Preface

Crop variety trials are the most valued and best-funded research among applied agricultural researches. Regardless of the economic developmental level and the budget situation, crop variety trials are conducted every year in every region for every major crop of the region. Breeders rely on variety trials to select superior breeding lines to release as new cultivars; farmers rely on variety trials to choose suitable crop cultivars to grow in their farms. Processors rely on variety trials to decide where and of which cultivars to source their grains or other crop products to process.

The direct outcome from crop variety trials is data; the ultimate outcome from crop variety trials is information on the target region, the test locations, and the genotypes, thereby correct decisions can be made on the genotypes for the target region. Data analysis is the process to extract useful information and draw conclusions from the data.

Data analysis is the process to extract useful information and draw conclusions from the data.

Data analyses performed by most researchers conducting variety trials are quite simple, in spite of numerous new and advanced methods advocated by statisticians. In most variety trial systems, the annual report of variety trials is limited to the following aspects: (1) Genotype-by-trait two-way tables for each trial (location), with summary statistics for each trait, such as trial mean, standard error, and least significant difference. (2) Genotype-by-location two-way tables for each trait in absolute values. (3) Genotype-by-location two-way tables for each trait in values relative to the trial mean or to a check. Presenting relative values is one step forward, which serves as a means to remove the environmental main effects and facilitates data summary across trials. (4) Genotypic means across all locations and/or locations within subregions. This is another step forward as this gives genotypic values for the region or subregions, thereby any genotype-by-location interactions across the whole region or a subregion are removed. Genotypic values for a trait can then be used to rank the genotypes, which become the basis for selecting genotypes and recommending cultivars. (5) In addition to genotypic means for the current year, some reports also include genotypic means across recent 2–5 years, when applicable. Genotypic ranking based on data from multiyears is certainly more credible as any genotype-by-year interaction and genotype-by-location-by-year interactions would be removed.

Primitive as it may appear, these simple data summary and analyses are quite effective, as evidenced by the continuous progress in cultivar development and crop production in various crops worldwide. However, the analyses may be improved by asking a few questions. First, when summarizing across all test locations, it is assumed that there are no repeatable genotype-by-location interactions (GL) within the target region represented by these locations. Is this true? When summarizing across locations within subregions, it is assumed that there are repeatable genotype-by-subregion interactions and there are no repeatable GL within subregions. Are these true? If the answer to any of these questions is “no” or “not sure,” then the data summary system may be suboptimal and should be improved. The process to answer these questions is “mega-environment analysis.” Second, the genotypic means across locations and years are calculated under the assumption that all test locations are equally representative of the target mega-environment and equally informative about the genotypes. Are these true? If the answer to any of these is “no” or “not sure,” then the system may be also suboptimal and needs to be improved. The process to answer these questions is “test location evaluation.” Third, two genotypes ranked the same based on genotypic means may be quite different in their specific adaptations or stability across the target region. This is the issue of “stability analysis,” which has been a buzzword in variety trial data analysis. Many stability indices have been proposed during the last 50 years but none is widely used by practical researchers. This is because the researchers are more confused than enlightened by these indices. A clear guidance is needed in this aspect to improve the precision and accuracy of genotype evaluation and cultivar recommendation. Fourth, decisions on genotypes have to be based not only on a single trait like yield but also on quality and other traits; unfortunately desirable traits are often undesirably associated. Genotype evaluation based on undesirably associated traits is a difficult task such that most variety trial reports leave this untouched. However, this is a decision that must be made, and tools and guidelines are needed. Finally, variety trial data analysis and decision-making have been hindered not only by knowledge but also by the availability of relevant, intuitive, verifiable, and user-friendly software. Although many comprehensive, powerful software packages are available, they are designed for use more by professional statisticians than by hands-on plant breeders and agronomists; although statisticians and breeders try to work closely, there is always a large gap between them due to different knowledge base and different research interests.

This book is written to fill the gaps. It is written to help researchers conducting crop variety trials to answer various questions and provide solutions in variety trial design, conduct, data management, data analysis, and decision-making. It starts with the definition of heritability in the framework of multiyear, multilocation variety trials, which is the theoretical foundation of variety trials and crop improvement. Heritability is the measure of the usefulness of the variety trials in variety evaluation. All practical measures in variety trials, from design, conduct, to data analysis, have a single purpose; it is to improve the heritability of variety trials so that superior genotypes can be effectively identified for the target environment (Chapter 1). There are three levels of variety trial data: single trial, multilocation trials in a single year, and multilocation trials in multiple years. The analytical techniques needed include conventional methods such as analysis of variance, variance component analysis, linear correlation, multiple regression, and graphical methods particularly biplot analysis (Chapter 2).

Biplot analysis was first developed by Gabriel (1971) and has become a popular method in variety trial data analysis in the name of “GGE biplot” following some of our work (Yan et al., 2000; Yan, 2001; Yan and Kang, 2003). Biplot analysis is a powerful data visualization tool and can be used to graphically address many research questions including those listed above. However, biplot analysis has not been used properly and adequately in many publications. This is understandable as it is still a new technique to most agricultural researchers and its properties and utilities are still being discovered and developed. The principles of biplot analysis, frequently asked questions, and frequently seen mistakes related to biplot analysis constitute a fair portion of this book (Chapters 3–6). Biplot analysis and conventional statistical analyses are jointly used in the analysis of different levels of data for a single trait (e.g., yield) to address the following issues: spatial or field trend analysis for single-trial data (Chapter 7), mega-environment analysis based on data from single and multiple years (Chapters 8 and 12), test location evaluation based on data from single and multiple years (Chapters 8 and 13), and genotype evaluation based on data from single and multiple years (Chapters 8 and 14). Genotype evaluation and decision-making based on multiple traits are addressed in Chapter 9. In addition, Chapter 10 illustrates the use of biplot analysis in studying trait associations in different environments, which can be extended to quantitative trait loci (QTL) identification based on phenotypic data from multiple environments. Chapter 11 illustrates the use of biplot analysis in studying location-by-trait patterns; this is a new application of biplot analysis and can be useful for processors to identify locations or regions for sourcing grains with desirable quality profile.

Chapter 15 describes a relational database system for storing, managing, and utilizing multilocation, multiyear variety trial data. Chapter 16 describes experimental designs for crop variety trials and breeding nurseries. Most of the biplot analysis and conventional analyses, plus data management and experimental design, are conducted using the GGEbiplot software (www.ggebiplot.com). So the modules and functions of GGEbiplot are systematically but succinctly introduced in the penultimate chapter (Chapter 17). As a crop breeder and the developer of the software I use it for almost all aspects in my breeding work, from experimental design to data management, data analysis, and decision-making. Its high efficiency and user-friendliness allowed me time to write research papers and edit/review manuscripts for many scientific journals, in addition to running a productive oat breeding program.

I write this book as a hands-on plant breeder. All issues addressed in this book are real problems identified from my own breeding work. In fact, since my target...