Presentations

Models were developed at two contrasting spatial scales to assess the vulnerability of groundwater to agricultural impacts. First, a nonlinear regression model (GWAVA) was developed at the national scale (R-squared = 0.80) to predict nitrate concentration in shallow groundwater (typically < 5 m deep). GWAVA is empirical but is more physically realistic than typical regression approaches. It has a structure that separates nitrogen (N) sources from factors that enhance or restrict nitrate transport and accumulation in groundwater. Input data were spatially averaged to minimize small scale variability so that the large scale influences of N loading, climate, and aquifer characteristics could be examined. Second, a mechanistic model (the Root Zone Water Quality Model or RZWQM2) was used to predict unsaturated zone N fluxes for the San Joaquin Valley of CA. RZWQM2 is a one-dimensional unsaturated zone fate and transport model that also simulates crop growth and the effects of agricultural management practices. The model was used to predict daily nitrate concentration in deep seepage so that responses at the water table to weather, irrigation, and other factors could be evaluated. The model was calibrated by inverse modeling to data collected throughout the unsaturated zone and yielded an index of agreement (d) value of 0.60 (d is interpreted like R-squared but is more sensitive to differences between predicted and observed values). RZWQM2 predicted that 67 percent of the N loss from the unsaturated zone occurred through plant uptake (144 kg N/ha-yr) and that the remainder occurred primarily through deep seepage (71 kg N/ha-yr). Pulses of nitrate at the water table were predicted to occur in response to irrigation. The two modeling approaches are complementary. Whereas GWAVA is primarily a spatial tool to compare levels of contamination in areas, RZWQM2 predicts changes in contaminant fluxes in response to changing conditions at the land surface. Empirical models such as GWAVA require fewer inputs and are easily applied at large spatial scales to identify impacted areas for monitoring and/or more detailed mechanistic modeling. The latter require more input data and typically are applied at the field scale. Because they have an explicit temporal component, mechanistic models can be used to determine the time required to realize potential effects from alternative management practices. Additionally, mechanistic models can help identify variables for potential incorporation into empirical models, such as soil hydraulic properties and plant characteristics.