Genomic selection has the potential to increase the efficiency of breeding programs in perennial crops. In apple, genomic predictions have been reported only in a limited number of studies and to our knowledge genomic selection is hardly applied. The very first step to build up an efficient genomic selection program is to develop an experimental design with adequate plant material to test predictions. In apple, this is now made possible with the joint efforts of the European FruitBreedomics Consortium in developing the apple REFPOP, a population dedicated to genomic predictions and genome wide association analyses, along with appropriate genotyping and phenotyping tools. Here, we present the long-term genomic selection apple project built around the apple REFPOP. This population is composed of 570 replicated genotypes planted in six countries and was phenotyped for the first time in 2018 at all six sites for yield, phenology and fruit quality traits. Phenotyping for these and additional traits will be repeated in the coming years. Genotyping data were available through FruitBreedomics for most of the genotypes and were obtained with the combination of a medium, 20K SNP array, and the high-density Axiom® Apple487K SNP Affymetrix array via imputation. We provide insights into the genetic features of the apple REFPOP with a specific focus on linkage disequilibrium and structural patterns. By combining several phenotypic variables recorded across countries and years and high-density genotyping, our goal is to use the apple REFPOP as a training population to predict the potential of applied populations in multiple environments and for multiple traits. To face the increasing complexity of the data collected, machine learning methods will be tested. Finally, we also discuss different strategies to calibrate and implement genomic prediction models into modern, climate-ready breeding programs. This work is partially funded by the EU-H2020 project N°817970 INVITE