Livestock agriculture (e.g., dairy, beef, pork, poultry) in the USA is tending rapidly toward operations where a large number of animals are concentrated in a relatively small area. The economies of scale are counterbalanced by the dangers of pollution from inadequate treatment of animal waste. Traditional methods of treatment involve lagoon retention and subsequent spreading on fields but the sheer volume of production seems to be outstripping these and other technologies. Surface-water runoff finds its way into streams and rivers, ultimately polluting all downstream segments of the watershed. The topic of this paper is spatio-temporal statistical modeling of (log) nitrate concentration in the upper North Bosque watershed, which is a region of concentrated dairy operations. A model is fitted from daily data collected over a period of 15 months, at 17 stream monitoring sites throughout the watershed. Optimal predictions of unknown nitrate concentration, at all stream locations at any given time, are obtained, along with a measure of their variability. The model allows for policy changes to be made, and assessed, based on the consequent spatio-temporal predictions.