 45, 5–32, https://doi.org/10.1023/A:1010933404324.
Burke, P. C., A. Lamers, G. Carbin, M. J. Erickson, M. Klein, M. Chenard, J. McNatt, and L. Wood, 2023: The excessive rainfall outlook at the Weather Prediction Center: Operational definition, construction, and real-time collaboration. Bull. Amer. Meteor. Soc., 104, E542–E562, https://doi.org/10.1175/BAMS-D-21-0281.1.
Chase, R. J., D. R. Harrison, A. Burke, G. M. Lackmann, and A. McGovern, 2022: A machine learning tutorial for operational meteorology. Part I: Traditional machine learning. Wea. Forecasting, 37, 1509–1529, https://doi.org/10.1175/WAF-D-22-0070.1.
Doswell, C. A., III, H. E. Brooks, and R. A. Maddox, 1996: Flash flood forecasting: An ingredients-based methodology. Wea. Forecasting, 11, 560–581, https://doi.org/10.1175/1520-0434(1996)011<0560:FFFAIB>2.0.CO;2.
Dougherty, K. J., J. D. Horel, and J. E. Nachamkin, 2021: Forecast skill for California heavy precipitation periods from the High-Resolution Rapid Refresh model and the coupled ocean–atmosphere mesoscale prediction system. Wea. Forecasting, 36, 2275–2288, https://doi.org/10.1175/WAF-D-20-0182.1.
Douglas, M. W., R. A. Maddox, K. Howard, and S. Reyes, 1993: The Mexican monsoon. J. Climate, 6, 1665–1677, https://doi.org/10.1175/1520-0442(1993)006<1665:TMM>2.0.CO;2.
Dowell, D. C., and Coauthors, 2022: The High-Resolution Rapid Refresh (HRRR): An hourly updating convection-allowing forecast model. Part I: Motivation and system description. Wea. Forecasting, 37, 1371–1395, https://doi.org/10.1175/WAF-D-21-0151.1.
Dunn, L. B., and J. D. Horel, 1994a: Prediction of central Arizona convection. Part I: Evaluation of the NGM and Eta model precipitation forecasts. Wea. Forecasting, 9, 495–507, https://doi.org/10.1175/1520-0434(1994)009<0495:POCACP>2.0.CO;2.
Dunn, L. B., and J. D. Horel, 1994b: Prediction of central Arizona convection. Part II: Further examination of the Eta model forecasts. Wea. Forecasting, 9, 508–521, https://doi.org/10.1175/1520-0434(1994)009<0508:POCACP>2.0.CO;2.
ElSaadani, M., W. F. Krajewski, and D. L. Zimmerman, 2018: River network based characterization of errors in remotely sensed rainfall products in hydrological applications. Remote Sens. Lett., 9, 743–752, https://doi.org/10.1080/2150704X.2018.1475768.
Evans, C., S. J. Weiss, I. L. Jirak, A. R. Dean, and D. S. Nevius, 2018: An evaluation of paired regional/convection-allowing forecast vertical thermodynamic profiles in warm-season, thunderstorm-supporting environments. Wea. Forecasting, 33, 1547–1566, https://doi.org/10.1175/WAF-D-18-0124.1.
Gourley, J. J., J. M. Erlingis, Y. Hong, and E. B. Wells, 2012: Evaluation of tools used for monitoring and forecasting flash floods in the United States. Wea. Forecasting, 27, 158–173, https://doi.org/10.1175/WAF-D-10-05043.1.
Gowan, T. A., J. D. Horel, A. A. Jacques, and A. Kovac, 2022: Using cloud computing to analyze model output archived in Zarr format. J. Atmos. Oceanic Technol., 39, 449–462, https://doi.org/10.1175/JTECH-D-21-0106.1.
Grim, J. A., J. O. Pinto, and D. C. Dowell, 2024: Assessing RRFS versus HRRR in predicting widespread convective systems over the eastern CONUS. Wea. Forecasting, 39, 121–140, https://doi.org/10.1175/WAF-D-23-0112.1.
Gutzler, D. S., and Coauthors, 2009: Simulations of the 2004 North American monsoon: NAMAP2. J. Climate, 22, 6716–6740, https://doi.org/10.1175/2009JCLI3138.1.
Herman, G. R., and R. S. Schumacher, 2018: Flash flood verification: Pondering precipitation proxies. J. Hydrometeor., 19, 1753–1776, https://doi.org/10.1175/JHM-D-18-0092.1.
Hill, A. J., and R. S. Schumacher, 2021: Forecasting excessive rainfall with random forests and a deterministic convection-allowing model. Wea. Forecasting, 36, 1693–1711, https://doi.org/10.1175/WAF-D-21-0026.1.
Hsieh, W. W., 2023: Introduction to Environmental Science. Cambridge University Press, 647 pp.
Hu, M., S. G. Benjamin, T. T. Ladwig, D. C. Dowell, S. S. Weygandt, C. R. Alexander, and J. S. Whitaker, 2017: GSI three-dimensional ensemble–variational hybrid data assimilation using a global ensemble for the regional Rapid Refresh model. Mon. Wea. Rev., 145, 4205–4225, https://doi.org/10.1175/MWR-D-16-0418.1.
Iowa State University, 2023: Iowa Environmental Mesonet. Accessed 10 October 2023, https://mesonet.agron.iastate.edu/plotting/auto/.
James, E. P., and S. G. Benjamin, 2017: Observation system experiments with the hourly updating Rapid Refresh model using GSI hybrid ensemble–variational data assimilation. Mon. Wea. Rev., 145, 2897–2918, https://doi.org/10.1175/MWR-D-16-0398.1.
Loken, E. D., A. J. Clark, and A. McGovern, 2022: Comparing and interpreting differently designed random forests for next-day severe weather hazard prediction. Wea. Forecasting, 37, 871–899, https://doi.org/10.1175/WAF-D-21-0138.1.
MacDonald, L. M., and C. J. Nowotarski, 2023: Verification of Rapid Refresh and High-Resolution Rapid Refresh model variables in tornadic tropical cyclones. Wea. Forecasting, 38, 655–675, https://doi.org/10.1175/WAF-D-22-0117.1.
Maddox, R. A., C. F. Chappell, and L. R. Hoxit, 1979: Synoptic and mesoscale aspects of flash flood events. Bull. Amer. Meteor. Soc., 60, 115–123, https://doi.org/10.1175/1520-0477-60.2.115.
Maddox, R. A., D. M. McCollum, and K. W. Howard, 1995: Large-scale patterns associated with severe summertime thunderstorms over central Arizona. Wea. Forecasting, 10, 763–778, https://doi.org/10.1175/1520-0434(1995)010<0763:LSPAWS>2.0.CO;2.
Marjerison, R. D., M. T. Walter, P. J. Sullivan, and S. J. Colucci, 2016: Does population affect the location of flash flood reports? J. Appl. Meteor. Climatol., 55, 1953–1963, https://doi.org/10.1175/JAMC-D-15-0329.1.
Martinaitis, S. M., S. B. Cocks, M. J. Simpson, A. P. Osborne, S. S. Harkema, H. M. Grams, J. Zhang, and K. W. Howard, 2021: Advancements and characteristics of gauge ingest and quality control within the Multi-Radar Multi-Sensor system. J. Hydrometeor., 22, 2455–2474, https://doi.org/10.1175/JHM-D-20-0234.1.
Martinaitis, S. M., and Coauthors, 2023: A path toward short-term probabilistic flash flood prediction. Bull. Amer. Meteor. Soc., 104, E585–E605, https://doi.org/10.1175/BAMS-D-22-0026.1.
Mazon, J. J., C. L. Castro, D. K. Adams, H.-I. Chang, C. M. Carrillo, and J. J. Brost, 2016: Objective climatological analysis of extreme weather events in Arizona during the North American monsoon. J. Appl. Meteor. Climatol., 55, 2431–2450, https://doi.org/10.1175/JAMC-D-16-0075.1.
McGovern, A., K. L. Elmore, D. J. Gagne II, S. E. Haupt, C. D. Karstens, R. Lagerquist, T. Smith, and J. K. Williams, 2017: Using artificial intelligence to improve real-time decision-making for high-impact weather. Bull. Amer. Meteor. Soc., 98, 2073–2090, https://doi.org/10.1175/BAMS-D-16-0123.1.
Meyer, J. D. D., and J. Jin, 2016: Bias correction of the CCSM4 for improved regional climate modeling of the North American monsoon. Climate Dyn., 46, 2961–2976, https://doi.org/10.1007/s00382-015-2744-5.
Meyer, J. D. D., and J. Jin, 2017: The response of future projections of the North American monsoon when combining dynamical downscaling and bias correction of CCSM4 output. Climate Dyn., 49, 433–447, https://doi.org/10.1007/s00382-016-3352-8.
Murphy, M. J., J. A. Cramer, and R. K. Said, 2021: Recent history of upgrades to the U.S. National Lightning Detection Network. J. Atmos. Oceanic Technol., 38, 573–585, https://doi.org/10.1175/JTECH-D-19-0215.1.
NCEI, 2023: Storm events database. NOAA, accessed 5 December 2023, https://www.ncdc.noaa.gov/stormevents/.
Powell, J. T., 2023: Analysis of southwestern Utah precipitation events associated with flash flooding. M.S. thesis, Dept. of Geological and Mining Engineering and Sciences, University of Utah, 91 pp.
Risanto, C. B., C. L. Castro, A. F. Arellano Jr., J. M. Moker Jr., and D. K. Adams, 2021: The impact of assimilating GPS precipitable water vapor in convective-permitting WRF-ARW on North American monsoon precipitation forecasts over northwest Mexico. Mon. Wea. Rev., 149, 3013–3035, https://doi.org/10.1175/MWR-D-20-0394.1.
Schumacher, R. S., 2017: Heavy rainfall and flash flooding. Oxford Research Encyclopedia of Natural Hazard Science, Oxford University Press, 1–41, https://doi.org/10.1093/acrefore/9780199389407.013.132.
Schumacher, R. S., A. J. Hill, M. Klein, J. A. Nelson, M. J. Erickson, S. M. Trojniak, and G. R. Herman, 2021: From random forests to flood forecasts: A research to operations success story. Bull. Amer. Meteor. Soc., 102, E1742–E1755, https://doi.org/10.1175/BAMS-D-20-0186.1.
Seaman, M., D. Church, and J. Cunningham, 2024: Leveraging probabilistic high resolution model guidance to improve flash flood forecasting across southern Utah. 28th Conf. on Probability and Statistics, Baltimore, MD, Amer. Meteor. Soc., 10.3, https://ams.confex.com/ams/104ANNUAL/meetingapp.cgi/Paper/430507.
Serra, Y. L., and Coauthors, 2016: The North American monsoon GPS transect experiment 2013. Bull. Amer. Meteor. Soc., 97, 2103–2115, https://doi.org/10.1175/BAMS-D-14-00250.1.
Sharif, R. B., E. H. Habib, and M. ElSaadani, 2020: Evaluation of radar-rainfall products over coastal Louisiana. Remote Sens., 12, 1477, https://doi.org/10.3390/rs12091477.
Smith, J. A., M. L. Baeck, L. Yang, J. Signell, E. Morin, and D. C. Goodrich, 2019: The paroxysmal precipitation of the desert: Flash floods in the southwestern United States. Water Resour. Res., 55, 10 218–10 247, https://doi.org/10.1029/2019WR025480.
Speiser, J. L., M. E. Miller, J. Tooze, and E. Ip, 2019: Comparison of random forest variable selection methods for classification prediction modeling. Expert Syst Appl., 134, 93–101, https://doi.org/10.1016/j.eswa.2019.05.028.
Sun, J., and Coauthors, 2014: Use of NWP for nowcasting convective precipitation: Recent progress and challenges. Bull. Amer. Meteor. Soc., 95, 409–426, https://doi.org/10.1175/BAMS-D-11-00263.1.
Thompson, G., M. K. Politovich, and R. M. Rasmussen, 2017: A numerical weather model’s ability to predict characteristics of aircraft icing environments. Wea. Forecasting, 32, 207–221, https://doi.org/10.1175/WAF-D-16-0125.1.
Wilks, D. S., 2011: Statistical Methods in the Atmospheric Sciences. 3rd ed. International Geophysics Series, Vol. 100, Academic Press, 676 pp.
Xu, M., G. Thompson, D. R. Adriaansen, and S. D. Landolt, 2019: On the value of time-lag-ensemble averaging to improve numerical model predictions of aircraft icing conditions. Wea. Forecasting, 34, 507–519, https://doi.org/10.1175/WAF-D-18-0087.1.
Yang, L., J. Smith, M. L. Baeck, and E. Morin, 2019: Flash flooding in arid/semiarid regions: Climatological analyses of flood-producing storms in central Arizona during the North American monsoon. J. Hydrometeor., 20, 1449–1471, https://doi.org/10.1175/JHM-D-19-0016.1.
Yu, G., B. J. Hatchett, J. J. Miller, M. Berli, D. B. Wright, and J. F. Mejia, 2023: Seasonal storm characteristics govern urban flash floods: Insights from the arid Las Vegas Wash watershed. J. Hydrometeor., 24, 2105–2123, https://doi.org/10.1175/JHM-D-23-0002.1.
Yussouf, N., and K. H. Knopfmeier, 2019: Application of the Warn-on-Forecast System for flash-flood-producing heavy convective rainfall events. Quart. J. Roy. Meteor. Soc., 145, 2385–2403, https://doi.org/10.1002/qj.3568.
Zhang, J., and Coauthors, 2016: Multi-Radar Multi-Sensor (MRMS) quantitative precipitation estimation: Initial operating capabilities. Bull. Amer. Meteor. Soc., 97, 621–638, https://doi.org/10.1175/BAMS-D-14-00174.1.
Zhang, J., L. Tang, S. Cocks, P. Zhang, A. Ryzhkov, K. Howard, C. Langston, and B. Kaney, 2020: A dual-polarization radar synthetic QPE for operations. J. Hydrometeor., 21, 2507–2521, https://doi.org/10.1175/JHM-D-19-0194.1.
Zhang, W., 2023: The dry and hot American Southwest under the present and future climates. Atmos. Ocean. Sci. Lett., 16, 100340, https://doi.org/10.1016/j.aosl.2023.100340.
