Temporal variability in irrigated land and climate influences on salinity loading across the Upper Colorado River Basin, 1986-2017

Freshwater salinization is a growing global concern impacting human and ecosystem needs with impacts to water availability for human and ecological uses. In the Upper Colorado River Basin (UCRB), dissolved solids in streams compound ongoing water supply challenges to further limit water availability and cause economic damages. Much effort has been dedicated to understanding dissolved solid sources, transport, and management in the region, yet temporal variability in loading from key sources such as irrigated lands and the influence of climate on dissolved solids loading remains unknown. Quantifying the contributions and temporal variability of dissolved solids loads from irrigated lands may benefit salinity management efforts. This study applies a time-varying (dynamic) modeling approach to predict annual dissolved solids loads across the UCRB from 1986 through 2017. Between 66% and 82% of the total accumulated dissolved solids load in the basin is from groundwater (storage and baseflow). Our findings link climate, irrigation, and groundwater, and confirm large storage contributions that have declined slightly with time. Dissolved solids loads increase during wet periods and decrease during dry periods, although the relative contributions of different sources vary little with time. Irrigation enhances loading efficiency relative to unirrigated areas through runoff and groundwater, and can locally be a major source of dissolved solids where irrigation occurs. Results indicate that loads from irrigated areas increase when irrigated area and/or water available for runoff increase. Increased regional aridification over the study period may have contributed to decreasing stream salinity through both quicker surface runoff and lagged groundwater storage processes. Study results may be relevant to salinity management in arid environments where water availability is limited and where irrigation influences salinity loading to streams.


Introduction
Recognition of the global threat that freshwater salinization poses to water availability for both humans and ecosystems is growing (Kaushal et al 2018, Cañedo-Argüelles 2020).Increasing salinity can limit water availability for humans through damage to infrastructure from corrosion, mobilization of lead and other pollutants into water supply systems (Stets et al 2018), poor taste, reduced agricultural returns, or causing water to be too saline for use.Salinization of freshwater ecosystems can alter ecosystem function and services, reduce biodiversity, and ultimately contribute to biological degradation of aquatic ecosystems via increased physiological stress and toxicity to aquatic species (Cañedo-Argüelles 2020 and references within).Moreover, multi-constituent interactions can enhance ecosystem stress and water quality degradation (Kaushal et al 2018, Cañedo-Argüelles 2020).Salinity in freshwater systems has both natural and anthropogenic sources.Natural sources include dissolution of saline rock formations Not subject to copyright in the USA.Contribution of U.S. Geological Survey.and discharge from saline springs (Anning and Flynn 2014).Anthropogenic sources include deicers, and land use such as irrigation, urbanization, and pasture/rangeland (Anning and Flynn 2014).While agriculture is a recognized source of salinity in many areas, the impacts and loads are not well quantified relative to other sources at most temporal and spatial scales (Thorslund et al 2021).
Salinity has long been recognized as a problem in the Colorado River Basin for both the United States and Mexico and is estimated to cause economic damages to infrastructure and crop production of $300-$400 million per year (Colorado River Basin Salinity Control Forum 2020).To address Colorado River salinity challenges, the United States and Mexico developed treaty obligations for the United States to deliver water of specified salinity to Mexico (United States and Mexico 1973).Basin states (Arizona, California, Colorado, Nevada, New Mexico, Utah and Wyoming) established the Colorado River Basin Salinity Control Forum in 1973 to facilitate cooperation among states and federal agencies to comply with the Clean Water Act.In 1974 the U.S. Environmental Protection Agency established a basinwide salinity control policy which included salinity concentration criteria and implementation plans to protect against increased economic damages to infrastructure and crop production (Colorado River Basin Salinity Control Forum 2020).Salinity control measures have focused on improving irrigation delivery systems, reducing point-source loads, and reducing soil erosion (Colorado River Basin Salinity Control Forum 2020).
Primary sources of salinity, or dissolved solids, have been characterized in the Upper Colorado River Basin (UCRB, figure 1).A multidecadal average evaluation estimates that natural sources, including the dissolution of minerals in soils and rocks, contribute the most (62%) to the approximately 6.4 million tons/year of dissolved solids delivered to the outlet of the UCRB, followed by irrigated agricultural lands (32%), and saline springs (6%) (Miller et al 2017).Much of the natural salinity comes from erosion and dissolution of sedimentary formations deposited in marine, lacustrine, or brackish environments including marine shales, salt anticlines, and coal-bearing formations (Liebermann et al 1989, Miller et al 2017).Baseflow (groundwater flowing to streams) is a key pathway for generating and delivering an estimated 89% of the dissolved solids loads to streams in the basin (Rumsey et al 2017).Climate and land use/cover can affect dissolved solids loading to streams (Kenney et al 2009, Miller et al 2017).Irrigated lands deliver greater salinity loads relative to their land area (yield) compared to geologic sources, though geologic sources contribute more to the total load due to their greater total area within the basin (Miller et al 2017).Salinity regimes and their primary predictors have also been classified (Bolotin et al 2022).
Over the past century, salinity in tributaries of the UCRB has declined significantly (Rumsey et al 2021).Much of the decline has been attributed to landscape processes, which may include changes to irrigation practices or climate as well as land cover and land use.However, specific mechanistic drivers (and their temporal variability) for these trends have not been identified nor have contributions from known salinity delivery mechanisms, like irrigation, been quantified over time.
Prior work has established a quantitative understanding of the spatial distribution and sources of salinity loads in the UCRB under long term average conditions (Kenney et al 2009, Miller et al 2017) and identified century-scale declines in basin loads at gages (Rumsey et al 2021), yet no study has quantified temporal variability in total and source-specific loads across the basin.This study quantifies salinity loading to UCRB streams from irrigated lands and other sources over time by developing a model of the temporal variability of dissolved solids sources and transport across the UCRB.Application of a dynamic modeling framework supports exploration of the processes driving temporal variation in dissolved solids loading.Quantification of salinity contributions over time from factors that are known to influence salinity loading in the basin, including the role of irrigated agriculture and interactions with climate, may inform continued management.Our modeling approach includes important storage pathways operating at timescales greater than a year that have not been previously considered.Understanding temporal variability of salinity loads may allow salinity managers to identify effective salinity control measures, quantify control-project reductions in loads, and to project possible load ranges in the future.

Methods
To quantify temporal and spatial variability in dissolved solids loads, sources, and transport in the UCRB, we developed a dynamic Spatially Referenced Regression on Watershed attributes (SPARROW, Schwarz et al 2006, Smith 2012, Smith et al 2014, Schmadel et al 2021) model that accounts for temporal variability of salinity sources and processes that transport salinity from the landscape to streams, as well as lagged storage.SPARROW models track constituent sources and transport through large basins and have been widely applied to predict water quality and quantity at regional to continental scales and to quantify factors that control the spatial variation in loading to streams (Preston et al 2011, Ator 2019, Hoos and Roland 2019, Robertson and Saad 2019, Wise 2019, Wise et al 2019).We used the dynamic model to predict annual dissolved solids loads, with source attribution, at 10 789 reaches in the UCRB for water years 1986-2017.Sources of uncertainty for the model and results include potentially not including SPARROW is a spatially explicit, hybrid statistical and process-based model that estimates constituent loads (dissolved solids here) in streams by linking monitoring data with information on watershed characteristics and constituent sources, and routes loads through a stream network.SPARROW's hybrid features improve the accuracy of predictions and support process interpretations (Alexander et al 2019).Non-linear least squares regression with mass balance constraints are used to characterize the spatial relationships between estimated constituent loads and variables representing constituent sources and transport processes.Those relationships are used to predict constituent loads at ungaged reaches throughout the stream network (Schwarz et al 2006).
We adapted the existing UCRB long-term average dissolved solids model (Miller et al 2017, see supplementary information (SI) 1.1 for model description) to estimate sources and transport of annual dissolved solids loading for all reaches in the basin from water years 1986 to 2017.Adaptation of the existing model assumes that the same sources and processes that control salinity over the long term apply on an annual basis.This assumption is justified based on prior work evaluating salinity sources (Kenney et al 2009, Miller et al 2017), but the transition to an annual time step permits exploration of additional processes as we demonstrate here.We applied 32 years of annual estimates of climate, irrigated land area, and dissolved solids loads to the model, and added lagged storage within a dynamic modeling approach described in Smith (2012), Smith et al (2014), andSchmadel et al (2021) to predict time-varying dissolved solids loads.This annual approach does not explicitly represent changes on a seasonal time step.
Long-term average SPARROW models assume that the change in dissolved solids in storage is zero (Schwarz et al 2006) whereas dynamic simulation of dissolved solids loading accounts for mass lagged in storage, including in soil and groundwater, that delays dissolved solids loading to streams (Smith 2012, Smith et al 2014).Storage is represented by a simple, recursive, one-period lag specification that includes storage occurring at timescales longer than a year.Although the mass in storage is unknown, the flux of mass (L s,t,i ) released from storage (s) for each catchment (i) per time period (t) is estimated as: where α s is a coefficient that represents the average fraction of load delivered to the stream from storage, L t−1,i is the load delivered to the stream in the previous time step, and f s,t,i is a land to water delivery function that uses explanatory variable data (see Schmadel et al 2021 for complete explanation).
The annual dynamic dissolved solids model was calibrated using datasets described below to predict spatial and temporal variability in dissolved solids sources and transport.Many different model configurations were tested, and the final configuration of variables and settings was selected based on maximizing the yield, load, and adjusted R 2 , minimizing the overall model root mean square error, inclusion of variables with significant p-values (p-value < 0.1), and residual plots.Model fit statistics are based on comparisons between observed and predicted values in natural log space.

Data description and processing
SPARROW models require estimates of salinity loads at points along the stream network and spatial datasets of explanatory variables that represent likely sources and landscape transport processes (which interact with sources to enhance or reduce salinity loading).Model inputs include time-constant salinity sources (geology, saline springs, and baseflow dissolved solids loads) and land-to-water delivery processes (slope and stream network).In addition to the time-constant data, we applied time-varying input data including annual estimates of dissolved solids loads (calibration data), discharge, climate (precipitation and evapotranspiration), and irrigated area to the dynamic salinity model to estimate the relationship between explanatory variables and estimated loads at points, and predict annual dissolved solids loads and source-specific loads across the basin.

Time-constant source and landscape transport data
The generalized geologic categories (figure 1(a)) and saline spring loads previously used in (Buto et al 2017, Miller et al 2017, 2022) were used to represent two natural sources of salinity to waterways in our model.These are held constant as geology does not vary over short timescales, and temporally distinct spring salinity records are limited (this is a source of uncertainty in the model).
Predicted baseflow dissolved solids loads were used as another source of mixed natural and anthropogenic salinity in the dynamic dissolved solids model.A long-term average baseflow dissolved solids model was developed based on the existing longterm average dissolved solids model for total streamflow dissolved solids loads (Miller et al 2017) to predict spatially distributed baseflow dissolved solids loads to UCRB stream that were used as a source in the dynamic stream dissolved solids SPARROW model.The long-term average total streamflow salinity model was adapted in the following ways: 1) springs and irrigated agriculture were not included as sources of baseflow salinity, 2) catchment mean elevation was not included as a landscape transport variable, 3) predicted loads were adjusted to correspond with estimated loads at calibration sites, 4) all stations were assigned least squares weights of 1, and 5) the model was calibrated using estimates of baseflow dissolved solids loads (see SI 1.2).Sources in the baseflow dissolved solids model include generalized geologic sources, and total area of irrigated agricultural land (sum of flood-irrigated agricultural land on Mesozoic and other lithology and sprinkler irrigated lands on any lithology) on any lithology (Miller et al 2017(Miller et al , 2022)).Transformations were applied to available water, slope, and iron oxide concentrations to approximate normal distributions.Available water, slope, soil thickness, soil iron oxide concentration, and percent range land interacted with all geologic sources.This approach captures the spatial distribution of baseflow salinity loads, but neglects temporal variability and assumes less fluctuation than surface water loads.Assuming that baseflow salinity changes more slowly than surface water, this approach, while simple, may approximate baseflow patterns within the model.Moreover, interactions with climate variables (described below) may improve this approximation by introducing temporal variability linked to climate.Catchment mean slope was used as a time-constant variable that influences the transport of dissolved solids from the watershed to streams (Buto et al 2017).Long-term average incremental dissolved solids loads (Miller et al 2017(Miller et al , 2022) ) were used as the initial storage values in the dynamic model.

Time-varying dissolved solids load estimation
Annual dissolved solids loads at gages were estimated using daily streamflow and discrete dissolved solids concentration data from the U. S. Geological Survey's National Water Information System (U.S. Geological Survey 2021), following methods for dissolved solids load estimation described in Rumsey et al (2021).Discharge and dissolved solids concentration data were applied to Weighted Regression on Time, Distance, and Season (Hirsch et al 2010) to estimate annual dissolved solids loads at 189 sites throughout the basin (see SI 1.3).Loads were used as calibration data for the dynamic dissolved solids model.

Time-varying source and landscape transport data
Irrigated land area from the IrrMapper dataset (Ketchum et al 2020) was used to represent irrigated lands (see SI 1.4).Loads from irrigated areas differ based on the underlying geology (Miller et al 2017).Therefore, the area of irrigated land on both Mesozoic and other lithologies from Buto et al (2017) for each catchment was calculated.
Catchment-total or mean annual precipitation, actual evapotranspiration, and runoff were obtained from Milly and Dunne (2022).Variables were generated using a monthly water balance model (McCabe and Markstrom 2007) that was modified to improve representation of snow processes, as described in Milly and Dunne (2020).

Time-varying discharge
To estimate reach mean streamflow for each year, runoff from Milly and Dunne (2020) was accumulated through the stream network and bias corrected using mean annual streamflow from gages throughout the basin.Discharge was adjusted using a correction factor based on the ratio of the measured and estimated discharge (see SI 1.5 and 1.6).

Predicted loads
Incremental and total dissolved solids loads were estimated for each of the 10,789 reaches for every year in the UCRB.Incremental load is the load generated within each catchment and transported to that catchment's outlet.Total load is the sum of the incremental load generated within a catchment plus the load delivered from upstream catchments.Basin totals were calculated as sums of all incremental loads in the basin.Sources of loads were tracked through time and space to quantify the contribution of irrigated lands to dissolved solids loads.Correlations among variables were assessed using Kendall correlation tests (Kendall 1938) due to the non-normal distribution of many variables.Kendall's Tau statistic is a rankbased measure of the strength of the monotonic relationship between two variables and is appropriate for non-normally distributed data (Helsel et al 2020).

Results
Model statistics and diagnostics plots indicate a good fit for both the streamflow and baseflow dissolved solids models to the calibration data (tables 1, 2 and SI figure 1).The load R 2 values of 0.96 for both the streamflow dissolved solids and baseflow dissolved solids models indicate that the models explain 96% of the variability in dissolved solids loads.The yield R 2 , which removes area-load correlations and is therefore a more useful measure of model performance, was 0.71 and 0.70 for the stream dissolved solids and baseflow dissolved solids models, respectively.Root mean square error values for the streamflow dissolved solids and baseflow dissolved solids models were 0.49 and 0.51 respectively, indicating that the average error in model predictions (associated with one standard deviation of the model error) was 49% and 51%, respectively.The residuals are normally distributed for both models (SI figure 1).
The best-performing dynamic dissolved solids model (the model hereafter) tracks and quantifies contributions from springs, generalized geology, irrigated land on Mesozoic and other lithologies, baseflow, and storage sources (table 1).Catchment available water (precipitation minus evapotranspiration), the previous year's available water, and catchment mean slope represent transport and delivery of dissolved solids to streams in thFe model.Available water interacts with all sources except springs (point sources) and storage terms.The prior year's available water interacts with all sources except springs and the initial storage.Slope interacts with all sources except springs, irrigated lands, and storage terms.Springs and baseflow do not influence storage as they contribute directly to the stream.
The long-term average baseflow dissolved solids model includes geologic and irrigated lands sources (table 2).Delivery transport variables including   available water, slope, soil thickness, rock iron oxide concentration, and fraction of rangeland coverage interact with these sources to enhance or reduce baseflow dissolved solids loads.High-yield lithologies have higher coefficients than low yield lithologies, and irrigated lands have the highest coefficient.
All delivery transport variables have positive coefficients except soil thickness.
The baseflow dissolved solids model load predictions represent long-term average loads to streams.Because available water interacts with these average loads in the model, the baseflow source represents quick-flow groundwater discharge that is influenced by within-period climate.The sum of the dissolved solids loads from storage and baseflow (defined here as groundwater loads that take more than or less than one year to discharge to streams, respectively) represents the total load to streams from groundwater.Predicted contemporaneous loads from geologic units, irrigated lands, and springs are representative of loads generated and transported (via runoff) or discharged to streams within one year.

Sources of dissolved solids
Across all 32 years, irrigated areas on Mesozoic lands had the highest coefficient and are therefore the most efficient source of salinity to streams (88 metric tons per square kilometer per year), followed by irrigated areas on other lands (45 metric tons per square kilometer per year, approximately 50% lower yields than irrigated areas on Mesozoic lands, table 1).Geologic source coefficients were lower still, and range from 0.33 to 3.50 metric tons per square kilometer per year, with the high-yield Paleozoic and Mesozoic rocks having the highest coefficients and therefore highest yields.The dimensionless coefficient for springs (point sources) was 0.81, indicating that the estimated loads from springs used as input to the model may be overestimated, possibly due to the lack of temporal variability in this variable.
Although irrigated lands have the highest yields relative to geologic sources (table 1), long-term storage (34%-61%) and quick-flow baseflow (20%-34%) contribute the largest fraction of dissolved solids loads in the basin (figure 2) due to their greater spatial coverage relative to irrigated lands.For all years, contemporaneous loads from springs and geologic sources contribute between 6%-12% and 6%-10% respectively to the total load generated in the basin each year.Contemporaneous irrigated lands on Mesozoic lithologies contribute 3%-9% while contemporaneous irrigated land on other lithologies contribute 2%-6% of the total contemporaneous load generated over the study period (figure 2).These results clearly indicate significant contributions from irrigated lands.However, their total contributions are likely underestimated here because some proportion of the high-salinity waters from irrigation and geologic sources enter groundwater and remain in storage for a year or more before entering streams.Relative contributions of individual sources to storage cannot be explicitly separated though because storage lumps all contributing sources.

Spatial and temporal variability
Dissolved solids total loading and loading from irrigated lands varies with time in the UCRB (figure 2), with the highest total loading occurring in the early years and during wetter years for both total and irrigated land sources.Similar to long-term average assessments, irrigated lands on Mesozoic lithologies generate greater loads than irrigated lands on other lithologies consistently through time (figure 2).Irrigated lands generate more dissolved solids during wet years than dry years (figure 2) like non-irrigated lands, suggesting that potential reductions in irrigation during wet years do not substantially reduce loads from irrigated areas.
The relative contributions of sources to the total load varies by subbasin (figure 3).For example, saline springs are sometimes the greatest dissolved solids source in the Colorado Headwaters subbasin, and in other subbasins, springs are a minor contributor.In the Upper Colorado-Dolores, White Yampa, and Lower Green subbasins, baseflow and storage contribute similar loads to streams.Conversely, in the Gunnison, Upper Colorado-Dirty Devil and San Juan subbasins, baseflow contributes much less than storage.In the Great Divide-Upper Green, baseflow is often a greater relative source than storage.For all years, contemporaneous loads from irrigated land runoff contribute between 0% and 18% of the total incremental load of each subbasin, with median loads less than 10% (figure 4).

Importance of groundwater and climate in dissolved solids loading
Between 66% and 82% of the total accumulated dissolved solids load (what would be measured instream) in the basin is from groundwater (storage and baseflow, figure 2), and this has declined slightly with time from the beginning of the study period.Assuming all storage occurs in groundwater, this range is within prior estimates of groundwater contributions to stream dissolved solids loads in the basin, which range between 30% and 93% (Warner et al 1985).Our estimate is slightly lower than a recent study by Rumsey et al (2017), which estimated that on average 89% of dissolved solids load originates from baseflow.One explanation for the different estimates may be that the Rumsey et al (2017) study focused on 69 individual sites whereas we quantify groundwater contributions across the whole basin.
Generally, dissolved solids loads increase with more available water for runoff and decrease with less available water (figure 5).Available water drives much of the interannual variability in basin-wide total loads and baseflow loads, as indicated by similar interannual patterns (figure 2) and a positive, significant correlation between the two variables (figure 5).The early to mid-1980s was a wet period in the basin (Lukas and Payton 2020), whereas a drought beginning in 2000 has persisted to the present (Williams et al 2020, 2022), and salinity loads align with these patterns.
Model results show three years (1986)(1987)(1988) have higher than expected total loads relative to the amount of water available for runoff (figure 5).We note that the residuals during these years are similar to residuals during other years (SI figure 1) and that higher loads were noted at many sites during this period, indicating that the high loads during this period are not a result of model initial conditions.Moreover, storage initial conditions represent long-term average loads and are therefore closer to mean values than maximum values for the dynamic predictions.The high loads in the late 1980s and gradual decline with time may be related to increased loading associated with more available water and enhanced dissolved solids accumulation in storage during several preceding wet years ( ˜1982-1985, NOAA National Centers for Environmental information 2022).Subsequent declines may be related to reduced loading associated with less available water and reduced release of dissolved solids from storage with time as climate dried.A transition from dry to wet years is associated with enhanced loading, whereas a transition from wet to dry years is associated with reduced loading, as indicated by coefficient signs on the available and previously available water terms.The loads generated from irrigated lands during the late 1980s are generally within the range of loads generated in all years (figure 5), suggesting that the high loads during these years are not caused by irrigation.
The relationship between climate and salinity loading demonstrated in model results provides evidence that the long-term reductions in salinity  Rumsey et al (2021).The mid-late 1990s features several wet years when salinity loads increased.However, after 2000 there was less available water for runoff during wet years as during wet years in the 1990s and available water is more consistent from year to year, resulting in reduced salinity increases during wet years and more consistent salinity loads with time (figure 5).

Importance of irrigation in dissolved solids management
Greater available water is also positively and significantly correlated with greater loads from irrigated lands (figure 5).However, the greater coefficients for irrigated lands relative to nonirrigated lands (geologic sources) indicate that irrigated lands are more efficient at generating dissolved solids loads to streams than unirrigated lands.This is consistent with prior work, which has identified irrigated agricultural lands as high-yield sources (Kenney et al 2009, Miller et al 2017).
Contemporaneous runoff contributions of irrigated lands to dissolved solids loading in the basin were relatively small, and dissolved solids from irrigated lands stored in groundwater take time to reach surfacewater.Therefore, salinity management efforts focused on surface runoff processes may not produce immediate results, but rather will take time to work through the groundwater system.Moreover, the interactions of climate, storage, and irrigation may be important considerations when planning salinity management; climate influences interannual variability of load generation, while the majority of the loads come from storage, and irrigation acts to increase loading efficiency of landscapes.

Future work
In addition to groundwater, storage could also occur in other places including reservoirs where precipitation of dissolved solids within reservoirs could reduce loads below dams (Deemer et al 2020).Reservoir effects have not been sufficiently quantified across the basin for inclusion in this model and therefore represent a source of uncertainty in our study.Quantifying and incorporating reservoir effects on salinity loads could improve source attribution and representation of transport processes in the model.The dynamic dissolved solids model uses predictions of baseflow dissolved solids that do not change with time as a source of salinity.Although groundwater tends to vary less with time than surface water, baseflow dissolved solids trends have been identified at some sites (Rumsey et al 2017).Therefore, incorporation of time-varying baseflow dissolved solids loads could improve model predictions.

Conclusions
The dynamic modeling approach applied here is a first-order approximation of storage processes that provides a fundamental step towards quantifying time-varying changes in dissolved solids loads across a large river basin.Groundwater, represented by quick-flow baseflow and lagged storage that includes contributions from all sources that pass through storage, contributes the most to stream dissolved solids loads.Climate is an overall driver of loading while irrigation practices can dramatically enhance loading effectiveness.Loads from irrigated lands peak during wet years and decline during dry years, which may have long term implications under continuing aridification of the region (Overpeck and Udall 2020) and the potential need to increase irrigation to sustain crops.Given the close connection between salinity loading and climate, future climate change impacts to basin hydrology, including runoff and groundwater contributions to streams (Miller et al 2021a(Miller et al , 2021b)), may also affect salinity loading to streams.
While this study is focused on the UCRB, results are relevant to salinity management in arid environments where water availability may be limited and where irrigation influences salinity loading to streams.Salinity managers in such areas may want to consider multiple transport pathways and timescales that result in variable near-and long-term effects on salinity loading in management planning.

Figure 2 .
Figure 2. Plots showing (a) sum of incremental contemporaneous loads from all sources, (b) sum of incremental contemporaneous loads from irrigated lands, (c) sum of available water (Milly and Dunne 2022), (d) sum of incremental contemporaneous loads by source as a percent of total load, and (e) sum of total accumulated load from storage and baseflow as a percent of total accumulated load with time for the Upper Colorado River Basin.

Figure 3 .
Figure 3. Plot showing sum of incremental contemporaneous loads by source with time as a percent of total load for subbasins of the Upper Colorado River Basin.

Figure 4 .
Figure 4. Boxplots of annual sums of incremental contemporaneous loads from irrigated land sources as a percent of total load for subbasins of the Upper Colorado River Basin.

Figure 5 .
Figure 5. Plots showing (a) available water compared to annual sum of incremental dissolved solids loads from all sources, (b) annual total area of irrigated lands compared to annual sum of incremental dissolved solids loads from all sources, (c) available water compared to annual sum of contemporaneous incremental dissolved solids loads from irrigated lands over Mesozoic and other lithologies, (d) annual total area of irrigated lands compared to annual sum of contemporaneous incremental dissolved solids loads from irrigated lands over Mesozoic and other lithologies, (e) available water compared to annual sum of incremental dissolved solids loads from baseflow, and (f) annual total area of irrigated lands compared to annual sum of incremental dissolved solids loads from baseflow for the Upper Colorado River Basin.

Table 1 .
Dissolved solids model coefficients and statistics.

Table 2 .
Baseflow dissolved solids model coefficients and statistics.
a Variance inflation factor.b Degrees of freedom.c Sum of squared errors.d Mean square error.e Root mean square error.
Alexander R B and Wolock D M 2011 SPARROW modeling to understand water-quality conditions in major regions of the United States: a featured collection introduction J. Am.Water Resour.Assoc.47 887-90 Robertson D M and Saad D A 2019 Spatially referenced models of streamflow and nitrogen, phosphorus, and suspended-sediment loads in streams of the Midwestern United States U.S. Geological Survey Scientific Investigations Report 2019-5114 74 Rumsey C A, Miller M P, Schwarz G E, Hirsch R M and Susong D D 2017 The role of baseflow in dissolved solids delivery to streams in the Upper Colorado River Basin Hydrol.Process.31 4705-18 Rumsey C A, Miller O L, Hirsch R M, Marston T M and Susong D D 2021 Substantial declines in salinity observed across the Upper Colorado River Basin during the 20th century, 1929-2019 Water Resour.Res.57 1-21 Schmadel N M, Harvey J W and Schwarz G E 2021 Seasonally dynamic nutrient modeling quantifies storage lags and time-varying reactivity across large river basins Environ.Res.Lett.16 095004 Schwarz G E, Hoos A B, Alexander R B and Smith R A 2006 The sparrow surface water-quality model: theory, applications and user documentation U.S. Geological Survey, Techniques and Methods 6-B3 1-248 (available at: https://pubs.usgs.gov/tm/2006/tm6b3/) Smith R A 2012 Use of dynamic SPARROW modeling in characterizing time-lags in nitrogen transport in the Potomac River Basin Scientific and Technical Advisory Committee, Chesapeake Bay Program Smith R A et al 2014 Use of MODIS data in dynamic SPARROW analysis of watershed loading reductions AGU Fall Meeting Abstracts vol 2014 pp H33-08 Stets E G, Lee C J, Lytle D A and Schock M R 2018 Increasing chloride in rivers of the conterminous US and linkages to potential corrosivity and lead action level exceedances in drinking water Sci.Total Environ.613-614 1498-509 Thorslund J, Bierkens M F P, Oude Essink G H P, Sutanudjaja E H and van Vliet M T H 2021 Common irrigation drivers of freshwater salinisation in river basins worldwide Nat.Commun.12 4232 U.S. Geological Survey 2021 USGS water data for the nation: U.S. geological survey national water information system database (available at: https://waterdata.usgs.gov/nwis)Udall B and Overpeck J T 2017 The twenty-first century colorado river hot drought and implications for the future Water Resour.Res.53 2404-18 United States and Mexico 1973 Minute 242 of the U.S.-Mexico International Boundary and Water Commission of 1973 (available at: www.usbr.gov/lc/region/g1000/pdfiles/min242.pdf) Warner J W, Heimes F J and Middelburg R F 1985 Ground-water contribution to the salinity of the Upper Colorado River Basin US Geological Survey Water-Resources Investigations Report 84-4198 1-92 Williams A P, Cook B I and Smerdon J E 2022 Rapid intensification of the emerging southwestern North American megadrought in 2020-2021 Nat.Clim.Change 12 232-4 Williams A P, Cook E R, Smerdon J E, Cook B I, Abatzoglou J T, Bolles K, Baek S H, Badger A M and Livneh B 2020 Large contribution from anthropogenic warming to an emerging North American megadrought Science 368 314-8 Wise D R 2019 Spatially referenced models of streamflow and nitrogen, phosphorus, and suspended-sediment loads in streams of the Pacific region of the United States U.S. Geological Survey Scientific Investigations Report 2019-5112 64 Wise D R, Anning D W and Miller O L 2019 Spatially referenced models of streamflow and nitrogen, phosphorus, and suspended-sediment loads in streams of the Southwestern United States U.S. Geological Survey Scientific Investigations Report 2019-5106 66