Modeling Spatial and Temporal Variation in Natural Background Specific Conductivity

Area of Science:

• Environmental Science
• Hydrology
• Water Quality Assessment

Background:

• Understanding temporal and spatial variations in stream dissolved mineral levels is crucial for assessing freshwater salinization.
• Assessing background mineral levels is essential for setting restoration goals and identifying vulnerabilities to climate events like drought.

Purpose of the Study:

• To develop a predictive model for natural background specific conductivity (SC) in all stream segments across the contiguous United States.
• To provide monthly predictions of SC from 2001 to 2015, aiding in water resource management and environmental monitoring.

Main Methods:

• A random forest model was developed using 11,796 observations from minimally impaired streams.
• Static predictors (geology, soils, vegetation) and temporal predictors (climate) were used to train and validate the model.
• Model validation involved an additional 92 stream segments, achieving high accuracy.

Main Results:

• The model explained 95% of the variation in SC among validation observations, with a mean absolute error of 29 microS/cm.
• High predictive accuracy (Nash-Sutcliffe efficiency = 0.85) was achieved across the study period, though bias was noted in specific regions.
• National predictions revealed significant spatial variation in SC, with higher levels anticipated in the desert southwest and plains, and reflected drought impacts.

Conclusions:

• The developed random forest model accurately predicts background specific conductivity in U.S. streams, offering a valuable tool for water quality assessment.
• The model's ability to capture temporal and spatial variations aids in understanding salinization dynamics and climate change impacts on freshwater resources.
• Findings highlight the importance of considering geological, soil, vegetation, and climate factors in predicting stream water quality.