A Study on Crop Phenotype Prediction by Integrating Environmental Data Collection and AI-Based Data Augmentation Techniques

Global food security challenges require advanced breeding strategies that integrate genomics, phenomics, and artificial intelligence.This study aims to improve phenotype prediction accuracy in tomato breeding by leveraging genotype data augmentation and semi-supervised learning.A total of 192 tomato accessions were cultivated under greenhouse conditions, and genotypic, phenotypic, and environmental data were collected for five key traits: fruit weight, height, width, firmness, and brix.We propose a 1D CNN-based genotype augmentation framework to expand the original dataset and a pseudo-labeling strategy to effectively utilize unlabeled data.Environmental variables such as temperature, humidity, and others were integrated through statistical feature extraction over the growth period to better reflect cultivation conditions.Phenotype prediction was performed using 18 regression models, including both tree-based and deep learning architectures, and the impact of different network structures was comparatively evaluated.Results show that genotype augmentation consistently improved predictive performance, with tree-based models such as LightGBM and CatBoost exhibiting the largest gains.Additional comparisons with state-of-the-art models confirmed the robustness of the proposed approach.These findings providing a practical strategy for data-limited breeding scenarios and scalable integration with multi-omics and environmental datasets.