(Georgia Institute of Technology, 2011-04)
Ssegane, Herbert; Tollner, E. W.; Mohamoud, Yusuf; Rasmussen, Todd C.; Dowd, John F.
The study explored use of causal feature selection algorithms to select dominant watershed variables that drive high, medium, and low flows. A two step approach was implemented. The first step minimized variable redundancy by examining variable relevance, variable redundancy, and conditional relevance of variable pairs whose correlation was greater than 0.9. The second step used six algorithms that seek to reconstruct a Bayesian network structure around a target variable for each flow percentile. Nineteen (19) flow percentiles were used to characterize high, medium, and low flow conditions of 26 Piedmont watersheds in the Mid-Atlantic. The algorithms included: (1) Grow-Shrink (GS); (2) interleaved-Incremental Association Markov Boundary (interIAMB) (3) Incremental Association Markov Boundary with Peter-Clark (IAMBnPC); (4) Local Causal Discovery (LCD2); (5) HITON-PC; and (6) HITON-MB. A new method was developed to quantify the reliability of each algorithm and its performance was compared to existing reliability methods. The effect of the initial number of variables on the final variable set selected by each algorithm was tested. Fusion of the algorithms was used to determine the overall dominant features for each flow percentile.
(Georgia Institute of Technology, 1997-03)
Burkingstock, Bryan K.; Bush, Parshall B.; Berisford, Yvette C.; Dowd, John F.; Feild, James B.; Johnson, West W.; Rasmussen, Todd C.; Taylor, John W.
Imidacloprid has been shown to control pine tip
moths (Rhyacionia spp.) in loblolly pine seedlings (Pinus taeda
L.). We describe an imidacloprid dissipation study on a small
watershed near Downs, Georgia. Field data are utilized to
calibrate the computer model VS2DT (Variably Saturated 2-
Dimensional Transport) that uses a fmite-difference
approximation of the advection-dispersion equation to simulate
contaminant transport through variably saturated porous media.
The site is simulated in a vertical cross section (X-Z plane) of
a treated watershed from the soil surface to the water table
(approximately 60 ft) spanning 260 ft horizontally. The cross
section is centered on an ephemeral stream which drains the
watershed, but for symmetry reasons, only half (130 ft
horizontal distance) of the cross section is modeled. Modeling
results show that imidacloprid moves more rapidly through
sandy soils than sandy clay loam soil material. No imidacloprid
concentrations above the detection limit of 0.6 ppb are
predicted for the lysimeters in the unsaturated zone or in groundwater.