Currently, little guidance is available for the design of accurate gri
d-cell layouts for spatially distributed models like the Agricultural
Nonpoint Source Pollution (AGNPS) model. Inaccuracies in the grid-cell
input could negate the advantages provided by the spatially distribut
ed approach to watershed (field) representation. The objective of this
study was to develop procedures for designing grid-cell layouts using
geostatistical methods. Geostatistical spatial analysis tools were us
ed to select a base cell size and locate cell sub-divisions for use in
AGNPS. The overall spatial correlation structure of SCS Curve Number
data was used to select the base cell size. The specific criterion use
d to select the base cell size was the average squared residual of kri
ged estimates for randomly selected locations within the study area. T
he cell size with the minimum average squared residual was selected as
the base cell size. Next, cells that wan-anted sub-divisions were ide
ntified using the squared residual of kriged estimates for each cell i
n the base cell layout. The grid-cell layout resulting from the applic
ation of the geostatistical methods and the squared residual criteria
had an average squared residual less than the grid-cell representation
without sub-divisions. Furthermore, the cells that were sub-divided f
ell within the most heterogeneous areas of the watershed. The geostati
stical methods provided quantitative information about the error assoc
iated with various grid-cell designs. This information was then used t
o design accurate grid-cell layouts that preserved the spatial heterog
eneity of the original data. Further development and application of th
e geostatistical methods used to design grid-cell layouts will lead to
greater improvements in the simulation accuracy of spatially distribu
ted models including AGNPS.