Fall Seminar Series - September 18, 2003
University of Minnesota
School of Statistics
College of Liberal Arts

Modelling Replicated Weed Growth Data Using Spatially-Varying Growth Curves

Sudipto Banerjee
Division of Biostatistics
University of Minnesota

Thursday, September 18, 2003
4:00 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:30 PM, 300 Ford Hall

Abstract

  Weed growth in agricultural fields constitutes a major deterrent to the growth of crops, often resulting in low productivity and huge losses for the farmers. Therefore, proper understanding of patterns in weed growth is vital to agricultural research. Recent advances in Geographical Information Systems (GIS) now allow geocoding of agricultural data, which enable more sophisticated spatial analysis. Our current application concerns the development of statistical models for conducting spatial analysis of growth patterns in weeds. Our data comes from an agricultural experiment conducted in ten locations over a sixteen hectare field in Waseca, Minnesota, that recorded growth of the weed Setaria spp., which is known to be particularly detrimental to the growth of crops. We capture the spatial variation in Setaria spp. growth using spatially-varying growth curves. An added challenge is that these designs are spatially replicated, with each plot being a lattice of sub-plots. Therefore, spatial variation may exist at different resolutions - a \emph{macro} level variation between the plots and \emph{micro} level variation between the sub-plots nested within each plot. We develop a Bayesian hierarchical framework for this setting. Flexible classes of models result which are fitted using simulation-based methods.