Evaluation of conditional treatment effect of salt stress on tomato sugar content using causal machine learning: A pilot study

Abstract

Exposing tomatoes to salt stress has been reported to increase the fruit sugar content (°Brix); however, the causal impact of this treatment under varying environmental conditions remains unclear. In this pilot study, a causal inference analysis was conducted using a Causal Tree to analyze the factors that influence the effect of salt-stress treatment on improving the °Brix in tomatoes. Data were collected from a single greenhouse using one cultivar over multiple cultivation periods, totaling 707 fruits. Propensity score matching was applied to reduce the covariate imbalance between the salt (NaCl) treatment and control groups. Using a Causal Tree, conditional average treatment effects were then estimated to assess heterogeneity in treatment impact based on environmental variables, such as temperature, vapor pressure deficit (VPD), and photosynthetically active radiation (PAR). Treatment with NaCl significantly increased °Brix compared to the control, but the magnitude of the effect varied depending on the environmental conditions. The Causal Tree analysis based on the accumulated cultivation data identified specific combinations of environmental factors under which the impact of NaCl treatment on °Brix enhancement was more pronounced. We estimated the Conditional Average Treatment Effect (CATE), defined as the difference in the proportion of producing high-sugar fruits (°Brix > 6%), between the NaCl treatment and control groups. For instance, the estimated CATE reached as high as 0.75 when the temperature was less than 25°C, the VPD was at least 0.84 kPa, and the PAR was below 13 mol ‧ m-2 ‧ d-1. In contrast, the CATE dropped to 0.031 when the temperature was between 20 and 25°C and the VPD was below 0.84 kPa. These results suggest that the interplay between temperature and VPD significantly modulates the efficacy of NaCl treatment for increasing tomato sugar content.