Can Spatial Modeling Substitute for Experimental Design in Agricultural Experiments?

Crop Science

Ernst, Oswaldo^3¹  Cadenazzi, Mónica^1²  Terra, José^4³  González-Reymundez, Agustín^2⁴  Borges, Alejandra^1⁵ 

[1]Dep. of Crop Production, Facultad de Agronomía, Univ. de la República, Estación Experimental Mario Cassinoni, Paysandú, Uruguay^3

[2]Estación Experimental Treinta y Tres, Instituto Nacional de Investigación Agropecuaria (INIA), Ruta 8, Km 281, Treinta y Tres 33000, Uruguay^4

[3]Statistics Dep., Facultad de Agronomía, Univ. de la República, Garzón 780, Montevideo, Uruguay Dep. of Agronomy, Univ. of Wisconsin–Madison, 1575 Linden Dr., Madison, WI 53706^5

[4]Statistics Dep., Facultad de Agronomía, Univ. de la República, Garzón 780, Montevideo, Uruguay current address, Dep. of Epidemiology and Biostatistics, Michigan State Univ. 909 Fee Rd., B601 West Fee Hall, East Lansing, 48823 MI^2

[5]Statistics Dep., Facultad de Agronomía, Univ. de la República, Garzón 780, Montevideo, Uruguay^1

关键词: ALPHA;    α-lattice incomplete block experimental design;    AR(1);    spatially correlated error model with one-dimensional autoregressive process;    Best_Gen;    proportion of times the true 15% superior genotypes are recovered;    COR;    Pearson’s correlation coefficient between true and estimated effects;    CRD;    completely randomized experimental design;    EXP(2);    spatially correlated error model with two-dimensional exponential process;    MSEP;    mean square error of prediction;    NSC;    no spatial correction model;    PREP;    partially replicated experimental design;    PREP g;    partially replicated experimental design with fixed number of genotypes;    PREP n;    partially replicated experimental design with fixed number of experimental units;    RCBD;    randomized complete block experimental design;    YSD;    yield standard deviation;   

DOI  :  10.2135/cropsci2018.03.0177

学科分类：农业科学（综合）

来源: Crop Science

【 摘 要 】

One of the most critical aspects of agricultural experimentation is the proper choice of experimental design to control field heterogeneity, especially for large experiments. However, even with complex experimental designs, spatial variability may not be properly controlled if it occurs at scales smaller than blocks. Therefore, modeling spatial variability can be beneficial, and some studies even propose spatial modeling instead of experimental design. Our goal was to evaluate the effects of experimental design, spatial modeling, and a combination of both under real field conditions using GIS and simulating experiments. Yield data from cultivars was simulated using real spatial variability from a large uniformity trial of 100 independent locations and different sizes of experiments for four experimental designs: completely randomized design (CRD), randomized complete block design (RCBD), α-lattice incomplete block design (ALPHA), and partially replicated design (PREP). Each realization was analyzed using different levels of spatial correction. Models were compared by precision, accuracy, and the recovery of superior genotypes. For moderate and large experiment sizes, ALPHA was the best experimental design in terms of precision and accuracy. In most situations, models that included spatial correlation were better than models with no spatial correlation, but they did not outperformed better experimental designs. Therefore, spatial modeling is not a substitute for good experimental design.

【 授权许可】

CC BY