Electrical-Response Prediction of Leaf-Based Electric-Field Energy Harvesters Using Dehydration-Related Variables and Synthetic Data

Crop-monitoring systems require sustainable, self-powered sensing solutions. Leaf-based Electric-Field Energy Harvesters (Leaf-EFEHs) offer a potential approach; however, the relationship between foliar moisture content (FMC) and their electrical response remains insufficiently understood. This work characterizes this relationship in Brassica rapa pekinensis and Spinacia oleracea Leaf-EFEHs under controlled dehydration and exploratorily evaluates whether synthetic data can provide incremental improvements in the prediction of open-circuit voltage (VOC) and short-circuit current (ISC). Fifty leaves per species were electrically characterized at eight dehydration stages, producing two species-specific datasets of 400 observations each. Five regression models and five synthetic-data generation methods were evaluated using measurement-level partitioning and 20 repeated leaf-level partitions. The electrical response varied nonlinearly during dehydration, with an initial increase followed by a decline in Brassica rapa pekinensis and an overall decline followed by stabilization in Spinacia oleracea. Under measurement-level partitioning for Brassica rapa pekinensis, Random Forest achieved the highest baseline performance, with R2=0.865 for VOC and R2=0.859 for ISC. Gaussian noise produced the smallest distributional discrepancies, while Mixup best preserved the correlation structure. Synthetic-data augmentation did not consistently improve predictive performance. Under leaf-level validation, temporal interpolation produced the largest gain for the Brassica rapa pekinensis MLP predicting VOC (ΔR2=+0.0619), whereas Gaussian noise produced the largest Spinacia oleracea gain for the MLP predicting ISC (+0.0330). Overall, augmentation utility was model-, strategy-, and species-dependent, emphasizing the need to evaluate synthetic data through predictive utility, preservation of the dehydration process, and generalization to unseen leaves.