Temporal and spatial variations in the frequency of compound hot, dry, and windy events in the central United States

Simultaneous low humidity, high temperature, and high wind speeds disturb the water balance in plants, intensify evapotranspiration, and can ultimately lead to crop damage. In addition, these events have been linked to flash droughts and can play a critical role in the spread of human ignited wildfires. The spatial patterns and temporal changes of hot, dry, and windy events (HDWs) for two time periods, 1949 to 2018 (70-years) and 1969 to 2018 (50-years) were analyzed in the central United States. The highest frequencies of HDWs were observed at stations in western Kansas and west Texas. Annually, the highest number of events happened concurrently with the major heat waves and droughts in 1980 and 2011. Temporally, an overall decrease in the HDWs was significant in the eastern regions of North Dakota and South Dakota, and an upward trend was significant in Texas and the western part of the Great Plains. Significant trends in HDWs co-occurred more frequently with significant trends in extreme temperatures compared with low humidity or strong wind events. The results of this study provide valuable information on the location of places where HDWs are more likely to occur. The information provided could be used to improve water management strategies.


Background
Desiccated corn acreage in southern Kansas in 1888 39 provides a historic example of the impact of HDWs on cropland. An extended and severe drought was reported in Kansas in 1887 and in the next year (1888) a series of hot winds destroyed 30% of the corn crop in south-central Kansas. Between 1883 and 1888 nearly 40 occurrences of HDWs were reported in the central United States including locations in Kansas, Nebraska, the Dakotas, Texas, and Arkansas 39 . The majority of the events were in June, July, and August. In September 1931 and following a month without rain, seven consecutive days of hot winds were reported at Ashland, Kansas, including a report of 49 °C (120 °F) for September 5th 40 . Similar reports from observers in the Great Plains and other states documented crop damage from only a few hours of HDWs.
A variety of terms and definitions are used in the literature to describe HWDs. Lydolph and Williams 11 suggested considering the co-occurrence of low relative humidity, high temperature, and stronger winds to define an hourly HDW event. They identified five different classes of HDWs based on relative humidity less than 30%, wind speed less than or greater than 7 m/s, and temperature higher than 29 °C (Table 1). Leathers and Harrington 9 classified HDWs or "furnace winds" with a temperature higher than 35 °C (a threshold considered critical for crop development), relative humidity lower than 30%, and wind speed equal to or greater than 7 m/s. Reid et al. 17 categorized wildfires in Oklahoma into six classes and showed that different classes of wildfires are more probable when the average wind speed is in the range of 3.5 m/s (for class 1) to 7 m/s (for class 6) when relative humidity ranges from 15 to 30%. Kruger et al. 16 found that in the southern Great Plains, the favorable For the impact of wind speed, a 25% reduction in grass leaf extension was discovered with an increase of wind speed from 1 to 7.4 m/s 43 . Besides, a 50% increase in Helianthus annuus (sunflower) water requirement 44 and a 50% reduction in the dry weight of marigolds grown 45 when wind speed is higher than 7 m/s (15 miles per hour).
Our analysis indicates that in the Great Plains, the average wind speed is 4.5 m/s and 4.3 m/s for 70-years and 50-years periods, respectively. Wind speed of 7 m/s is equal to the 85th percentile threshold for both periods. The 90th percentile of wind speed for the entire Great Plains is equal to 8.2 m/s and 7.7 m/s for 70-years and 50-years periods, respectively. Considering the mentioned information in the literature and high wind speed in the Great Plains, the 7 m/s threshold was considered which is also the same as in previous studies 9,11 that analyzed the geography of HDWs in the United States.
For relative humidity, a leaf water deficit and small root water deficit were reported for relative humidity values between 30 and 35% 46 or about 25% 47 . Hoffman et al. 48 analyzed the impact of low (25%) and high (90%) relative humidity on cotton plants and found a significantly higher leaf diffusion resistance when the humidity is low. O'Leary and Knecht 49 revealed that the yield of bean plants decreased by 40% when the RH decreased from 70-75 to 35-40%. For Lettuce sativa, growth will decrease when the relative humidity decreases to 35-40% 50 .
Seasonally, a majority of HDWs are reported in late spring and summer worldwide. In Siberia, the "Sukhovey" winds occur primarily in May and June, with very few in late summer 10 . In the Great Plains, USA, HDWs mainly occured in mid-summer, however, the less-frequent HDWs of the southeastern United States occur primarily during less humid periods in September, June, or May 11 . In China, most HDWs are reported in May and June 12 . A study from a wind tower in Algeria documented a HDW event blowing from the deserts during the hot season with a wind speed of 4.7 m/s in July 54 . Across Africa, the "Khamsin", "Sharav", and "Sirocco'' HDWs mostly occur in late spring and early summer. As HDWs are usually expected between May and September, the warm season in the central United States (May through September) was selected for this study.

Results and discussion
Two periods were selected to study temporal and spatial changes in HDWs. A 70-years period (1949-2018) was studied using 27 stations and a 50-years period (1969-2018) was assessed using 44 stations (Fig. 1). The frequency of hourly HDW events was determined at each station for each month in the warm season. Then, monthly values were summed to calculate an annual (warm season) total. No matter which time period was considered, the highest annual frequencies of HDWs occurred at stations in western Kansas and Texas (Fig. 1b,c). Across www.nature.com/scientificreports/ all stations analyzed, the mean annual frequency of HDWs ranged from less than 1.0 to more than 60.0 h, with the largest number occurring in southwest Kansas (Dodge City). This finding reinforces results from a previous study analyzing 1948-1993 data that identified Dodge City, Kansas, as the hotspot 9 . This location coincides with a high-speed-wind region in the United States [55][56][57][58] . Analysis of the geographic pattern of mean monthly wind speed for 1961-1990 showed the highest values in the Great Plains, with the largest values in Kansas and Texas in June, July, August, and September 55 . Spatial analysis of wind-power density showed the highest classes of wind speed (greater than 7 m/s) in southwest Kansas, northwest Nebraska, and North Texas in the summer 58 .
Using the information in the 2011 National Land Cover Database (NLCD 59 ), greater frequencies of HWDs were located in croplands and grasslands. Four interpolation methods including Inverse Distance Weighted (IDW), Natural Neighbor, Ordinary Kriging, and Universal Kriging were used to interpolate the point-based data. Two statistical error indicators including the mean absolute error (MAE) and root mean square error (RMSE) were then used to test the results of the interpolation methods. IDW was then selected based on the minimum error calculated for both MAE and RMSE.
Then, average linkage 60 and K-mean 61,62 clustering methods were applied to cluster the data based on the frequency of HDWs, extreme high temperature (higher than 35 °C), extreme low relative humidity (less than 30%), and extreme high wind speed (higher than or equal to 7 m/s). First, the optimal number of clusters were specified using silhouette method 63 and then the clusters were visualized as maps (Fig. 2). Based on the silhouette method 63 , four and eight clusters were specified as the optimal number of clusters for the 50-year and 70-year periods, respectively. The clustering results were a little different based on the two methods. However, the main grouping pattern were the same especially for the 50-year period ( Fig. 2a,b). Cluster 1 groups together stations with lower numbers of HDW events and proportionally high number of extreme high wind speed events that includes stations along the Texas coast and those in the northeastern part of the study area. Cluster 2 includes the stations from central Kansas south and westward into the Texas Panhandle. These are the stations that are most at risk of having a HDW event. Cluster 3 groups stations with the highest number of extreme low relative humidity events and includes stations in the west. From eastern Kansas southward into east Texas, stations are grouped as cluster 4 ( Fig. 2a,b). These stations recorded a higher number of extreme high temperature but a lower number of extreme low relative humidity and extreme high wind speed events. Clustering of the longer time series, produces a larger number of optimal clusters, with Cluster 2 representing the stations with the highest number of HDW events (Fig. 2c,d). The additional clusters produced with the analysis of 70 years of data suggest within group differences for the 50 year solution. For example, the low frequency of HDW events cluster from the 50-year analysis (Cluster 1) include Clusters 4 and 7 in the 70-year analysis (Fig. 2).
Year  The monthly frequency of hourly HDWs showed the same pattern of occurrence for both periods with a maximum in July (Fig. 4). Analyzing the monthly pattern of single variable extremes showed the highest probability of extreme temperature (higher than 35 °C) occurred in July and August. However, extreme wind events (higher than or equal to 7 m/s) occurred mostly in May and June and the distribution was almost equal in all months for dry relative humidity extremes (less than 30%).
Stations in the southern Great Plains had the highest frequencies of monthly values of HDWs (Fig. 5). Extreme temperature events mostly occurred in Texas, Kansas, and Oklahoma. For relative humidity, the majority of  showed no significant relationship between the annual frequency of HDWs and the annual frequency of extreme relative humidity. However, a significant correlation was determined for both wind and temperature extremes with HDWs. An afternoon maximum is a predominant characteristic of the diurnal pattern of HDW events (Fig. 6). Hourly observations document that HDWs occur between 10 a.m. and 11 p.m. with a maximum frequency in the afternoon (4:00 p.m. and 5:00 p.m. local time). However, The highest diurnal frequency of HDWs was different in diverse stations ranging between 2:00 p.m. and 6:00 p.m. (Fig. 7).
The statistical t test analysis showed no statistically significant difference on the diurnal pattern of HDWs between 50-year and 70-year periods. Analyzing the separate components of HDWs, extreme high temperature events were most limiting for the occurrence of HDW events (Fig. 6). Same as HDWs, the maximum occurrence of extreme high temperature events, extreme low relative humidity, and extreme high wind speed were all discovered at 4:00 and 5:00 p.m. Hourly analysis of the three components documents that wind and RH can meet HDW criteria at any hour of the day, but temperature exceedance is limited to 10 a.m. to 11 p.m. (Fig. 6b,c).
The Mann-Kendall trend test was used to analyze any station-based trend in the frequency of HDWs. From 1949-2018 and using annual totals, 30% (75% positive) of the stations showed a significant trend (Fig. 8). When analyzed on a monthly basis, August had the highest percentage of significant trends (18%; 11% positive). Negative trends in August occurred in eastern areas of South Dakota and North Dakota. Other months did not have stations with a decreasing trend. Positive monthly trends were most significant in Texas, Kansas, Colorado, and Montana.
From 1969-2018, 27% (67% positive) of all stations showed a significant trend in the annual HDW event total. For this analysis period, the majority of significant positive trends (87%) were in Texas. This finding can be linked to the higher number of HDW events associated with dry periods later in the time series (e.g., 2011). All statistically significant negative trends occurred in eastern portions of Nebraska, South Dakota, and North Dakota. The decrease was consistent with the decrease of extreme low humidity, high temperature, and wind speed. Analysis of monthly frequencies (Table 2)   www.nature.com/scientificreports/ demonstrates the importance of data period on the results. It is interesting to note that when the data are summarized for the entire warm season (compared to individual monthly results), a larger number of stations have a statistically significant trend. Figure 8 shows the temporal changes of HDWs at stations that experience at least one HDW in each warm season. Station location is identified in Fig. 9. Trends in extremes of temperature, wind speed, and relative humidity were also analyzed to understand the influence of each variable on temporal changes in HWDs (Fig. 9). For the 50-years period, 45% of stations had a significant trend for extreme temperature (41% positive) and relative humidity (43% positive) events. Among stations that had a significant trend for high wind speed in the 50-years period (36%), there was a similar number of stations with positive and negative trends (18%). For the 70-years period, 59% of stations had a significant www.nature.com/scientificreports/ trend in extreme temperature events that were mostly positive (52%). The changes in extreme low relative humidity were less significant (19%; 11% positive) over the longer period. However, high wind speed had a significant negative trend for a majority (52%) of stations. Only one station located in Texas had a significant positive trend for high wind speed in the 70-years period (Fig. 9).
The changes in extreme temperature events are consistent with global climate change 2,29,30 . Only two stations located in the north-eastern part of the study area had a significant negative trend for both time periods. This area might be considered as part of the "warming hole" in the United States where temperature and extreme temperature events have downward trends [68][69][70][71][72] . Future changes in the occurrence of compound extreme hot and dry days during crop-growing seasons will negatively influence crop yield 73 .
Although water vapor is increasing in the atmosphere 2,70,74 , the frequency of extreme low humidity events has also been increasing in western Great Plains. At 57% of the stations, the upward trend in extreme temperature events corresponds with an upward trend in the frequency of low relative humidity events in the 50-years period. Applying coupled global climate models, an increase was discovered in the annual frequency of dry days in the central United States 75 . When the time series does not include the southern Great Plains drought of the 1950s, the upward trends in extreme temperature and low relative humidity are statistically significant for stations across Texas. The downward trend of extreme wind events is consistent with previous study 36 in which a decline was discovered in the 90th percentile and annual mean wind speed in the United States. Here, a decrease in HDWs was discovered at the stations with a decline in all three single extremes (Fig. 9). However, an increase of HDWs may affect water resource management challenges associated with the greater evapotranspiration 18 .

Summary and conclusions
Spatial and temporal variations of HDWs were analyzed for the central United States (including the Great Plains) over two different periods of 1969-2018 (50-years) and 1949-2018 (70-years). For the 50-years period, there were more stations (44 stations) compared to the 70-years period (27 stations) that helped lead to a better understanding of spatial patterns. HDWs were defined as compound hourly events with high wind speeds (higher than or equal to 7 m/s), high temperature (higher than 35 °C), and low relative humidity (lower than 30%). Frequency analysis showed a greater occurrence of HDWs in western Kansas southward into Texas. This was consistent with the spatial pattern of extreme wind (wind speeds higher than 7 m/s) and extreme temperature (temperatures higher than 35 °C) events. The monthly analysis showed a greater probability of HDWs in July, with fewer occurrences in May and September. Compound HDW events were mainly observed in the afternoon with the highest frequencies at 4:00 and 5:00 p.m.
Results document the influence of the data period on the trend analysis. However, most stations located in Texas and the western Great Plains showed an upward trend in the HDWs. A downward trend in HDWs was found in the northeastern portion of the study region, consistent with the decrease in the extreme temperature, humidity, and wind speed events over time. Temporally, the central United States experienced large numbers   The higher temperatures associated with climate change may increase the frequency of extreme HDWs. HDW events have implications not only for those involved in crop production. Wind-driven wildfires and increased evapotranspiration effects on water management systems are potential impacts of these compound extreme meteorological events. Adaptation and mitigation strategies may need to be adjusted to cope with the negative impacts of these events.

Methods
The HDW definition advanced by Leathers and Harrington 9 was used in this study. After calculating the frequency of HDWs in each annual warm season, changes in the frequency of these events were analyzed spatially and temporally, considering frequency in different years and individual months. Then, the Mann-Kendall trend test 76,77 , which is widely used for climate variables [78][79][80] , was applied to determine the existence of any monotonic trend in the frequency of HDWs over time. In addition, the widely used Pearson and Spearman correlation test 81,82 was applied to look for associations between HDWs and extreme temperature, humidity, and wind speed. A two-sided significant level of 0.05 was used for all parameters. For HadISd data, all times are provided as coordinated universal time (UTC). However, the central United States contains two different time zones, Mountain and Central. To better understand the diurnal changes of HDWs, the stations were separated based on their time zone and then local time was calculated for each station. Average linkage 60 from hierarchical clustering methods and K-mean 61,62 from non-hierarchical or partitional clustering methods were selected to cluster stations using R packages 83,84 . Study area. Previous studies 9,11 showed the greatest probability of HDWs in the Great Plains in the central United States. The higher number of extreme HDWs in this region can be explained in part by the relatively www.nature.com/scientificreports/ high mean annual wind speed and a summer maximum in mean wind speeds 37 . The relative flatness of the Great Plains and a lack of tree cover are contributing factors. With an area of 2,898,107 km 2 , the ten states that contain the Great Plains span from Texas in the south to Canada in the north, and from the Rocky Mountains eastward to Kansas. In this study, weather observing sites within 10 states of Montana, Wyoming, Colorado, New Mexico, North Dakota, South Dakota, Nebraska, Kansas, Oklahoma, and Texas, were analyzed.
Data. Sub-daily temperatures, relative humidity, and wind speed data were obtained from HadISD for 1949-2018 to analyze the long-term (70-years) changes of compound HDWs in the central United States. HadISD (version 3.0.1.201906p) is a station-based, sub-daily, quality-controlled dataset available from the Met Office website (https ://www.metoffi ce.gov.uk/hadob s/hadis d). The dataset uses a subset of the Integrated Surface Database (ISD) 85 from the National Oceanic and Atmospheric Administration (NOAA) National Centre for Environmental Information (NCEI), which initially provided data with temporal coverage beginning in 1973 86 . The HadISD provides the NCEI data in a more easily accessible format. They also have performed more detailed quality control on data and merged some stations to make longer records where this was reasonable to do so 86,87 . An update of the HadISD data added new stations and extended the temporal coverage back to 1931 87 . The most recent version of HadISD data includes 7677 stations with global coverage filtered by updated quality-control methods 87 . The new sets of HadISD data contain sub-daily humidity and heat-health measurements (e.g., heat index and apparent temperature). Stations with less than 10% missing data were selected for this study. In addition, stations with an entire year missing were excluded from the study to prevent bias in the results of temporal trend analysis.