Numerous breeding efforts have improved crop yields by screening new high-yield cultivars through yield trials in multiple environments. However, the accumulated data from these trails has not been effectively upcycled to guide future breeding programs because of the strength of the genotype by environment (G × E) interaction. Here, we propose a new method that accounts for these interactions by using a weather-driven crop growth model; the recorded yields of each cultivar were expressed using a unique linear regression in response to the potential yield (Yp) calculated by a weather-driven crop growth model. We call this approach the YpCGM method. We applied this method to 72 510 independent datasets from yield trials of paddy rice that used 237 core cultivars (n = 20 to 6342 field trials) measured at 110 locations in Japan during the 38 years from 1980 to 2017. The 237 core cultivars were selected from a pedigree matrix of 14 032 Japanese cultivars by using the k-medoids method (k = 200 clusters), and the genomic information for 91 800 single-nucleotide polymorphisms was obtained. The genotypic coefficients of yield-ability and yield-plasticity differed among the 237 cultivars, with values ranging from 2.5 to 7.3 t/ha and from –0.23 to +0.95, respectively. Genomic prediction validated the values of these two parameters by leave-one-out cross-validation based on the pedigree and genome information. Our study represents a novel “big data” method for managing and generalizing G × E interactions using a crop growth model supported by large amounts of field data.