Positive effects of public breeding on U.S. rice yields under future climate scenarios

This data repository offers comprehensive resources, including datasets, Python scripts, and models associated with the study entitled, "Positive effects of public breeding on U.S. rice yields under future climate scenarios". The repository contains three models: a PCA model for data transformation, along with two meta-machine learning models for predictive analysis. Additionally, three Python scripts are available to facilitate the creation of training datasets and machine-learning models. The repository also provides tabulated weather, genetic, and county-level rice yield information specific to the southern U.S. region, which serves as the primary data inputs for our research. The focus of our study lies in modeling and predicting rice yields, incorporating factors such as molecular marker variation, varietal productivity, and climate, particularly within the Southern U.S. rice growing region. This region encompasses Arkansas, Louisiana, Texas, Mississippi, and Missouri, which collectively account for 85% of total U.S. rice production. By digitizing and merging county-level variety acreage data from 1970 to 2015 with genotyping-by-sequencing data, we estimate annual county-level allele frequencies. These frequencies, in conjunction with county-level weather and yield data, are employed to develop ten machine-learning models for yield prediction. An ensemble model, consisting of a two-layer meta-learner, combines the predictions of all ten models and undergoes external evaluation using historical Uniform Regional Rice Nursery trials (1980-2018) conducted within the same states. Lastly, the ensemble model, coupled with forecasted weather data from the Coupled Model Intercomparison Project, is employed to predict future production across the 110 rice-growing counties, considering various groups of germplasm.

This study was supported by USDA NIFA 2014-67003-21858 and USDA NIFA 2022-67013-36205.