Skip to main content
ARS Home » Pacific West Area » Maricopa, Arizona » U.S. Arid Land Agricultural Research Center » Water Management and Conservation Research » Research » Publications at this Location » Publication #170164
Title: TESTING APPROPRIATE ON-FARM TRIAL DESIGNS AND STATISTICAL METHODS FOR COTTON PRECISION FARMING

Author
item GRIFFIN, TERRY - PURDUE UNIV IN
item Fitzgerald, Glenn
item KANBERT, DATTIB - PURDUE UNIV IN
item LOWENBERG-DEBOER, J - PURCUE UNIV IN

Submitted to: National Cotton Council Beltwide Cotton Conference
Publication Type: Abstract Only
Publication Acceptance Date: 1/5/2005
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Precision agriculture (PA) has reduced the cost of data collection, but most on-farm comparisons are large block, split field or paired field comparisons with little or no replication. For logistical reasons, relatively few producers use split planter, strip trial or other on-farm designs derived from classical small plot experiments. The general objective of this study was to determine if spatial analysis could help farmers make better use of limited replication data they currently collect with cotton yield monitors (YM). Specific objectives are to describe a method for testing alternative experimental designs for cotton farming while addressing spatial autocorrelation (SAC) using two regression methods. Since the introduction of cotton YM lagged that of the combine, less research has been conducted on its use. In 2001, 37% of U.S. corn acres and less than 2% of U.S. cotton acres were harvested with a YM. Now that a substantial number of cotton farmers are collecting YM data, spatial analyses and experimentation is becoming sought after. Cotton production presents unique problems to on-farm comparisons. The potentially dense cotton YM data can feasibly be collected on-the-go, and planned on-farm comparisons can be implemented, harvested and analyzed with PA. For instance, several cotton inputs such as midseason insecticides, growth regulators, and defoliants are applied with aerial applicators. If the farmer wanted to compare insecticides, larger treatment blocks would be easiest to implement. With dense YM data, suspect data points may be removed from analysis leaving an adequate number of measurements. Using a series of simulated fields with known SAC ranging from negligible to strong, different levels of replication were evaluated. Monte Carlo simulations show that as SAC increases in the field, the accuracy of ANOVA under classical assumptions substantially decreases while ANOVA corrected for SAC correctly estimates the true model parameters at the 5% level. Both models had similar accuracy in the absence of SAC. Results suggest that large block cotton YM data with little or no replication can be useful if SAC is correctly modeled. Furthermore, farmers can make reliable decisions based on unreplicated or limited replication comparisons, if that data is analyzed appropriately. To demonstrate how these methods apply to cotton farms, large block tillage comparisons were used. Four replicated treatments including three minimum plus conventional tillage were applied to irrigated cotton at the University of Arizona's Maricopa Agricultural Center 40 km south of Phoenix. Electromagnetic induction readings were correlated to sand content and used as a continuous covariate in the analysis. Preliminary results indicate that ANOVA using a spatial regression model provides more accurate results compared to standard ANOVA. When a standard AVOVA was used, significance levels indicated none of the treatments were significantly different. However, when a spatial ANOVA was used, two treatment variables were different from the mean at the 5% level and one at the 10% level. These results show that when SAC is taken into account, information about local variations over the production surface is gained. Using ordinary ANOVA analysis, these important effects would have not been identified.