|
Activity Number:
|
377
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #301041 |
|
Title:
|
Spatially Balanced Designs for Experiments on Autocorrelated Fields
|
|
Author(s):
|
Harold M. van Es*+ and Cindy L. van Es and Carla Gomes and Meinolf Sellmann
|
|
Companies:
|
Cornell University and Cornell University and Cornell University and Cornell University
|
|
Address:
|
1005 Bradfield, Ithaca, NY, 14853-1901,
|
|
Keywords:
|
design ; autocorrelation ; agronomy ; spatial
|
|
Abstract:
|
Despite the availability of a multitude of advanced design, most agronomic field experiments (at least 90%) are implemented through randomized complete block designs (RCBD). Use of blocks is generally justified by spatial autocorrelation of soil fertility patterns due to regional variations in drainage, erosion, etc. We demonstrated in earlier work that randomized treatment allocation to plots can cause biases and imprecision under most field conditions due to autocorrelation and trends. We propose spatially balanced experimental designs that are inherently robust to nonrandom field variability (autocorrelated or trended), and use standard designs that are optimized in terms of average spatial separation of treatments. We are using two approaches to develop spatially balanced CBDs: a heuristic local search approach using simulated annealing, and an exact method using combinatorics (for Latin Squares only at this time). We discovered that local search methods provided satisfactory results for smaller designs only. Results with a combinatoric approach are promising and we generated totally spatially balanced Latin Squares up to order 18.
|
