Spatial Patterns and Controls on Wind Erosion in the Great Basin

The Great Basin of the western United States is experiencing dramatic increases in wildfire and Bromus species invasion that potentially accelerate wind erosion and plant community change. We used a wind erosion model parameterized for rangelands and standard ecological monitoring data sets collected at 10,779 locations from 2011 to 2019 to characterize potential wind erosion in the Great Basin, assess relationships between factors affecting wind erosion, and quantify effects of wildfire and invasive Bromus species on aeolian horizontal sediment flux, Q. There were 403 monitoring plots (∼3.7% of plots) with Q > 100 g m−1 d−1. Median values for the highest Q category (>100) ranged from 196.5 to 308.5 g m−1 d−1. Locations with Q > 100 g m−1 d−1 were associated with dry, low elevation areas of the Great Basin with low perennial grass and perennial forb cover, and with large bare gaps between plants. Areas with high perennial grass, perennial forb, and shrub cover had small Q (≤10 g m−1 d−1). Substantial wind erosion was predicted in areas that have experienced wildfires, but areas with multiple wildfires had a lower predicted probability of Q particularly as invasive Bromus species cover increased. Modeled Q was up to two orders of magnitude higher post‐wildfire (median 44.2 g m−1 d−1) than in intact or annual grass‐invaded regions of the Great Basin (median 0.4 g m−1 d−1). Our results reveal the complex interplay among plant community composition, wildfire, and the amount of bare ground controlling wind erosion on Great Basin rangelands.

• Great Basin wind erosion is associated with dry and low-elevation areas with large amounts of bare ground • Little aeolian sediment transport occurs in sagebrush steppe where perennial grasses, forbs, and shrub cover are extensive • Bromus species cover suppresses wind erosion except following wildfire when pulses of aeolian sediment transport increase by orders of magnitude

Supporting Information:
Supporting Information may be found in the online version of this article.
In the western United States (US), wind erosion is of great concern when it leads to degradation of rangelands that are important economic, ecological, and cultural resources.The Great Basin, encompassing ∼520,000 km 2 , contains the largest area of managed rangelands in the western US (Morris & Rowe, 2014).The Great Basin is comprised of a heterogeneous mix of ecosystems with different soil types, salinity, precipitation, and elevation gradients that influence vegetation cover across sagebrush desert, salt-desert, intermountain sagebrush, woodlands, and sub-alpine forests (Stringham, Novak-Echenique, Blackburn, Coombs, et al., 2015;Stringham, Novak-Echenique, Blackburn, Snyder, & Wartgow, 2015;Stringham et al., 2017Stringham et al., , 2019)).Thus, wind erosion in the Great Basin is anticipated to be influenced by the interactions between climate, topography, soil, and vegetation community attributes that are further influenced by a suite of disturbances.Data on the spatial patterns and controls on wind erosion and dust emission from the Great Basin are minimal (Hand et al., 2017).Yet, efforts to increase sampling and long-term monitoring of vegetation communities within the Great Basin are currently underway, which provides an opportunity to examine these interactions at multiple spatial scales (Kachergis et al., 2022).
Studies examining atmospheric dust over the Great Basin have elucidated coarse spatial patterns of potential source areas.Remote sensing data suggest the Great Basin is a minor contributor to dust aerosols in the US (Ginoux et al., 2012;Prospero et al., 2002).Furthermore, analyses of spring and summer fine dust concentrations (PM 2.5 ) over the continental US have also not indicated the Great Basin is a major dust source (Hand et al., 2016(Hand et al., , 2017)).Hennen et al. (2022) identified dust emissions in the Great Basin were largest in the Great Salt Lake Area, with smaller emissions from the Owyhee Plateau associated with high wind speeds interacting with low surface roughness.Prospero et al. (2002) identified likely dust sources in the Great Basin with the characteristic attributes of being low-lying areas that accumulate alluvial sediments, salt flats, playas, and ephemeral or former lakes.
Within the Great Basin, field studies on dust emission have focused on effects of wildfire, which due to a reduction in vegetation cover can significantly increase erosion rates (Miller et al., 2012;Roehner et al., 2020;Vega et al., 2019).For example, Sankey et al. (2009a) measured large horizontal sediment mass fluxes (>5 kg m −1 over a 30-day period) in fire scars in shrublands of the Snake River Plain 2 months after wildfire.Similarly, a dust event 2 months after the Jefferson Fire in 2010 in southern Idaho produced an average horizontal sediment mass flux of 640 kg m −1 over a 2-week period (Wagenbrenner et al., 2013).Modeled estimates of post-fire PM 10 emissions were ∼32.1 Tg from 2012 fire scars in the Great Basin (Wagenbrenner et al., 2017).In addition to increasing bare ground, fire leaves organic compound residues on the soil surface that can reduce water infiltration capacity and potentially reduce the threshold shear velocity for soil particle motion (Ravi et al., 2006(Ravi et al., , 2007)).However, the importance of wildfire as a driver of regional wind erosion in the Great Basin relative to other potential causes is yet to be determined.
If wildfire is an important control on wind erosion in the Great Basin, then it is likely that spatial patterns and the timing of wind erosion are also influenced by the widespread invasion of non-native, annual Bromus species.Spatial patterns of native sagebrush-steppe vegetation and wildfire regimes in the Great Basin can be highly influenced by cheatgrass (Bromus tectorum) invasion (Balch et al., 2013;Bradley et al., 2018;Pilliod et al., 2017).Current estimates indicate ∼210,000 km 2 of the Great Basin is covered with ≥15% cheatgrass (Bradley et al., 2018).Cheatgrass invasion into sagebrush ecosystems has a positive, or amplifying, feedback effect on wildfire.Once established, this feedback decreases plant species richness and increases the cheatgrass fine fuel load that in turn accelerates wildfire frequency and intensity-further promoting cheatgrass establishment (Balch et al., 2013).This cheatgrass-wildfire cycle has emerged as an important driver of land cover change in the Great Basin-especially in arid, low-elevation areas with low perennial herbaceous vegetation cover and variable precipitation (Chambers et al., 2016;Williamson et al., 2020).Knowledge of the interactions between invasive species, such as cheatgrass, and wildfire frequency on wind erosion could directly inform post-wildfire management of Great Basin ecosystems to reduce the impacts of accelerated wind erosion.
Here, we characterize the magnitude, spatial patterns, and controls on aeolian sediment transport in the Great Basin.We examined the relationships among weather, soil, vegetation cover, elevation, disturbance, and aeolian sediment transport rates to describe the underpinning controls.We then quantified the effects of invasive Bromus species that includes cheatgrass species cover and wildfire on sediment transport rates.Our approach used an 10.1029/2023JG007792 3 of 20 aeolian sediment transport model parameterized for rangelands, applied to long-term ecological monitoring data to investigate regional patterns and drivers of wind erosion within the Great Basin.

Data and Modeling Approach
Data collected by the Bureau of Land Management's (BLM) Assessment, Inventory, and Monitoring (AIM) program were used to make regional inferences about wind erosion in the Great Basin.AIM collects standardized data on foliar cover, vegetation height, and size of canopy gaps, along with other vegetation and soil indicators across stratified and randomly sampled rangeland plots (Herrick et al., 2018;Taylor et al., 2014;Toevs et al., 2011).AIM data were used in two ways: (a) as inputs into an aeolian sediment transport model (described below) to provide indicators of wind erosion in the Great Basin, and (b) to identify how vegetation characteristics, including plant functional groups and species (e.g., cheatgrass) influence where, when, and why Great Basin rangelands are eroding.
Aeolian sediment transport was estimated for 10,779 BLM AIM monitoring plots across the Great Basin using the Aeolian EROsion (AERO) model (Edwards et al., 2022).AERO predicts horizontal sediment flux, Q (g m −1 d −1 ) following Kawamura (1951) as a function of surface wind friction velocity u s* following Okin (2008) and the surface threshold wind friction velocity u *t following Shao and Lu (2000).Soils were assumed to be dry and the Fécan et al. (1999) moisture correction was not used.AERO was parameterized using a Generalized Likelihood Uncertainty Estimation (GLUE) framework (Edwards et al., 2022).The parameterization used data on soil properties, vegetation foliar cover, vegetation height and spatial distribution, and measured Q from five National Wind Erosion Research Network (NWERN) sites, as described by Edwards et al. (2022) and Webb et al. (2016).
The GLUE framework identified 453 parameter sets with the majority of AERO predictions accurate within 90% prediction uncertainty bounds over several orders of magnitude of measured Q from the NWERN sites, and linear regression slope of 1:1, indicating the model performs well across a range of soil types and vegetation communities (Edwards et al., 2022).
Inputs to AERO were produced for each of the 10,779 AIM monitoring plots established in the Great Basin from 2011 to 2019 (Toevs et al., 2011).Surface (0-5 cm) soil texture for each plot location was acquired from the SoilGrids 100 m spatial resolution digital soil map (Ramcharan et al., 2018).The fractional cover of bare soil, vegetation canopy gap size distribution, and average vegetation height was calculated from AIM line-point intercept, canopy gap intercept, and vegetation height data (Herrick et al., 2018), respectively, using the terradactyl R package (McCord et al., 2022).Data were collected at monitoring plots once per year using sample designs guided by BLM national, state, and field office objectives for land health monitoring and assessment (Kachergis et al., 2020).Probability distributions of wind speed at 10 m above ground level were acquired for each plot location from gridded 3 hr and 1° spatial resolution data from the National Centers for Environmental Prediction North American Regional Reanalysis (Mesinger et al., 2006).AERO was run for each monitoring plot using 453 GLUE parameter sets, producing probabilistic estimates of potential Q based on the monitoring plot vegetation properties determined at the time of sampling and wind speed probabilities.Here, we used the median AERO predictions for the monitoring plots.Medians were used for our analyses because they describe the central tendency of skewed distribution of AERO predictions.

Stratification of the Great Basin and AIM Data
We used the Natural Resources Conservation Service Major Land Resource Area (MLRA) designations to stratify the Great Basin into nine regions (Figure 1).MLRAs are characterized by similarity in climate, soil, geology, topography, and vegetation (Natural Resources Conservation Service-USDA, 2006).Stratification of the monitoring plots by MLRA sought to account for regional variability and facilitate comparisons among areas with differing ecosystem attributes that may influence wind erosion, as well as potentially different land uses and management.We note that AIM data for MLRA 11 in 2011, and MLRAs 24, 25, and 28B in 2019 were not available for our analyses.Missing data were not expected to change overall results given the large number of AIM monitoring plots used in the analyses.

Spatial Patterns of Sediment Transport by Wind
Modeled sediment transport (Q) for AIM monitoring plots was mapped to identify regional patterns of Q within and among Great Basin MLRAs.We characterized modeled Q using base-ten order-of-magnitude categories.The categories were one of five levels: none (Q = 0), 0 < Q ≤ 1, 1 < Q ≤ 10, 10 < Q ≤ 100, and Q > 100 g m −1 d −1 .While previous studies have measured and reported rates of aeolian sediment flux that were orders of magnitude higher than our largest Q category (g m −1 d −1 vs. kg m −1 d −1 , e.g., Ravi et al., 2012;Sankey et al., 2009a), we did not further stratify our largest Q category to incorporate those reported by previous studies because the proportion of monitoring plots with modeled Q > 100 g m −1 d −1 represented a small fraction of the total number of plots across all MLRAs (approximately four percent, Table 1).These five categories encapsulated the variability in the magnitude of aeolian sediment transport reported in previous studies in similar biogeographical conditions (Alvarez et al., 2012;Sankey et al., 2009a) and disturbances (Sankey et al., 2009b;Wagenbrenner et al., 2013) as those in the Great Basin (Belnap et al., 2009;Munson et al., 2011;Nauman et al., 2018).The proportion of plots and the 25th, median, and 75th percentiles with modeled Q in each category were calculated for each MLRA.The proportions were then compared across levels of modeled Q between MLRAs to account for unbalanced sampling of monitoring plots across years.Using proportions for comparisons reduces bias introduced due to unbalanced sampling across years (e.g., not sampling the same number of plots across each year in each MLRA) and differences in the number of monitoring plots falling under each Q category.
Precipitation is an important control on Q in drylands, interacting with spatial patterns of soils and vegetation to influence soil erodibility and aerodynamic sheltering (Bergametti et al., 2016;Gherardi & Sala, 2015;Yoshioka et al., 2007).To examine the effect of variable precipitation on modeled Q, we calculated standardized anomalies from 30-year precipitation averages assuming stationarity for the western US for each of the years 2011-2019  and plotted the means for each year against normalized modeled Q.We calculated the standardized precipitation anomaly as mean annual precipitation minus the 30-year precipitation mean from 1991 to 2020 divided by the standard deviation of the 30-year mean.Scaling and normalizing modeled Q followed a similar procedure.Modeled Q for monitoring plots was subtracted from the mean Q across the Great Basin assuming stationarity and the resulting value was divided by the standard deviation of Q.We also calculated the previous winter season (December-January-February [DJF]) precipitation average for AIM sampling dates and locations to evaluate how the proportion of plots with varying modeled Q differed due to winter precipitation across MLRAs that typically supports vegetation growth.

Cross Correlation Analysis of Factors Explaining Variability in Wind Erosion
Spatial and temporal variability of modeled Q was related to foliar cover of different plant species (e.g., invasive annual grasses) and growth forms (i.e., annual and perennial forbs, grasses, and shrubs), soil, topography, and climate (Webb et al., 2014).To assess the relative influence of these controls on modeled Q, we ran Spearman's rank correlation analyses across the Great Basin and within each MLRA.Modeled Q was natural log transformed to account for skewness in the data.To include monitoring plots with modeled Q = 0 g m −1 d −1 a value of 1 g m −1 d −1 was added to Q so that natural log transformation resulted in a zero value in subsequent analyses.Potential explanatory variables included elevation, precipitation, surface (0-5 cm) soil texture, and ground cover indicators calculated from the AIM monitoring data.We used a cross-correlation analysis to assess how factors known to affect wind erosion and dust emission were related as a way to not overinterpret their effects when aggregating point data to larger spatial scales.
For each monitoring plot, elevation data were extracted from a 90 m digital elevation model of the western US (Hanser, 2008).Yearly, winter season (DJF), and 30-year (1991-2020) precipitation means were acquired from 4 km gridded PRISM Climate Group data (PRISM Climate Group, 2022; http://prism.oregonstate.edu).Percent sand for each plot was extracted from the 100 m SoilGrids data product for the conterminous US (Ramcharan et al., 2018).AIM data were used to calculate sagebrush (Artemesia) species cover, invasive Bromus species cover, annual and perennial forb cover, and annual and perennial grass cover using the terradactyl R package (McCord et al., 2022).We chose these cover indicators because they describe dominant plant species (sagebrush cover), invasive plant species (including B. tectorum cover), and functional plant species groups that are ecologically important controls on wind erosion (Galloza et al., 2018;Mayaud & Webb, 2017).The cover indicators describe the components of the total foliar cover and its structure that are inputs to AERO, and so provide understanding of vegetation controlling wind erosion in the Great Basin beyond the functional representation of ground cover in AERO.

Influence of Wildfire and Invasive Bromus Species on Wind Erosion
The interactions between wildfire, invasive Bromus species, and wind erosion are potentially complex, with the timing, frequency, and intensity of wildfire and the extent of invasive Bromus species influencing protective ground cover and soil properties that may exacerbate or attenuate wind erosion.Recognizing this complexity, as a first step toward understanding the interactions, we assessed how the magnitude of modeled Q changed across monitoring plots with time since fire and measured cover of invasive Bromus species.We computed the number of wildfires associated with each AIM monitoring plot using the BLM National Fire Perimeters Polygon data set (https://data.doi.gov/dataset/blm-natl-fire-perimeters-polygon-d33d9).We extracted the wildfire incident name and discovery date of each wildfire polygon that overlapped with AIM monitoring plot locations.We extracted the total number polygons that overlapped each AIM monitoring plot and used only the number of wildfires that occurred before the date of AIM sampling in the logistic regression (described below).Monitoring plots with wildfire occurrence after AIM sampling were grouped with plots having no wildfire occurrence in the analysis.
To visualize the effect of post-fire wind erosion, a subset of AIM monitoring plots sampled within 1 year of a wildfire occurrence discovery date were plotted against modeled Q and invasive Bromus species cover.The within 1 year wildfire occurrence was determined as the difference between the most recent wildfire discovery date and the date an AIM monitoring plot was sampled.Only monitoring plots that had a difference greater than zero or less than or equal to 1 year were used.
We used logistic regression to model the relationship between the probability of modeled Q occurrence and MLRAs, sampling year, the number of wildfires, and invasive Bromus species cover associated with monitoring plots.Modeled Q was converted to a binary response variable.AIM plots with modeled Q > 0 were assigned a 1; otherwise, 0 if modeled Q = 0.A binary response in Q facilitated comparison of potential wildfire effects because large variability and skewness in Q occurred when aggregating values at broad spatial scales like those of MLRAs.The number of historical wildfires associated with AIM monitoring plots as a predictor variable was used to test wildfire effects on the probability of Q occurring.To account for the majority of monitoring plots having no wildfire occurrence, a dummy/indicator variable was added to the model that was coded as 1 to account for no wildfires and zero for plots with one or more wildfires.We used MLRA as a categorical predictor and AIM sampling year as a continuous predictor to control for broad spatial scales and sampling year.Percent cover of invasive Bromus species measured at monitoring plots was used as a continuous predictor.A quasi-binomial distribution was used to account for the difference in observed versus expected variance described by our set of predictor variables.Analyses were conducted using R v4.2.1.

Spatial Patterns of Wind Erosion in the Great Basin
Across the Great Basin, the majority of monitoring plots sampled in 2011-2019 had modeled Q between 0 and ≤10 g m −1 d −1 (Table 1 and Figure 2).The proportion of sampled plots with modeled Q > 10 g m −1 d −1 was highest for MLRAs 24, 27, and 29 and the lowest for MLRAs 11, 23, 25, and 26.Thirty-two percent of monitoring plots in the northern most MLRA, the Snake River Plains of southern Idaho (MLRA 11), had no predicted horizontal sediment flux.These monitoring plots were located along the north-central edge of the Great Basin.Only 9% of monitoring plots in MLRA 11 were predicted to have Q > 10 g m −1 d −1 , predominately along the south-western border.The Fallon-Lovelock Area (MLRA 27) in west-central Nevada had 53% of monitoring plots with modeled Q > 10 g m −1 d −1 .We found clusters of monitoring plots with high and extreme modeled Q located in the periphery of the eastern MLRA 23, northwestern MLRA 24, and western MLRA 25.
Large spatial and temporal variability were evident in the magnitude of modeled Q, indicated by changes in the proportion of sampled plots with different magnitudes of Q across sampling years (Figure 3).From 2013 to 2015, the proportions of sampled plots with predicted Q > 100 g m −1 d −1 increased across all MLRAs.The annual proportions of Q were inversely related to previous winter precipitation.For example, the proportion of plots with zero or low modeled Q was most prevalent for MLRAs when previous winter precipitation was relatively large as in 2017.Conversely, the proportion of monitoring plots with medium to large modeled Q (>1 g m −1 d −1 ) was higher when the previous winter precipitation was small as in 2012-2015.
Total annual precipitation was variable among years and MLRAs of the Great Basin.We found most MLRAs received less than average precipitation from 2011 to 2013 (Figure 4).The third consecutive year of drier than

Influence of Soils, Landforms, and Vegetation on Wind Erosion
The amount (e.g., % cover indices) and distribution (e.g., % canopy gaps) of vegetative cover were important controls on aeolian sediment transport in the Great Basin (Figure 5).In general, horizontal sediment flux (Ln Q) was negatively correlated with indicators of vegetation cover.As expected, total foliar cover was negatively correlated with Ln Q across the Great Basin.The strength of the association between total foliar cover and Ln Q varied among MLRAs.The correlation estimate ranged between ρ = −0.65 (MLRA 27) to ρ = −0.32(MLRA 23, p < 0.001 for all statistically significant comparisons, Figures S1-S9 in Supporting Information S1).Interestingly, we found no relationship between total foliar cover and Ln Q in MLRA 11.Relationships between total foliar cover and plant species/groups comprising monitoring plots affect wind erosion because total foliar cover is a composite estimate of the cover of plants species/groups.Indeed, we found the largest positive correlation between total foliar cover and perennial grass cover for all functional groups.Thus, similar to total foliar cover, perennial grass cover was negatively correlated to Ln Q.The strength of the association between perennial grass cover and Ln Q was the highest for MLRA 11 and lowest for MLRA 27.Modeled Ln Q was negatively correlated with perennial forb cover, marginally to weakly negatively correlated with annual forb cover and annual grass cover, and marginally negatively correlated with sagebrush (ARTR cover in Figure 5) species cover.
The magnitude and direction of the relationships among horizontal sediment flux (Ln Q), topography, climate, and soil varied across the Great Basin (Figure 5).First, we found no significant association between elevation and vegetation foliar cover across the Great Basin.However, stratifying by regions, elevation and foliar cover were positively correlated for six of nine MLRAs (Figures S1-S9 in Supporting Information S1).Elevation was associated with winter season and mean annual precipitation and Ln Q was negatively correlated with elevation (ρ = −0.19,p < 0.001).Ln Q was positively correlated with the percentage of sand (0-5 cm) at monitoring plots (ρ = 0.14, p < 0.001).However, this result differed among MLRAs.The correlation ranged between 0.09 and 0.3 (p < 0.05 across comparisons) for five MLRAs (Figures S1-S9 in Supporting Information S1).Ln Q and percent sand were not correlated for MLRA 24, 25, and 29.In MLRA 28A, Ln Q and percent sand were weakly negatively correlated (ρ = −0.1,p < 0.001).

Effects of Fire and Invasive Bromus Species on Wind Erosion
A greater number of fires occurred in the northern portion of the Great Basin from the Snake River Plains (MLRA11) southwest to the Owyhee High Plateau (MLRA 25) (Figures 1b and 6); these two MLRAs also had the highest invasive Bromus species cover.Logistic regression model results demonstrated that the probability of Q occurring varied with MLRA, year that monitoring plots were sampled, and one or more fires (Table 2).Increased Bromus species cover and the number of fires occurring prior to AIM monitoring were found to have significant negative relationships to the predicted probability of Q (Figure 6).The probability of Q occurring decreased by two percent for every one percent increase in invasive Bromus species cover (Table 2).Furthermore, the probability of Q occurring declined by ∼20% for each increase in the number of fires.While most AIM monitoring plots had no fire association (n = 8,406 monitoring plots), 2,373 plots had at least one fire.Regions with large number of fires (e.g., ≥3 fires) had large variability and higher median cover estimates of invasive Bromus species and total foliar cover than regions with monitoring plots with three or fewer fires (Figure 7).The probability of Q decreased with increasing invasive Bromus species cover but remained high with no fires.The probability of Q also decreased with increasing number of fires and increasing invasive cover.One-third of monitoring plots with invasive Bromus species cover sampled within 1 year after a fire had large to extreme modeled aeolian sediment transport rates (Figure 8).

Discussion
Understanding the geographic distribution and drivers of wind erosion on rangelands is vital for understanding dust impacts on Earth's systems, as well as how we may improve land management to mitigate soil loss and dust emissions.Within the Great Basin, modeled aeolian sediment fluxes had large variability within and among eco-geomorphic regions defined by MLRAs.Overall, we found that hot spots of larger modeled aeolian sediment fluxes occurred across the north-western portion of the Great Basin across the Owyhee Plateau south to the Fallon-Lovelock Area.Wind erosion in some of these hot spots was associated with playas within the Black Rock Desert, Carson Sink (Figure 2).Smaller hot spots were found in the Snake River Plains and are likely related to wildfire and vegetation treatments that result in low vegetation cover (Figure 2).Hot spots of the Great Salt Lake Area south of the Great Salt Lake are likely driven by low precipitation in combination with soils inhospitable to plant growth, such as playas (Figure 1 and Figure S7 in Supporting Information S1).Further study of the wind erosion hot spots is needed to establish in detail the within-region eco-geomorphic and land use drivers and controls.However, the eco-geomorphic characteristics of these hot spots were generally consistent with dust hot spots globally identified by Prospero et al. (2002), having low elevation, sandy soils, and low annual rainfall.In addition, the analysis of dust optical depth, DOD, over the Great Basin by Ginoux et al. (2012) identified regional dust hot spots that are consistent with our findings of elevated modeled wind erosion in the eastern Malheur High Plateau, Owyhee Plateau, the Fallon-Lovelock and Great Salt Lake areas, and across the Southern Nevada Basin and Range into the Mojave Desert.The overlap in regions with elevated modeled sediment transport described by this study to those in Ginoux et al. (2012) suggests a potential link between dust in the atmosphere measured through DOD originating from or being contributed to by regions we identified.
Among MLRAs, we found modeled aeolian sediment fluxes are likely responding to temporal variability in precipitation.We note that the pattern of Q and yearly proportions of the different orders of magnitude do not necessarily represent an MLRA entirely because the locations of sampled plots differed among years (Figure 3).Therefore, the proportions offer a snapshot in time of potential sediment fluxes within the MLRAs at the sampled plots.Our results suggest a lagged response of wind erosion to periods of up to 3 years where mean annual precipitation was below the 30-year mean (drought conditions) within the Great Basin (Figure 4).This lagged response may manifest itself by an order of magnitude increase in aeolian sediment transport rates compared to previous years of drier than average conditions.However, a longer time series would be needed to establish if there is a consistent temporal response of wind erosion to interannual precipitation variability within the Great Basin.The largest increases in modeled aeolian sediment flux across most MLRAs were associated with precipitation in 2011-2013 being below the 30-year mean.These MLRAs had low amounts of sagebrush and perennial grass cover with large amounts of bare soil and canopy gaps >1 m (Figures S1-S9 in Supporting Information S1).Moreover, annual precipitation above the 30-year mean may not necessarily reduce wind erosion as evidenced by little change in the proportions of modeled aeolian sediment fluxes (Figure 4).Because precipitation strongly influences plant production, we expected to find that effects of these conditions were linked to vegetation cover (Munson et al., 2019;Noir-Meir, 1973).In some Great Basin ecosystems and landforms, such as playas, vegetation cover may only ever reach a maximum amount of biomass/cover that is less than what would be expected with average to above average precipitation (Knapp et al., 2017).
The relationships among modeled aeolian sediment fluxes, elevation and soil texture and their interactions with precipitation and plant cover were also predicted to influence wind erosion.Modeled Q was larger in lower elevation areas than at higher elevations in the Great Basin (Figure 5 and Figures S1-S9 in Supporting Information S1) likely due to less plant cover driven by low precipitation at lower elevations (Munson et al., 2019).Modeled Q increased with percent sand and decreasing perennial forb and perennial grass cover, which was expected given higher observed plant productivity in finer textured soils in the Great Basin (Stringham, Novak-Echenique, Blackburn, Coombs, et al., 2015;Stringham, Novak-Echenique, Blackburn, Snyder, & Wartgow, 2015;Stringham et al., 2017Stringham et al., , 2019)).Similar variability in the complex interplay among precipitation, ground cover, and wind erosion has been identified elsewhere, such as the Australian rangelands (Webb et al., 2006(Webb et al., , 2009)), and is qualitatively consistent with geophysical explanations of wind erosion occurring at hot spots within broader rangeland settings (e.g., Gillette, 1999).Interestingly, however, we found that soil texture (% sand content) alone was not a strong predictor of patterns of potential wind erosion within the Great Basin (Figure 5).This was likely due to the AERO model parameterization of soil erodibility (u *t ) assuming all soils are loose, dry, and granulated; therefore, the effects of changes in soil crusting and dry aggregate stability that are influenced by soil texture are not accounted for in the model estimates of Q (Edwards et al., 2022).
Modeled Q was also correlated with plant functional group cover and structure across the Great Basin.Plant communities in the Great Basin vary by region and elevation but are predominantly sagebrush with herbaceous understories (Stringham, Novak-Echenique, Blackburn, Coombs, et al., 2015;Stringham, Novak-Echenique, Blackburn, Snyder, & Wartgow, 2015;Stringham et al., 2017Stringham et al., , 2019)).Both perennial grass and perennial forb cover had the most influence on potential wind erosion (Figure 5).Interestingly, sagebrush cover had only a weak effect on potential wind erosion and annual forb, annual grass, and invasive Bromus species cover had the weakest effects considering the Great Basin as a whole (Figure 5).However, we did find variability in these effects among MLRAs (Figures S1-S9 in Supporting Information S1) for example, in some potential wind erosion hotspots, there was a negative correlation between sagebrush cover and modeled Q, as in the Fallon-Lovelock area (Figure S6 in Supporting Information S1).This may be due to variable sagebrush cover across the MLRAs resulting from increased disturbances like wildfires exacerbated by the invasion of non-native Bromus species.
In the other hotspots, wind erosion potential was larger where perennial grass cover was lower, as in the eastern portion of the Malheur High (Figure S2 in Supporting Information S1) and Owyhee High Plateaus (Figure S4 in Supporting Information S1).Other studies have also shown a strong association between perennial grass loss and increased wind erosion risk on the Colorado Plateau (Belnap et al., 2009;Miller et al., 2011), in Chihuahuan Desert ecosystems (e.g., Bergametti & Gillette, 2010;Li et al., 2007;Webb et al., 2014) and in Great Plains grasslands (Ravi et al., 2012).Overall, the relationships between modeled Q and plant functional group cover variation at hotspots identified in this study are likely site specific.Further research is needed to elucidate the sediment transport responses to spatial variability to ecological potential, for example, using disturbance response grouping or state-and-transition models (Heller et al., 2022;Stringham et al., 2016).The spacing of vegetation, a key control on wind erosion and described by the cover of canopy gaps >1 m (Webb et al., 2021), was significantly influenced by perennial grass cover.Larger cover of canopy gaps was associated with reduced perennial grass cover, which likely explains why perennial grass cover had a persistently strong relationship with potential wind erosion (Figure 5).The negative correlation between canopy gaps >1m and perennial grass cover was due to decreased amount of bare ground with increasing perennial grass species cover which in turn minimized the amount of unvegetated plant interspaces (e.g., Gondard et al., 2003;Heller et al., 2022;Webb et al., 2021).The weak negative relationship between invasive Bromus species cover and gaps >1 m overall within the Great Basin should not be interpreted as this indicator not being important for patterns of wind erosion (Figure 5 and Figures S1-S9 in Supporting Information S1).On the contrary, the cheatgrass-fire relationship has major implications for patterns of wind erosion in the Great Basin (Hahnenberger & Nicoll, 2012;Jewell & Nicoll, 2011;Miller et al., 2013;Sankey, Germino, Sankey, & Hoover, 2012;Sankey et al., 2009a;Wang et al., 2019).Bromus species cover, which in the Great Basin is predominantly comprised of cheatgrass (B.tectorum), increases the probability of wind erosion.
Wildfire is highly influential in Great Basin ecosystems and its increase in frequency and severity is affecting wind erosion by contributing to changes in plant community structure and composition (Ellsworth et al., 2020; Mahood & Balch, 2019;Miller et al., 2013;Sankey et al., 2009a).Across the Great Basin, we found that increasing invasive Bromus species cover and an increasing number of fires resulted in decreased probability of wind erosion (Figure 6).Decreased wind erosion was expected when plant cover, from any plant species, increased (Breshears et al., 2003).However, while Bromus species invasion provides ground cover, its presence has significant tradeoffs including increasing the susceptibility of sagebrush steppe ecosystems to wildfire (Balch et al., 2013).Immediately post-fire, modeled wind erosion potential dramatically increased (Figure 8), which is consistent with field measurements (e.g., Ravi et al., 2012;Sankey et al., 2009a;Wagenbrenner et al., 2013).Pulses of wind erosion may occur after wildfire, in which aeolian sediment transport rates increase by orders of magnitude where there is sufficient supply of loose erodible material (Sankey, Germino, & Glenn, 2012).This supports our finding that, despite the general decrease in sediment fluxes with increasing Bromus species cover, the potential for wind erosion is consistently high with increasing Bromus species cover.The presence of Bromus species increases variability in plant cover (Figure 7)-plant cover being high pre-fire and low post-fire-and over time on average wind erosion increases.Thus, managing invasive Bromus species will also benefit soil health (e.g., carbon and nutrient cycling), and air quality as they are influenced by wind erosion (Bagchi et al., 2013;Hand et al., 2017;Wagenbrenner et al., 2013Wagenbrenner et al., , 2017)).
Insights into the spatial patterns and controls on wind erosion in the Great Basin were enabled by application of AERO to the large monitoring data set collected by the BLM AIM program.Using standardized data setssuch as AIM-to produce estimates of wind erosion potential provides high-quality data that allowed stronger comparisons among and within regions (Kachergis et al., 2022;Taylor et al., 2014).We recognize, however, that further work is needed to improve the utility of monitoring data and AERO to other regions, land uses, and circumstances of management concern.For example, the lack of data on military lands in Nevada precluded developing a complete picture of wind erosion in the Great Basin.Continued efforts to standardize monitoring across land ownerships and collaborations with various federal management agencies will aid in filling in data gaps (Kachergis et al., 2022).Additionally, while AERO has been parameterized for diverse rangeland conditions (Edwards et al., 2022), further parameterization is needed for post-fire wind erosion.For instance, it is known that fire can affect soil erodibility by influencing hydrophobicity and cohesion (Dukes et al., 2018;Ravi et al., 2006Ravi et al., , 2007) ) which can lead to physical crusting of soils post-fire.The effects of physical crusting, hydrophobicity, and cohesion resulting from wildfire on soil erodibility are not currently constrained or accounted for in AERO (Edwards et al., 2022).Future work parametrizing AERO will include these factors and their effects on horizontal sediment flux and dust emissions.Identifying potential wind erosion hotspots in the Great Basin has provided new insights into the location of potential dust sources in the Great Basin and a step toward identifying where dust may be transported and its downwind impacts.Describing these components of the Great Basin dust cycle will be pertinent to understanding the interactions between invasive plants, wildfire, and wind erosion across rangelands and other dryland ecosystems globally.

Conclusions
In the Great Basin region, modeling indicates wind erosion is likely to occur in drier, low-elevation areas including playas, as in the Fallon-Lovelock, eastern Malheur High Plateau, and the central portion of the Southern Nevada Basin and Range, with higher sand content, low amounts of perennial grass and perennial forb cover, and with large distributions of bare gaps between plants.Accelerated wind erosion also occurred in the Snake River Plains, western Owyhee Plateau, and Humboldt areas, which produce high amounts of plant biomass but have been invaded by Bromus species and experience repeat wildfires.Conversely, areas with high amounts of (preferably) native perennial and sagebrush cover (e.g., high total foliar cover), such as in the Central Nevada Basin and Range areas, likely experience low levels of wind erosion because low number of fires and increased plant cover reduces wind erosivity, shelters bare soil, and traps soil particles.We found that the probability of aeolian sediment transport occurring is high across the Great Basin with increasing Bromus species cover and slightly decreases for most regions with increasing number of fires.However, wind erosion rates can be orders of magnitude larger post-fire, especially in Bromus species-encroached landscapes.
While aeolian sediment transport occurs in more xeric parts of the Great Basin where vegetation cover is sparse, and less frequently in sagebrush steppe where shrub and perennial grass and forb cover are high, sediment transport rates are strongly affected by annual invasive Bromus species and wildfire that suppress and exacerbate wind erosion respectively.While invasive Bromus species provide ground cover that shelters the soil, the protection afforded comes with an elevated risk of wildfire and dramatic post-fire erosion.Considering our findings beyond the Great Basin ecosystems, our results reveal the complex interplay that can exist among the abiotic and biotic factors controlling rangeland wind erosion.Exotic plant invasion and native plant encroachment are common across rangelands globally.Continuing to build an understanding of the interactions with aeolian processes will be critical for managing rangelands to avoid accelerated erosion and associated impacts on land health and air quality.This research was funded by the U.S. National Science Foundation (NSF) and Natural Environment Research Council (NERC), Grant: 1853853.The authors gratefully acknowledge the contribution of Adrian Chappell for providing conceptual discussion and editing to the article, Sarah McCord for input and insights to BLM data, Darren James for statistical advice, and Brandi Wheeler, Katie Young, Dylan Burruss, and Joslynn Romero for input with analyses.This research was a contribution of the Long-Term Agroecosystem Research (LTAR) network supported by the U.S. Department of Agriculture (USDA).Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the US Government.

Figure 1 .
Figure 1.Delineation of the Great Basin study area by Major Land Resource Area designations (a), Bureau of Land Management wildfire perimeters colored by approximately 20-year increments from 1911 to 2019 (b), elevation (c), mean annual precipitation from 2011 to 2019 (d), and estimated percent sand (e) and clay (d) content in soils (Ramcharan et al., 2018).

Figure 2 .
Figure 2. Location and geographic distribution of 10,779 Assessment, Inventory, and Monitoring (AIM) plots where the Aeolian EROsion model was used to predict aeolian horizontal sediment flux, Q, in g m −1 d −1 .AIM monitoring plots were established and sampled from 2011 to 2019.

Figure 3 .
Figure 3. Proportions of Aeolian EROsion modeled aeolian horizontal sediment flux (Q; g m −1 d −1 ) colored by order of magnitude for monitoring plots sampled across nine Major Land Resource Areas comprising the Great Basin of the western United States from 2011 to 2019.Mean winter precipitation for each year is represented by the black points and the right y-axis.

Figure 4 .
Figure 4. Comparison between the proportions of monitoring plots normalized by values of Q across all Major Land Resource Areas and deviations from the 30 years (1991-2020) mean precipitation for the 9-year period of 2011-2019.The left-hand y-axis pertains to the bars representing proportions of Q magnitude and the righthand y-axis to the black points.Black points represent mean precipitation deviation for each year and bars represent the standard deviation of mean precipitation.

Figure 5 .
Figure 5. Spearman's rank cross-correlation matrix for the direction and magnitude of the relationship between aeolian horizontal sediment flux (Ln Q) and 14 variables related to plant cover, climate, and topography across the Great Basin.Red colors indicate positive and purple colors negative correlations.Darker coloration indicates relative magnitude of strength of pair-wise comparisons.Values between comparison indicate significant correlation coefficients at a p < 0.001.Missing comparison values indicate no significant relationship between comparison.Abbreviations in figure are ARTR for sagebrush, DJF for December-January-February (winter months), non-ARTR for non-sagebrush, and ppt for precipitation.

Figure 6 .
Figure 6.Predicted probability of horizontal sediment flux (Q > 0) against invasive Bromus species cover colored by the number of fires resulting from logistic regression model.

Figure 7 .
Figure 7.Comparison box-and-whisker plots of the distribution of invasive Bromus species cover and total foliar cover measured at monitoring plots as a function of the number of fires established in Major Land Resource Areas.Total foliar cover is a composite index that includes the proportion of invasive Bromus species cover if it was present at monitoring plots.

Figure 8 .
Figure 8. Modeled horizontal sediment flux (natural log transformed) against Bureau of Land Management Assessment, Inventory, and Monitoring (AIM) sampling date at monitoring plots colored by the percentage of invasive Bromus species cover.Data represent responses in Q of monitoring plots sampled in 2011-2019 within 1 year of the most recent fire.Symbols represent Major Land Resource Areas (MLRA).

Table 1
Comparison of Modeled Aeolian Horizontal Sediment Flux (Q) Across Major Land Resource Areas (MLRA) of the Great Basin

Table 2
Logistic Regression Model Assessing the Relationships Between Major Land Resource Areas, Year Monitoring Plots Were Sampled, the Percentage of Invasive Bromus Species Cover, and the Number of Fires Associated With Monitoring Plots on the Probability of Horizontal Sediment Flux (Q)