In the process to the marketing of cultivars, identification of superior test locations within multi-environment variety trial schemes is of critical relevance. It is relevant to breeding organizations as well as to governmental organizations in charge of cultivar registration. Where competition among breeding companies exists, effective and fair multi-environment variety trials are of utmost importance to motivate investment in breeding. The objective of this study was to use genotype main effect plus genotype by environment interaction (GGE) biplot analysis to evaluate test locations in terms of discrimination ability, representativeness and desirability, and to investigate the presence of multiple mega-environments in cotton production in the Yangtze River Valley (YaRV), China. Four traits (cotton lint yield, fiber length, lint breaking tenacity, micronaire) and two composite selection indices were considered. It was found that the assumption of a single mega-environment in the YaRV for cotton production does not hold. The YaRV consists of three cotton mega-environments: a main one represented by 11 locations and two minor ones represented by two test locations each. This demands that the strategy of cotton variety registration or recommendation must be adjusted. GGE biplot analysis has also led to the identification of test location superior for cotton variety evaluation. Although test location desirable for selecting different traits varied greatly, Jinzhou, Hubei Province, China, was found to be desirable for selecting for all traits considered while Jianyang, Sichuan Province, China, was found to be desirable for none. (Résumé d'auteur

China is one of the largest cotton producing countries in the world thanks to high yields, on which a variety registration system has mainly focused, so that a lack of quality is nowadays acknowledged as a weak point of the cotton industry in that country. The objective of this study was to check the hypothesis that bias in cultivar selection in favor of yield has been maintained through the application of an imperfect selection index (SI), but that a better outcome is possible. Our demonstration is based on an analysis of the data from ten years of cotton variety trials using genotype-by-trait biplots, implemented both for the cultivar selection index (SI) currently applied in China and for an adjusted selection index (ASI) that more effectively took into account the antagonism between yield and quality traits. The main findings were: 1) significant negative associations between yield and fiber quality hindered their simultaneous improvement; 2) registered genotypes were mainly determined by the SI which was primarily yield-oriented; 3) no progress in fiber quality was recorded unlike yield; 4) balanced progress in yield and quality is possible through an adjusted selection index (ASI) guided by genotype-by-trait biplot analysis.