Citation:

Smith, R. AND J. Renee Brooks. Linking spatial patterns with processes in river networks using stable isotopes: Evaporation in Mississippi River watershed. American Geophysical Union Annual meeting, Washington DC, Washington DC, December 10 - 14, 2018.

Impact/Purpose:

Managers need to understand how water sources, evaporation and storage affect runoff in large watersheds for managing non-point pollution, and water quality and quantity of freshwater resources. River isotopes provide an integrated signal of recharge sources and evaporation, however the processes transferring this signal along flowpaths are poorly understood. We used a new class of geostatistical models to infer processes influencing water quality and quantity in large river networks.

Description:

Understanding how evaporation and storage affect runoff in large watersheds is critical for management of freshwater resources. River isotopes provide an integrated signal of recharge sources and evaporation, however the processes transferring this signal along flowpaths are poorly understood. We used a new class of geostatistical models that account for unique spatial dependencies defining river networks (e.g., branching and longitudinal connectivity), called Spatial Stream Network (SSN) models, to quantify the relative role of in-stream versus landscape processes (e.g., irrigation, soils, surface water, and snowmelt) contributing to variability in hydrogen isotopes in the Mississippi River. To do so, we first developed a mass-balance prediction of river 2H/1H ratios and deuterium excess (d, deviation from global meteoric water) for each catchment by multiplying spatially gridded precipitation isoscapes and runoff estimates. We regressed predictions against measured river isotopes from the EPA’s National Rivers and Streams Assessment to test the degree to which river isotope ratios reflect precipitation isoscapes, and compared several SSN models with different autocovariance structures that accounted for Euclidean and network spatial dependency, or both. The spatial dimension defining deviations from mass-balance yields insight into how the isotope signal is modified along the river channel or across the landscape. SSN models including both autocovariance structures performed significantly better than non-spatial models, or models with Euclidean autocovariance only (dAIC >4). Euclidean and in-stream autocovariance explained 85% and 6% for 2H/1H, and 74% and 24% for d respectively, demonstrating a measureable effect of in-stream processes. To explore mechanisms contributing to these spatial patterns, we analyzed residuals of SSN models without spatial weighting, regressing measured river isotope ratios and the precipitation-based predictions. Mean catchment slope, baseflow index, VPD, % water body area, soil permeability, and elevation were correlated with non-spatially weighted residuals, suggesting that both recharge and evaporation drive observed spatial patterns. Our study demonstrates a novel approach for using spatial statistics to infer processes in river networks.

Record Details: