Modeling the effect of biostimulants on horticultural crop performance, Poinsettia: an approach combining non-target plant analysis methods and data science

International audience

anglais. Poinsettia (Euphorbia pulcherrima Willd.) is an important ornamental horticultural crop, which is potted in summer to reach a commercial stage in December in Europe. The Poinsettia culture in greenhouse starts in July and may be exposed to high temperatures. Then, this crop needs to deal with heatwave periods during the rooting phase. The purposes of this study is to assess whether the application of biostimulants promotes Poinsettia growth during the summer season, and whether such effects can be predicted early in the cultivation process.Initially, over a four-years cultivation period (2019 to 2022) involving 10 different types of biostimulants, we gathered phenotypic data and Near Infrared Spectroscopy (NIRS) measurements during two growth phases (5 and 28 days after potting). Additionally, climate data were recorded: Relative Humidity, total radiation and temperature. Crop performance was modelled using Machine Learning tools (GLMNET, RF and SVM) with the NIRS data, comprising over 1000 variables for a sample size of 200 plants.The study revealed a pronounced year-to-year effect, which hides the effect of biostimulants. Principal component analysis highlights the role of summer Relative Humidity on better root development of young plants. Biostimulants based on amino acids exhibited a slight advantage over other types, such as seaweed extracts and microorganisms. Generalized Linear Model with penalization was used to predict plant performance, expressed into commercial rating. The dataset was divided, with 70% used for algorithm training and 30% for testing the resulting model. This predictive approach showed that plant performance could be predicted from rooting variables at the start of the culture.Specific biostimulants consistently promoted plant growth at the end of summer across the four years (2019 to 2022) while others exhibited varying effects each year. The potential impact of those biostimulants on early-stage cultivation could be later assessed using NIRS data and machine learning tools.