Objective

The Maize GxE project is a collaborative effort that involves researchers from diverse areas of study. The datasets collected by the project are some of the largest public data of their kind and are therefore of broad interest to communities from genetics to agronomy to computer science and beyond. The competition was organized to connect these communities and others with interest in dissecting and exploring genotypic, environmental, and GxE information to predict hybrid maize performance in different environments across the US. The competition started on November 15, 2022, and ended on January 15, 2023. All the participants had access to the same curated data set, containing information collected on over 180,000 maize field plots and involving 4,683 hybrids. Participants were asked to create predictive models for maize grain yield for the 2022 Maize GxE project field trials, utilizing the existing Maize GxE project dataset and any other publicly available data. The trait of interest was grain yield, and the competitors were asked to submit absolute grain yield (Mg ha− 1) adjusted to 15.5% moisture for each hybrid in each location where data had been collected during the 2022 field season. The winner of the competition was the model with the lowest average root mean squared error (RMSE) across locations when compared with the actual yield data obtained in 2022.

Data description

The Prediction Competition data are publicly available via CyVerse/iPlant. This dataset contains training and testing set data and has been structured according to the specifications outlined in Table 1.

  • Training_data: includes phenotypic, genotypic, soil, weather (downloaded from https://power.larc.nasa.gov), environmental covariate data, and metadata information from 2014 to 2021 for use in developing and training models.

  • Testing_data: includes genotypic, soil, weather, environmental covariate data, and metadata information for 2022 locations. Also, a submission template that contains the environments and hybrids that participants used to submit yield predictions.

Maize is cultivated as a hybrid crop, typically resulting from the cross of two inbred parents. Consequently, both the phenotypic data in the training and testing sets exhibit hybrid information. The genotypic data includes hybrid information generated in-silico from inbred genotypic data.

Table 1 Overview of Genomes to Fields 2022 Maize Genotype by Environment Prediction Competition data files
Full size table

Limitations

These datasets contain missing data. When working with large agricultural datasets, missing data is a common occurrence due to various factors such as data collection limitations, measurement errors, plot losses, and environmental events. The genotypic data provided contains hybrid information derived from inbred genotypic data, a common practice. However, depending on the study goals, this may pose limitations for specific types of analysis. In instances where precise GPS coordinates were not available for certain environments (i.e., a location in a particular year), field coordinates were estimated. Depending on the research objective, the unavailability of accurate GPS coordinates could impact the reliability of the results.