Salinity Yield Modeling of the Upper Colorado River Basin Using 30‐m Resolution Soil Maps and Random Forests

Abstract

Salinity loading in the Upper Colorado River Basin (UCRB) costs local economies upward of $300 million U.S. dollars annually. Salinity source models have generally included coarse spatial data to represent nonagriculture sources. We developed new predictive soil property and cover maps at 30-m resolution to improve source representation in salinity modeling. Salinity loading erosion risk indices were also created based on soil properties, remotely sensed bare ground exposure, and topographic factors to examine potential surface soil erosion drivers. These new maps and data from previous SPARROW models were related to recently updated records of salinity at 309 stream gauges in the UCRB using random forest regressions. Resulting salinity yield predictions indicate more diffuse salinity sources, with slightly higher yields in more arid portions of the UCRB, and less overall load coming from irrigated agricultural sources. Model simulations still indicate irrigation to be the major human source of salinity (661,000 Mg or 12%) and also suggest that 75,000 Mg (1.4%) of annual salinity in the UCRB is coming from areas with excessive exposed bare ground in high-elevation mountain areas. Model inputs allow for field-scale screening of locations that could be targeted for salinity control projects. Results confirm recent studies indicating limited surface erosional influence on salinity loading in UCRB surface waters, but impacts of monsoonal runoff events are still not fully understood, particularly in drylands. The study highlights the utility of new predictive soil maps and machine learning for environmental modeling.

Mapping erosion risk for saline rangelands of the Mancos Shale using the rangeland hydrology erosion model

“…The RHEM model characterizes slope using sine of the slope angle, which at low slope values is similar to percent slope. In order to provide information on slope shape for RHEM, the longitudinal convexity of the smoothed 30-m DEM was calculated using the topographic modeling function in the ENVI image Maps of statistically estimated rock fragment size, rock cover, and sodium adsorption ratio (SAR, Figure 3e) also were taken from Nauman et al (2019). For RHEM, rock cover was set to 0% if the estimated fragment size was less than 5 mm, and the estimated total rock cover was applied to the remaining areas ( Figure 3d).…”

“…Since we used datasets from Nauman et al (2019), that source should not be considered independent. Beyond the specific case of SAR, many soil properties across the Mancos Shale are the result of heterogenous, fine-scale deposition processes (Tuttle et al, 2014 the SAR map product was likely much more of an issue than the precipitation.…”

“…In fact, the outline of the Mancos Shale is quite visible as the areas of highest rangeland salt yields in Figure 5 of that report. Using new high‐resolution soil datasets that they developed statistically, Nauman, Ely, Miller, and Duniway (2019) applied random forest regression to predict salinity sources in the upper Colorado River basin (UCRB) with a 30‐m grid resolution. One of the most consistent predictors of salinity in that effort was an index of bare ground, particularly in mountainous terrain.…”

A hybrid approach for predictive soil property mapping using conventional soil survey data

“…Many areas of the UCRB have been identified as having “sensitive soils” by resource managers and stakeholders due to concerns about salinity loading (Nauman, Ely, Miller, & Duniway, 2019; Tillman, Anning, Heilman, Buto, & Miller, 2018), accelerated soil erosion by wind and water (Bailey, 1935; Duniway et al., 2019; McFadden & McAuliffe, 1997; Munson, Belnap, & Okin, 2011; Nauman, Duniway, Webb, & Belnap, 2018) and dust impacts on snowmelt runoff timing (Painter et al., 2010, 2018; Skiles et al., 2015). The UCRB is quite diverse, with elevations ranging from 353 to 4,274 m, including arid deserts up through alpine tundra.…”

“…Potential limitations or shortfalls of PSPM studies completed to date include (a) difficulty finding sufficient quantities of training and covariate data; (b) omission of many soil properties generally found in a conventional soil survey, like the soil survey geographic (SSURGO) database of the United States (e.g., salinity, erodibility, sodium adsorption ratio, gypsum, carbonates, sand fractions, and others) (Soil Survey Staff, 2018); and (c) inconsistent representation of uncertainty (Nauman & Duniway, 2019). Emerging trends show promise in improving PSPM, including the use of many more covariates, such as remotely sensed imagery summarized over large time series (Hengl et al., 2017; Maynard & Levi, 2017; Ramcharan et al., 2018), covariates and model learners that account for varying spatial scales of inference (Behrens et al., 2014; Behrens, Schmidt, MacMillan, & Viscarra Rossel, 2018; Behrens, Zhu, Schmidt, & Scholten, 2010b), and approaches that modify random forests (RFs) to better deal with inherent averaging bias (Hengl, Nussbaum, Wright, Heuvelink, & Gräler, 2018; Nauman et al., 2019; Nguyen, Huang, & Nguyen, 2015; Zhang & Lu, 2012). Because many studies rely on previously collected data that may not be a representative sample (Brus, Kempen, & Heuvelink, 2011), further exploration of patterns in model uncertainty may help fill sampling gaps.…”