Commercial horticultural variety trials and crop models: a data suitability assessment

Horticultural production systems are complex, driven by a combination of genetic, environmental, and management factors. Model-driven decision support tools may reduce this complexity, if suitable data to train and calibrate these crop models can be identified. Commercial variety trials may be a suitable data source. In this study, a systematic data suitability assessment of commercial pepper (Capsicum annuum L.), tomato (Solanum lycopersicum L.), and watermelon (Citrullus lanatus L.) variety trials was conducted for use with five crop models (AquaCrop, DSSAT, STICS, reduced order TOMGRO, reduced order de Koning) and a hypothetical machine learning yield prediction model. Using a novel data suitability assessment methodology, variables from a corpus of variety trial reports and raw data were evaluated against model-variable requirements using three suitability criteria: variable availability in the corpus, spatiotemporal resolution compatibility, and variable criticality for model functioning. Variety trial data were found suitable for only the machine learning yield prediction model. Unsuitability was driven by the low availability of temporally dynamic structural characteristic variables, which often required destructive sampling. However, many potentially measurable variables were cumulatively laborious and beyond trial scope. It is therefore recommended that variety trial data be used only with machine learning yield prediction models until practitioners of theory-informed modeling can collaborate with variety trial operators to develop crop models with reasonable variable requirements. It is also recommended that variety trial operators supply raw data when possible and make use of reporting systems that minimize the burden of data entry because raw data contains crucial modeling variables.