Incorporation of auxiliary information in the geostatistical simulation of soil nitrate nitrogen

In north-central Florida the potential risk for movement of nitrate into the aquifer is high due to the large extent of well-drained marine-derived quartz sand overlying porous limestone material coupled with high precipitation rates. Our objective was to estimate spatio-seasonal distributions of soil NO3-N across the Santa Fe River Watershed in north-central Florida. We conducted spatially distributed synoptic and seasonal sampling (September 2003 - wet summer/fall season, January 2004 - dry winter season, May 2004 - dry spring season) of soil NO3-N. Prior distributions of probability for NO3-N were inferred at each location across the watershed using ordered logistic regression. Explanatory variables included environmental spatial datasets such as land use, drainage class, and the Floridian aquifer DRASTIC index. These prior probabilities were then updated using indicator kriging, and multiple realizations of the spatial distribution of soil NO3-N were generated by sequential indicator simulation. Cross-validation indicated that smaller prediction errors are obtained when secondary information is incorporated in the analysis and when indicator kriging is used instead of ordinary kriging to analyze these datasets characterized by the presence of extreme high values and a nonnegligible number of data below the detection limit. The NO3-N values were lowest in September 2003 as a result of excessive leaching caused by large, intense tropical storms. Overall the NO3-N values in January 2004 were high and could be attributed to fertilization of crops and pastures, low plant uptake, and low microbial transformation during the winter period. Despite seasonal trends reflected by the values of observed and estimated NO3-N, we found areas that showed consistently high soil NO3-N throughout all seasons. Those areas are prime targets to implement best management practices.