Genomic Selection and Predictive Breeding Approaches for Enhancing Crop Productivity and Climate Resilience: A Review

Genomic selection (GS) and predictive breeding have emerged as practical approaches to accelerate genetic gain in crops confronted by rising temperatures, increasingly erratic rainfall and growing food demand. Unlike marker-assisted selection based mainly on a limited set of significant loci, GS uses genome-wide molecular markers and phenotypes from a training population to predict genomic estimated breeding values of untested candidates. This review synthesises foundations, prediction methods, training population optimisation, genotype-by-environment interactions, high-throughput phenotyping and operational breeding decisions. It examines the use of ridge-regression best linear unbiased prediction, genomic best linear unbiased prediction, Bayesian shrinkage models, nonlinear kernels and machine learning, and highlights why model comparisons must be evaluated using forward-in-time and independent-environment validation rather than randomly partitioned related materials alone. Evidence from maize, wheat and other crops indicates that GS can shorten selection cycles and improve the efficiency of decisions for complex traits, although performance depends on genetic relationships, heritability, trait architecture, reference population size and changing environments. Integration of weather covariates, crop development, managed-stress trials and remote-sensing phenotypes can improve the biological relevance of predictions for drought and heat resilience, but uncertainty remains substantial under novel climates. The review proposes a practical implementation framework emphasising target population of environments, data quality, economic gain per unit time, genomic diversity, farmer-relevant traits and transparent model updating. Challenges include limited phenotyping infrastructure, biased training data, computational capacity, seed-system bottlenecks and inequitable access to genomic resources. The central conclusion is that predictive breeding should be viewed as a continuously tested decision system, not a standalone algorithm: reliable phenotypes, representative environments and disciplined breeding operations are at least as important as statistical sophistication.