Degree Type


Date of Award


Degree Name

Doctor of Philosophy


Theses & dissertations (Interdisciplinary)


Water Resources

First Advisor

Steven E. Jungst

Second Advisor

R. S. Kanwar


Non-point source pollution in agricultural regions in the Midwest consists of two important components, namely, dislodgment of common pollutants, their movement across the landscape and subsequently transport through surface waters to their final destination. The objectives of this study were to: 1) model surface movement of agricultural pollutants for a succession of sub-watersheds in Bear Creek using AGNPS; 2) calibrate and validate the model for a small sub-watershed using available experimental plot data; 3) incorporate Monte Carlo simulation within the modeling process where field data was not available; 4) apply the calibrated and validated model to a small sub-watershed with a riparian buffer strip to observe mitigating effects on surface pollutants if any; 5) formulate a Markov random field model for use in modeling nitrate-nitrogen concentrations for a network of monitoring stations on the Des Moines River; 6) estimate, fit, and cross-validate the model and; 7) apply the model to dissolved oxygen and suspended solids to evaluate the model's specificity for each water quality variable. The agricultural non-point source pollution model (AGNPS) was used to model surface transport processes at the watershed level. Modifications were made to the existing AGNPS-ARC/INFO interface in order to automate feedlot, point source, and channel information. Feedlot information was obtained through a survey, while a probabilistic method was used to predict tile locations for the lower half of the watershed.;The model was calibrated for total runoff (inches), sediment yield (tons), soluble nitrate-nitrogen in runoff (lbs/acre), soluble nitrate-nitrogen concentration in runoff (ppm), phosphorus in sediment (lbs/acre), soluble phosphorus in runoff (lbs/acre) and soluble phosphorus concentration in runoff (ppm) to be found at the outlet of each watershed. The model was then applied to a small watershed to demonstrate the effect of riparian buffer strips in mitigating surface pollution in an agricultural field.;From graphical displays it was clear that the model predicted better under drier conditions than for excessively wet conditions. For 1997, which was a dry year, the observed and predicted values are almost equal. In contrast for 1998, which was a wet year, the difference between the observed and predicted values is very large. Corn consistently gave lower RMSE for all seven parameters calibrated (0.02, 0.09, 0.44, 0.09, 0.00, 0.05, 0.59) compared to either the combination of rowcrop and switchgrass (0.18, 0.09, 0.49, 0.11, 0.01, 0.14, 0.01) or row crop, switchgrass, shrubs and trees (0.11, 0.04, 0.18, 0.07, 0.02, 0.10, 0.00). The buffer strip in general seemed to have a mitigating effect on the pollutants as they moved over the surface.;Switchgrass seemed to be highly effective in reducing the movement of pollutants into the stream. Runoff volume was reduced by 19%, soluble nitrogen concentrations were reduced by 3%, soluble phosphorus concentrations were reduced by 3% for the early summer precipitation event. For larger watersheds, where field data was not available, Monte Carlo simulations were included in the modeling process. Smaller watersheds in the headwaters were the highest contributors of pollutants to the stream. EBCM3, which is a very small sub-watershed within BC9 showed a maximum concentration of soluble nitrogen (0.43 ppm), as well as a maximum concentration of soluble phosphorus (0.20 ppm), whereas BC9 showed an absence of both. For the second part of the research a statistical model was developed for modeling nitrate-nitrogen concentrations in the Des Moines River.;The model was formulated as a "conditionally specified model" in which parametric forms were assigned to conditional densities. Both "systematic" and "random" components were modeled effectively. A space-time metric was developed to represent spatial dependence in the "variable distance" model versus a purely spatial metric in the "fixed distance" model. An independent model was also fit, which only included systematic trend, in order to make comparisons between models. The "variable distance" or flow model performed better due to the ability to model greater variability in the system. The flow model had smaller mean squared errors for nearly all the stations. The improvement achieved from the mean-covariance model by adding just one parameter to get the flow model is better than that achieved by the distance model.;The variance and mean squared prediction error for 1982-1996 data for nitrate using all the data and then dropping station 6 while doing cross validation are very similiar. This indicates good predictive capability of the model for nitrate nitrogen. The mean square prediction error (MSPE) is only slightly higher for the prediction than it is when the data for all the stations is available. When applied to dissolved oxygen and suspended solids, the model did not perform as well. This indicated that the model was specific to nitrate-nitrogen which transported differently than dissolved oxygen or suspended solids. Indications of the specificity of the model were visible in the exploratory data analysis.



Digital Repository @ Iowa State University,

Copyright Owner

Shabana Hameed



Proquest ID


File Format


File Size

237 pages