The Signature of Climate in Fluvial Suspended Sediment Records

Arid regions are often characterized by exceptionally high rates of fluvial sediment transport, but the processes responsible for this apparent connection between climate and sediment transport remain unclear. We examined decades of continuous flow records and suspended sediment concentrations from 71 rivers across the United States by comparing the suspended sediment rating curve behavior, quantified using power law coefficients and exponents, to an aridity index. Results indicate that higher aridity correlates with both greater overall suspended sediment concentration and lower sensitivity of concentration to changes in discharge, demonstrating that rivers in arid locations on average have greater suspended sediment transport efficiency across most discharges, and achieve high transport rates at a relatively lower discharge than rivers in temperate climates. Furthermore, based on additional analyses of the Normalized Difference Vegetation Index, specific suspended sediment yield, and a hydrograph flashiness index, we attribute the relationships between sediment transport and climate primarily to differences in vegetation density, precipitation, and runoff, variables that all influence both sediment supply and riverbed grain sorting. Finally, we note that the observed contrasts in sediment transport behavior likely represent climate‐driven differences in the magnitude and frequency of sediment supply rather than annual suspended sediment load, which does not depend significantly on climate. This study highlights a critical connection between multiple interrelated climatic factors and sediment transport, an important finding for future hazard mitigation in a changing climate with rapidly shifting vegetation patterns and hydrology.


10.1029/2023JF007429
2 of 18 sediment transport typically occurs during large but infrequent flash floods (Cohen & Laronne, 2005;Laronne & Reid, 1993;Reid & Laronne, 1995).On a larger scale, Ludwig and Probst (1996) showed in a study of 60 world rivers that rivers in dry climates had higher mean annual total suspended sediment than rivers in temperate climates for a given specific runoff.Thus, although wetter regions have much greater total precipitation and streamflow than dry regions, this does not necessarily translate into greater rates of sediment transport, as has also been demonstrated using erosion rates across a large climate gradient along the Andes (Carretier et al., 2013(Carretier et al., , 2018)).
Notably, the relationship between sediment transport efficiency and climate is distinct from that between annual sediment yield and climate, a topic of considerable scrutiny over several decades.Early work by Fournier (1949) and Langbein and Schumm (1958) explored the connection between sediment yield and precipitation, putting forth distinct models for how climate affects sediment yield through its influence on runoff and vegetation cover.However, additional work over the following years suggests that there is no clear universal control of climate on sediment yields (Jansson, 1988;Milliman & Farnsworth, 2011;Syvitski & Milliman, 2007;Walling & Webb, 1983).The lack of an obvious signal between these two variables makes the aforementioned observations of high sediment transport efficiency in arid locations all the more noteworthy.That is, if the total amount of sediment rivers transport annually does not strongly depend on the climate, then the magnitude and frequency of sediment transport may instead explain differences in observed sediment transport rates across climates.
Past work may reveal potential links between sediment transport efficiency and climate through the shape of the suspended sediment rating curve.Syvitski et al. (2000), in an effort to predict rating curve behavior from numerous variables, found covariance between rating curves, mean discharge, flow variability, precipitation, and temperature among other factors.Although it featured few sites from arid climates, their study suggests a connection between climatic variables and the behavior of sediment transport beyond overall sediment yield.For example, there is indeed a clear correlation between climate and runoff variability, an important determinant of erosion on landscapes and sediment transport in river channels (Rossi et al., 2016).Vegetative cover is also intimately related to climate, which affects both runoff and soil erosion, and thus the total contribution of sediment to fluvial systems, as well as the magnitude and frequency of events that supply sediment to channels (e.g., Gyssels et al., 2005;Loch, 2000;Noble, 1965;Puigdefábregas, 2005).While some of the variables investigated by Syvitski et al. (2000) may themselves not directly have a strong impact on suspended sediment transport (such as temperature or latitude), they may reflect the combined effect of multiple climatic factors that could reasonably cause differences in the efficiency of sediment transport across climate types.However, a high-level detailed analysis of sediment transport behavior across a much larger range of climates is necessary to make such connections.
To uncover the potential links between climate and sediment transport efficiency, here we present an investigation of continental-scale patterns in direct suspended sediment measurements and explore possible mechanisms responsible for the observed trends.While there are many local catchment variables that influence rates of sediment transport in individual rivers-land cover, dominant lithology, vegetation type, or slope, for exampleexamining overall patterns across dozens of rivers may cut through the noise to reveal fundamental relationships between climate and sediment transport.This work is highly relevant to risk management, particularly with respect to flood hazards and infrastructure damage.Sediment supply and transport efficiency are directly related to variability in riverbed elevation and flood conveyance (Pfeiffer et al., 2019), which are particularly high in arid settings (Slater et al., 2019;Slater & Singer, 2013).This volatility in channel geometry leads to major erosion and sediment deposition problems near channels (e.g., Merritt & Wohl, 2003;Wilcox et al., 2016).Thus, understanding the processes driving high rates of sediment transport will help assess which rivers pose the greatest danger to nearby development, particularly in the face of shifting climatic and hydrologic conditions due to anthropogenic climate change.

Site Selection
For this study we use publicly available suspended sediment concentrations from the U.S. Geological Survey (U.S. Geological Survey, 2022).Although sediment transport occurs both as suspended load and as bedload, directly measuring bedload transport is resource-and labor-intensive (Reid et al., 1980) and would be infeasible for this large-scale study.Fortunately, the fraction of total load transported as bedload is relatively small except 10.1029/2023JF007429 3 of 18 in sand-bedded rivers with very low suspended sediment flux (Turowski et al., 2010).Thus, while suspended sediment and bedload are not equivalent, studying suspended sediment alone provides important insight into the overall patterns of sediment transport efficiency given the large proportion of sediment transported in suspension.
Suspended sediment loads in fluvial systems are also highly sensitive to land use (e.g., Abbott et al., 2018;Gao et al., 2013;Rossi et al., 2009).To reduce the anthropogenic signal in our data, we began with the HCDN-2009 dataset from the GAGES-II project (Falcone, 2011), which identifies U.S. Geological Survey gauge locations suitable for hydroclimatic study across the United States and Puerto Rico.These 743 HCDN-2009 sites have at least two decades of continuous flow records beginning in Water Year 1990, are currently in "reference" condition (i.e., not significantly altered by agriculture, dams, irrigation, or other land use in the watershed), have watersheds with no more than 5% impervious surface, and have not been labeled as unsuitable for study by evaluators from state Water Science Centers.
For each HCDN-2009 gauge location with available suspended concentration sediment data, we selected sites with at least 50 discrete field measurements of suspended sediment concentration beginning in the Water Year 1990 and spanning at least five unique years.A multi-year record is important to account for any years with anomalous weather patterns as well as for short-term oscillations in climate, which influence suspended sediment in rivers (Gray et al., 2015).
These high standards for site quality yielded 65 potential USGS gauges, but very few were in the southwestern United States, the driest region in our study area (Figure 1).In order to have a sufficient representation of sites across different hydroclimates, we also drew from gauges outside the HCDN-2009 dataset, examining sites from the "WestXeric" ecoregion in the GAGES-II database with a hydrologic disturbance index of less than 10 (the maximum in the database being 42), excluding one location with gravel excavation.This process resulted in 20 additional sites from the southwestern U.S.
It is possible that the more lenient standard for the hydrologic disturbance index at Xeric West sites introduced bias into our dataset.For example, dams have a large influence on the amount of fine sediment flowing through rivers, and although these structures are absent upstream of the majority of our selected sites, a few Xeric West locations have dams upstream of the USGS gauge station.However, we do not deem this disqualifying in the context of this study; dams trap fine sediment, often leading to lower suspended sediment loads and coarsening of bedload downstream (Draut et al., 2011;Galay, 1983;Grant et al., 2013), which would create the opposite effect of what we expect for arid streams in this region.Thus, if anything, the inclusion of these dammed sites would likely dampen the signal we observe between climate and suspended sediment behavior.Furthermore, in terms of possible geographic over-and under-representations, there was a relatively high number of sites in Idaho and Colorado, and five sites within a 300 km 2 quadrilateral near Lake Tahoe.However, we chose not to eliminate any additional sites for the sake of avoiding human bias in our site selection, while still taking note of these small irregularities.

Aridity Index as a Climate Metric
While many parameters can be used to describe a particular climate, arguably the most important aspect is whether that region is dry or humid.With this in mind, we use Aridity Index, or the ratio of mean annual precipitation to potential evapotranspiration, as our principal climatic parameter to assess the influence of climate on sediment transport.
We determined the Aridity Index at each of our gauge locations using a high-resolution raster dataset for the years 1970-2000 (Zomer et al., 2022).This dataset assigns Aridity Index values to 30 arc-second cells (several hundred meters depending on latitude), with lower values representing more arid locations, and higher values more humid.We calculated the mean Aridity Index for all raster cells within each watershed using the rasterstats Python package (Perry, 2023).Watershed boundaries for all USGS gauges were provided by the GAGES-II dataset (Falcone, 2011).

Suspended Sediment Analysis and Further Site Refinement
Continuous stream discharge data, typically recorded every 15 min, were obtained for each site over the period of time spanned by suspended sediment concentration data.If discharge measurements began after the earliest suspended sediment measurements, some suspended sediment records were discarded to match the timespan of available discharge data.Each suspended sediment measurement was paired with a corresponding discharge, based on timestamps rounded to the nearest 15-min interval.If no matching discharge was available within 15 min of the suspended sediment measurement, and no alternative record of discharge was available in a separate field measurement dataset, the data point was discarded.
The relationship between suspended sediment concentration and discharge was usually nonlinear, and for most sites, data behaved linearly in log-log space at moderate and high discharges (Figures 2a and 2b).In most cases, we excluded some low-flow data due to data resolution issues; suspended sediment concentration is typically reported as an integer, causing an artificial shallowing of the rating curve slope at low discharges (Figure 2c).In other cases, sediment transport appeared genuinely different at low flow conditions relative to stronger flows (Figures 2d-2f); however, given that the majority of sediment transport over time is expected to occur at moderate-to-high discharges (Wolman & Miller, 1960), deemphasizing low-flow conditions is also justifiable from a geomorphic standpoint.Following the reasoning above, we isolated data above a particular discharge where suspended sediment concentration and discharge began to behave linearly in log-log space based on a visual fit.For some sites, there was a log-log linear relationship between the two variables even at low discharges, in which case no lower discharge cutoff was applied and all data were used (Figures 2a and 2b).Linear regression was applied to the remaining data in log-log space to obtain a best-fit power function relating discharge to suspended sediment concentration in the form where SSC is suspended sediment concentration (mg/L) and Q is volumetric discharge (m 3 /s).The exponent, a, is dimensionless, while the coefficient, k, links discharge to concentration and thus has units of (mg L −1 ) (s m −3 ) a .Suspended sediment measurements of 0 mg/L could not be log-transformed and thus were excluded from the power law fit; fortunately, these values often occurred at low discharges, which are less significant in this study for the reasons stated above.Given the differences in rating curve behavior at low flow versus moderate to high flow (due to data resolution issues or true trends in sediment transport), some rating curves  2f).Typically, data resolution issues attributed to integer-based reporting of sediment concentrations were most common in temperate locations (Figure 2c), while rapid rises of suspended sediment concentrations at low flow and subsequent shallowing of the rating curve at higher flow conditions (Figure 2f) were more characteristic of hyper-arid locations in this study.Singular power law fits to these large data sets also ignore variations in suspended sediment rating curves over time due to hysteresis; however, this study aims to examine trends across much longer timescales than individual storm events or particular seasons, so grouping all data together is appropriate.In summary, while power law fits cannot perfectly capture the variety of relationships between suspended sediment and discharge, for the purposes of this large-scale study, these functions should adequately represent sediment transport behavior-particularly at moderate to high flows.
We use the exponent, a, and coefficient, k, in Equation 1 as important quantifiers of sediment transport behavior in this study.Together, these values capture how suspended sediment concentration responds to increasing discharge (a) as well as how much sediment is carried within the flow at any given discharge (k).In order to avoid drawing erroneous conclusions about exponents and coefficients for the rivers in this study, further data refinement was required.
After applying the discharge cutoff and a power law fit, some sites had no clear relationship between suspended sediment concentration and discharge or few remaining data points (Figure S1 in Supporting Information S1).Thus, to preserve data quality and avoid small exponents unrepresentative of true rating curve behavior, sites were discarded if regressions had p-values greater than 0.05 (8 sites) or were based on fewer than 50 measurements (6 sites) (Table S1).There was no apparent geographic or climatic pattern in these discarded sites (Figure S2 in Supporting Information S1).
The final set of 71 gauge locations spanned a large spectrum of aridity indices and geography (Figure 1).The median number of suspended sediment measurements per site (after regression) was 334, and the median timespan of a given dataset was ∼20 years (Table S2).

Results
Rating curve behavior, as described by the coefficient and exponent, varied considerably with aridity.Regression analysis indicates that increases in Aridity Index (AI)-i.e., a wetter climate-are generally associated with decreases in the coefficient (k = 0.797 AI −4.23 ; R 2 = 0.47, p < 0.001) and increases in the exponent (  = 1.28 + 0.368 ln(AI) ; R 2 = 0.20, p < 0.001) (Figure 3).These patterns are particularly apparent when plotting the exponent against the coefficient and coloring by the Aridity Index (Figure 4a); while there is some scatter in the data, suspended sediment rating curves for arid rivers (red) generally have larger coefficients and smaller exponents than for more temperate systems (blue).This figure also shows that overall coefficients and exponents are inversely correlated; rating curves appear unlikely to have both high coefficients and exponents.This inverse relationship between rating curve parameters has also been observed in prior studies (Asselman, 2000;Syvitski et al., 2000).Of course, this may simply be a consequence of mathematics; assuming there is some limitation on available sediment, a steeper slope in linear regression will force a smaller intercept.All rating curve coefficients, exponents, and corresponding watershed aridity indices are summarized in Table S2.
As another way to visualize these findings, we constructed two representative power-law rating curves for hypothetical arid and temperate streams by inserting the 10th and 90th percentile aridity indices from our site list (0.23 and 1.21, respectively) into the linear regressions between Aridity Index and both rating curve parameters (Figure 3), and plotting the suspended sediment concentration over the full range of discharges in our dataset (Figure 4b).In log-log space, the 10th percentile (arid) rating curve is shallower with a higher baseline suspended sediment concentration relative to the 90th percentile (temperate) rating curve.The large difference in the coefficient between the two locations causes the arid stream to have higher suspended sediment concentrations than the temperate stream across all discharges.However, the higher exponent for the representative temperate rating curve causes the suspended sediment concentration to approach that of arid streams as the discharge increases.The disparity in suspended sediment concentration between the two rating curves is particularly pronounced at low to moderate discharges, and for the suspended sediment concentration of the temperate stream to exceed that of the arid stream, we must extrapolate the rating curves to higher discharge values than those found in this study.

Discussion
Our results show that, on average, arid streams have greater suspended sediment concentrations for a given discharge-and thus greater suspended sediment transport efficiency-than streams in more temperate locations (Figure 4b).Furthermore, given the low coefficient and high exponent in their rating curves, rivers in temperate locations transport little suspended sediment at low and moderate flow conditions and must reach relatively high flow conditions to transport appreciable volumes of sediment.This highly nonlinear relationship between discharge and suspended sediment concentration is less evident in arid streams, which transport high concentrations of sediment across a wider range of discharges.Rating curve parameters, a and k, refer to the exponent and coefficient (respectively) of a power-law fit between discharge (Q; m 3 /s) and suspended sediment concentration (SSC; mg/L) in the form SSC = kQ a (Figure 2).Log axes were used when a particular variable was more normally distributed in log space.Linear regressions in semi-log (above) and log-log (below) space are also shown as red lines, with corresponding equations listed in red text along with an R 2 and p-value for the regression.More arid locations occur to the left of the plots, while more temperate sites tend toward the right.

Role of River Size in Rating Curve Behavior
Before discussing the physical explanation for climate-based contrasts in rating curve behavior, we consider the possibility that the size distribution of rivers in our study skewed our results.In a study of the Rhine River, Asselman (2000) found that with increasing discharge downstream, suspended sediment rating curves shallowed, with the regression coefficient (k) increasing and the exponent (a) decreasing.The author attributed this observation to the fact that the same increase in discharge on a small stream represents a proportionally higher change in flow relative to the mean than in a large river.Hence, smaller rivers would show steeper rating curves simply because of their relatively lower discharge range compared to larger rivers.River discharge may thus be a hidden factor in the broad patterns observed in suspended sediment rating curve behavior across climates.
To explore whether our rating curve trends are influenced by river size, we plotted the rating curve coefficient and exponent against the geometric mean discharge for each site, with discharge values of 0 m 3 /s removed from the calculation (Figure S3A in Supporting Information S1).Regression analyses reveal a weak positive relationship between mean discharge and the exponent a (R 2 = 0.14; p = 0.002), and a strong negative relationship between mean discharge and the coefficient k (R 2 = 0.54; p < 0.001).
While these regressions show a significant relationship between mean discharge and rating curve behavior, the relationships are the opposite of those from Asselman (2000); increasing discharge in our study correlates with steeper rating curves, rather than shallower.Such a relationship was also uncovered by Syvitski et al. (2000) among their study locations.This may in fact be related to climate; rivers in wetter regions tend to be larger, as is clear from the moderate positive relationship between Aridity Index and the geometric mean discharge (Figure S3B in Supporting Information S1).Our analysis reaffirms that climate likely plays a significant role in influencing suspended sediment rating curves in this broad study.

Sediment Transport Efficiency in Relation to Sediment Yield
Although our work has focused primarily on the distinctive behavior of suspended sediment transport rating curves, the possibility must be addressed that high rates of suspended sediment transport in arid locations (particularly at non-peak flow conditions) are simply a product of high overall sediment loads.As mentioned in the Introduction, multiple studies have examined relationships between annual specific sediment yields and various climatic factors such as precipitation, runoff, temperature, and vegetation cover (e.g., Fournier, 1949;Jansson, 1988;Langbein & Schumm, 1958;Wilson, 1973;Zhang et al., 2022).Perhaps the most well-known of these studies, by Langbein and Schumm (1958), posited that as precipitation increases, there is a tradeoff between greater runoff, which facilitates erosion, and vegetation density, which reduces it.In this model, sediment yield is maximized in arid and semi-arid climates, which could explain the results of our analysis.However, multiple subsequent studies have cast doubt on these findings, noting that there is no clear relationship between sediment yield and precipitation or runoff, at least on a global scale (Milliman & Farnsworth, 2011;Renwick, 1996;Walling & Webb, 1983;Wilson, 1973).Moreover, multiparameter models of fluvial sediment flux that take into account factors like climate, lithology, slope, and human influence found that climatic factors accounted for only 14% of all variability in fluvial sediment loads, while local geologic and topographic variables were much more important (Syvitski & Milliman, 2007).In summary, past work lends little support to the possibility that rivers in arid locations inherently transport more sediment annually than rivers elsewhere.
While the literature does not strongly suggest a dependence of sediment flux on climate, we can estimate the overall suspended sediment flux at each of our stations to confirm if our data demonstrate such a relationship.To do so, we approximated the annual specific suspended sediment yield or the total suspended sediment flux per unit watershed area over 1 year.We first predicted suspended sediment flux over time by multiplying each discharge record (taken every 15 min) by the suspended sediment concentration estimated from power law rating curves for each of the 71 gauges.(Notably, standard rating curves without a discharge cutoff were used so as not to severely misrepresent sediment transport at low flow-see "SSC-Q Plots" in Supplement.)We then integrated the fluxes with respect to time and divided by drainage area to determine the annual specific suspended sediment yield (kg yr −1 km −2 ), which we use as an approximation of the total sediment supply.
Power law regression revealed no significant relationship between aridity and specific suspended sediment yield at our study sites (R 2 = 0.021; p = 0.22) (Figure 5).Of course, not all proposed models in the literature are power functions, and perhaps a different curve shape would provide a better fit to our data below; a U-shaped curve similar to that of Fournier (1959), for example, may be more appropriate.Regardless of the chosen fit, our analysis demonstrates that the suspended sediment yield at arid locations is not meaningfully greater than that at temperate locations.These results therefore contradict the possibility that fluvial sediment loads are higher in our arid locations than the temperate ones.
Moreover, returning to the distinct trends we have observed in sediment rating curves, we see that overall sediment flux also offers limited ability to predict suspended sediment rating curve behavior for our sites: regression analyses showed no apparent relationship between specific sediment yield and either the rating curve exponent, a (R 2 = 0.001; p = 0.81) or coefficient, k (R 2 = 0.03; p = 0.17) (Figure 6).This suggests that sediment yield is also not particularly relevant when discussing the efficiency of sediment transport.
As is evident from our analyses and the established literature, fluvial sediment flux and the supply of sediment to rivers in this study must be examined with more nuance; our work has indicated clear differences in rating curve behavior across climates, but there is no compelling evidence for a significant relationship between the sediment yield and either climate or rating curve behavior.Without any obvious differences in overall sediment loads across climates, the only explanation for distinct suspended sediment rating curves must be the magnitude and frequency of individual sediment-bearing floods.Variability in a river's "typical" sediment transport event is certainly connected to climate; for example, rivers in arid climates may receive large influxes of sediment following a storm, but significant storm events are more infrequent overall than in temperate locations.This concept is supported by past work suggesting that rivers in arid regions generally have higher mean annual total suspended sediment for a given specific runoff than humid regions (Ludwig & Probst, 1996), and again points to efficiency as a key difference in the nature of suspended sediment transport across climates.
There are multiple climate-related variables that affect the magnitude and frequency of suspended sediment transport and could also explain the distinctive behavior of rating curves in this study.We review the most likely possibilities below.

Influence of Vegetation Cover on Sediment Supply
The simplest explanation for high fluvial suspended sediment concentrations in arid regions, particularly at low to moderate discharges, is a greater supply of fine sediment to these rivers at those flow conditions in particular.Sediment supply differences could be related to vegetation in the watershed; soil erosion is highly dependent on vegetation density, given that plants provide greater soil cohesion, increase interception and infiltration, and reduce overall runoff (Gyssels et al., 2005;Loch, 2000;Morgan, 2005).Indeed, vegetation removal is often associated with increased sediment supply to rivers through gradual erosion and shallow landsliding, as well as more frequent sedimentation events due to the lower threshold for significant erosion to take place (Istanbulluoglu et al., 2004;Warrick et al., 2013;Ziemer, 1981).Thus, although the relationship between mean annual precipitation and erosion rates is complicated both in the short and long term (Carretier et al., 2013(Carretier et al., , 2018;;Walling & Kleo, 1979), overall aridity may influence vegetation density in such a way that sediment supply to rivers is significantly different across climatic regimes.This was, in fact, a key component of the Langbein and Schumm (1958) model, which attributed an observed decrease in sediment yield at high precipitation rates to increasing vegetation density.The lower vegetation density in arid landscapes could allow for significant mobilization of sediment during even relatively modest runoff-generating storms, while much larger and more infrequent storms are required to move sediment in densely vegetated temperate landscapes.To investigate the potential role of vegetation in our data, we first estimated vegetation density in each of the 71 study watersheds using the Normalized Difference Vegetation Index (NDVI).We began with a 1-m resolution, 4-band aerial imagery collection acquired during agricultural growing seasons, provided by the USDA National Agriculture Imagery Program (NAIP) (U.S.Department of Agriculture, 2022).Using growing season imagery is preferable as it accurately reflects the presence of vegetation; because NDVI is essentially a measure of vegetation "greenness," estimating an annual mean would greatly underestimate the density of vegetation in deciduous-dominated areas.With this in mind, we used Google Earth Engine to calculate NDVI from NAIP imagery for the years 2010-2020 using the near-infrared (NIR) and red (R) bands: Notably, open water (e.g., lakes) should not be considered when evaluating the effects of land vegetation density on soil cohesion in watersheds, so we filtered out cells with NDVI values less than −0.5, representing locations that are almost certainly water.Finally, for each raster cell, we calculated the mean NDVI across all years, and then determined the mean NDVI value for all cells within each watershed using polygons provided by the GAGES-II dataset (Falcone, 2011).
First of all, we find a strong relationship between Aridity Index and NDVI (R 2 = 0.69; p < 0.001) (Figure S4 in Supporting Information S1).This confirms our assumption that vegetation density should be relatively lower in arid locations, potentially contributing to greater sediment mobilization during storms.
Next, the relationship between vegetation density and sediment transport can be examined by plotting the growing season NDVI against both the coefficient (k) and exponent (a) of the suspended sediment rating curves (Figure 7a).Regression analysis demonstrates a weak positive relationship between mean NDVI and the exponent (R 2 = 0.13; p = 0.003) and a moderate-to-strong negative relationship between mean NDVI and the coefficient (R 2 = 0.40; p < 0.001).
The observed relationship between vegetation density and rating curve behavior suggests that sediment supply is a potential control on suspended sediment transport efficiency between climates.In general, rivers with denser vegetation in their watersheds behave similarly to more temperate (higher Aridity Index) locations, with low rating curve coefficients and high exponents.This may explain the aforementioned behavior of sediment transport in temperate rivers, where suspended sediment concentrations are low during low-flow conditions but rise rapidly with discharge at higher flow; where there is abundant vegetation, only large and infrequent storms may be able to produce significant hillslope erosion (East et al., 2018).By contrast, in arid regions where vegetation is sparse or patchy, even relatively small storm events mobilize a significant amount of sediment from the surrounding landscape.This is particularly visible in individual suspended sediment plots from hyper-arid regions, where suspended sediment concentrations are often low or scattered at the lowest flow conditions but rapidly rise to sustained high concentrations across both moderate and high discharges (e.g., Arroyo Chico and Paria River: Figures 2b and 2f; Polacca Wash, Rio Puerco near Bernardo, Oraibi Wash: "SSC-Q Plots" in Supplement).
In summary, our results suggest that differences in vegetation density-and thus erosional thresholds-can explain why high rates of sediment transport occur at relatively lower discharges in arid locations than temperate ones.This is wholly distinct from the effect of vegetation on long-term suspended sediment flux, as there is no significant relationship between NDVI and the annual specific suspended sediment yield based on a power law fit (R 2 = 0.002; p = 0.70) (Figure S5 in Supporting Information S1).It is more likely that vegetation density influences the magnitude and frequency of sediment supply, thereby influencing overall patterns in suspended sediment transport efficiency across climates.

Flow Variability and Armoring
Another key climate variable that may influence suspended sediment transport efficiency is variability in precipitation, runoff, and streamflow.While mean annual precipitation has a major influence on overall aridity, the magnitude and frequency of individual precipitation events throughout the year are also quite significant.For instance, regions with moderate but frequent storms year-round will generally be less arid than locations with only very large, intense storms restricted to a few months.Indeed, across the United States, runoff variability is correlated with Aridity Index, with arid locations experiencing fewer intermediate-frequency runoff events than temperate locations (Rossi et al., 2016). 10.1029/2023JF007429 13 of 18 Differences in precipitation and runoff variability can affect fluvial sediment transport in multiple ways.For one, precipitation variability and vegetation density co-vary, such that regions with strong seasonal precipitation are typically less vegetated (Hooke, 2000;Lotsch et al., 2003).This can lead to higher sediment yields from the surrounding landscape in the ways we have described above; a less-vegetated watershed will likely have less soil cohesion and reduced infiltration, leading to greater runoff-driven erosion.
However, precipitation variability may also contribute to greater sediment transport efficiency not solely through increased sediment supply but also through its effect on the sorting of bedload sediment.Flow variability is a major control on whether a riverbed develops "armor"-a coarse, less mobile surface layer above a finer subsurface-which arises due to selective transport of fines under both low-level baseflow and longer flood recessions (Hassan et al., 2006).This could also explain the suspended sediment rating curve behavior we observe in this study, as fine sediment on a loose, unsorted bed surface can quickly be brought into suspension even by relatively low flows, thereby contributing to a relatively high, flatter rating curve (large coefficient, small exponent) (Asselman, 2000).Arid streams, where flow is more infrequent and variable, generally lack significant armor and thus have easily mobilized beds, leading to high sediment transport rates across all flow strengths.Meanwhile, temperate streams where armor is more developed have lower suspended sediment transport at low flow conditions and require streamflow to reach some threshold discharge before bedload is mobilized.After this point, suspended sediment concentrations increase quickly, reflected by a steep rating curve (low coefficient, high exponent) (Asselman, 2000).As an example of this climate contrast, Reid and Laronne (1995) compared  (Baker et al., 2004).Regression equations and the corresponding R 2 and p-values are shown in each plot.
10.1029/2023JF007429 14 of 18 rates of bedload transport between an ephemeral stream in Israel and a perennial stream in Oregon, claiming that the exceptionally high sediment flux in the ephemeral stream was related to a lack of a coarse armor layer on the riverbed.To the extent that suspended sediment is sourced from the bed of a river, the development of bed armor will strongly influence the mobility of the fine fraction (e.g., Wilcock & Crowe, 2003) and can thus plausibly explain the differences in suspended sediment mobility in arid and temperate settings that we document here.
In lieu of field measurements of surface and subsurface grain size (i.e., armor) at our study sites, we turned to flow variability to test whether bed surface grain size sorting could contribute to the climatic patterns we see in suspended sediment transport, as flow variability is a major control on armor formation (Hassan et al., 2006).We quantified flow variability at each site by calculating the Richards-Baker Flashiness Index (RBI), which compares the sum of incremental changes in discharge (Q, m 3 /s) to the total discharge across various timesteps, thus providing a non-dimensional index of how rapidly discharge changes over time (Baker et al., 2004): We used mean hourly discharges, calculated from continuous USGS streamflow data (U.S. Geological Survey, 2022), to determine the Richards-Baker Flashiness Index for each one of our gauge locations, limited to the time span of suspended sediment data used in this study.While Baker et al. (2004) predominantly use mean daily discharge to calculate RBI, they note that the chosen interval depends on the application.In this case, hourly averages of discharge are more appropriate because many flash floods in arid locations in our study area occur over the span of hours and minutes rather than days; mean daily discharge would thus likely dampen the signal of these brief storm events.
Regression analysis indicates that increases in RBI are generally associated with decreases in the exponent (R 2 = 0.16; p < 0.001) and increases in the coefficient (R 2 = 0.33; p < 0.001) (Figure 7b).That is, flashier streams parallel the behavior of more arid streams, with lower rating curve exponents and higher coefficients, though interestingly there is not a clear relationship between the Aridity Index and the Richards-Baker Flashiness Index (R 2 = 0.04; p = 0.09) (Figure S6 in Supporting Information S1).In short, the relationship between flashiness and the two rating curve parameters suggests that flow variability could also play a role in determining sediment transport efficiency, potentially through its effect on riverbed armoring.
It is worth noting that riverbed armoring is primarily a bedload process, and we have examined suspended sediment in this study.However, the two processes are closely linked.If a riverbed is unarmored, suspended sediment transport and bedload transport will be strongly correlated, partly because fine sediment can be sourced from the bed of the channel, and highly armored streams should lack significant suspended sediment concentrations or bedload transport rates at low flow conditions.Furthermore, a large supply of fine sediment from the watershed causes a finer channel bed, and thus high rates of transport in both the suspended load and bedload (Parker & Klingeman, 1982).Finally, the amount of fine sediment in alluvial channels has been shown in some settings to influence the movement of the coarse fraction (Dietrich et al., 1989;Iseya & Ikeda, 1987;Parker, 1990;Wilcock & Crowe, 2003), so high suspended sediment concentrations can also facilitate high bedload transport rates in streams.Thus, differences in flow variability between rivers in contrasting climates could contribute to similar differences in armoring, which may explain the distinctive behavior of suspended sediment rating curves.

Relative Roles of Vegetation and Flow Variability
Given the similar strength of the relationships between rating curve parameters and the two variables we examine here-vegetation density and flow flashiness-it is not immediately apparent which property best explains the connection between climate and sediment transport efficiency.While the two variables are only weakly correlated (R 2 = 0.10; p = 0.008) (Figure S7 in Supporting Information S1), the complicated relationships between hydrologic, ecologic, and geomorphic processes may make them impossible to disentangle.As noted before, precipitation variability influences vegetation density in a watershed, which in turn controls sediment supply to channels, but that variability also affects the riverbed armor formation, which itself can change the efficiency of sediment transport.Moreover, riverbed armoring is also largely determined by sediment supply (Hassan et al., 2006); a river cannot winnow fine grains from the bed if there is a large input of fine sediment to the channel.Therefore, while riverbed armoring may explain a component of the documented contrasts in fluvial sediment transport efficiency between arid and temperate locations, differences in grain size sorting could also reflect local flow variability or vegetation-driven sediment supply, both of which relate to climate.The complex relationship between aridity, flow variability, armoring, vegetation, and sediment transport is summarized in Figure 8.
Most likely, the compounded effects of both vegetation and flashiness contribute to differences in sediment transport efficiency, rather than any one parameter dominating.Indeed, the Aridity Index is a stronger individual predictor of rating curve parameters than either NDVI or flashiness (Figures 3 and 7), implying that sediment transport is determined by multiple factors that all depend on-or contribute to-the overall climate.
While vegetation density and precipitation variability are probably the principal factors controlling suspended sediment transport efficiency, there could be other variables that we did not account for.It seems likely, however, that any other climatic variables would be inseparable from either rainfall or vegetation patterns, in the same way that these two variables are interconnected themselves.For example, Syvitsky et al. (2000) found covariance between rating curve parameters and factors such as mean annual temperature and latitude, variables that should not intuitively influence sediment delivery to rivers.Additional analyses of climate proxies and sediment transport may contribute to a more complete understanding of sediment transport across different climates, but there is most likely no specific variable more responsible for the results of this study than the overall climate itself.

Conclusions
In this contribution, we examine suspended sediment concentrations and continuous flow records from 71 U.S. Geological Survey gauges across North America, finding that increasing aridity coincides with higher coefficients and lower exponents in power-law relationships between discharge and suspended sediment concentration.This rating curve behavior demonstrates that rivers in arid locations generally transport sediment more efficiently than those in temperate regions.Arid streams transport a large volume of suspended sediment across a wide range of discharges, including low to moderate flow conditions, while suspended sediment transport in temperate streams is minimal at low discharges, but scales quickly as the discharge increases.
We attribute the contrast in sediment transport efficiency mainly to differences in vegetation density and precipitation patterns, which in turn affect both sediment supply and streamflow variability.Based on regression analyses, we interpret the relationship between climate and sediment transport behavior as likely due to the combined effect of multiple climatic factors, some of which we may not have considered.Notably, further analyses of specific suspended sediment yield reveal no clear relationship between climate and overall suspended sediment loads, suggesting that the distinct patterns in sediment rating curves are a result of differences in how and when sediment transport occurs across climates.
These results are highly relevant to risk management, particularly with respect to flood hazards and infrastructure damage.Variability in river bed elevation and flood conveyance in the United States is particularly high in arid settings (Slater & Singer, 2013;Slater et al., 2019), leading to major erosion and sediment deposition problems near channels (e.g., Merritt & Wohl, 2003;Wilcox et al., 2016).Our results lend credence to these observations and suggest some mechanistic explanations.Moreover, these hazards are likely to be exacerbated by anthropogenic climate change; presently, new non-perennial and flashier streams are emerging throughout the southern and southwestern US (Overpeck & Udall, 2020;Zipper et al., 2021), and climate change has also led to shifts in vegetation distribution and the frequency of extreme rainfall events and droughts, which have profound consequences for streamflow, river morphology, and sediment transport (East & Sankey, 2020).It is thus increasingly important to understand the inherent properties of rivers, which may shed light on how climate change will affect sediment transport behavior and natural hazard risks.

Figure 1 .
Figure 1.Map of the United States with the final set of 71 USGS gauges in this study colored by Aridity Index (Zomer et al., 2022).HCDN-2009 sites (Falcone, 2011) are shown as circles, and sites from the expanded Xeric West dataset are shown as diamonds.

Figure 2 .
Figure 2.Several example power-law fits between discharge and suspended sediment concentration are organized in columns by aridity index (AI).For some sites, no discharge cutoff was applied (a, b), but in most cases, some low-discharge measurements were excluded from rating curve calculations due to data resolution issues and the goal of capturing sediment transport behavior at moderate to high flows (when most transport occurs).Calculated rating curves sometimes underestimated (c, d) or overestimated (e, f) suspended sediment concentrations at excluded points.Corresponding information about the power law regression (red line) is also shown, including the equation, N, R 2 , and p-value."Cutoff" values in plot titles refer to the upper quantile of data points (by discharge) used in the regression.

Figure 3 .
Figure3.Aridity index versus two rating curve parameters for 71 stream gauges in the United States.Rating curve parameters, a and k, refer to the exponent and coefficient (respectively) of a power-law fit between discharge (Q; m 3 /s) and suspended sediment concentration (SSC; mg/L) in the form SSC = kQ a (Figure2).Log axes were used when a particular variable was more normally distributed in log space.Linear regressions in semi-log (above) and log-log (below) space are also shown as red lines, with corresponding equations listed in red text along with an R 2 and p-value for the regression.More arid locations occur to the left of the plots, while more temperate sites tend toward the right.

Figure 4 .
Figure 4. Results of suspended sediment rating curve analysis.(a) Coefficients and exponents of power-law rating curves between discharge and suspended sediment concentration, colored by Aridity Index (AI) with more arid sites in red and more temperate sites in blue.The coefficient and exponent pairs for the 10th and 90th percentile AI, estimated using regressions between AI and the two rating curve parameters, are labeled with red and blue crosses, respectively.(b) Representative rating curves between discharge and suspended sediment concentration based on coefficients and exponents estimated at the 10th (arid) and 90th (temperate) percentile Aridity Index (AI), shown as crosses in Panel (a).Sediment concentrations are plotted on the highest discharge represented in this dataset, and cut off at 0.01 m 3 /s as an arbitrary low value (0 m 3 /s cannot be shown on a log scale).

Figure 5 .
Figure 5. Power law regression between Aridity Index and the annual specific suspended sediment yield for the 71 sites in this study.Regression equation, R 2 , and p-value are shown in the bottom right.

Figure 6 .
Figure6.Plots of estimated annual specific suspended sediment yield versus power-law rating curve parameters a and k for the 71 USGS gauges used in this study.Linear regressions (in semi-log and log-log space, respectively) do not demonstrate a compelling relationship between sediment yield and rating curve behavior for the sites in this study.

Figure 7 .
Figure7.Regressions between rating curve power-law parameters (exponent, a, and coefficient, k) and two possible environmental predictors of sediment transport behavior: (a) Normalized Difference Vegetation Index (NDVI; left) and (b) flow flashiness via the Richards-Baker Flashiness Index (RBI; right)(Baker et al., 2004).Regression equations and the corresponding R 2 and p-values are shown in each plot.

Figure 8 .
Figure 8. Flow chart demonstrating the relationships between two key climatic variables (precipitation and vegetation) and sediment transport efficiency.