Root coumarin content is an important indicator for quality evaluation and harvest decision making in Qianhu, but its measurement requires destructive sampling and cannot be naturally paired with repeated non-destructive leaf observations. This paper proposes a multi-source temporal data model that aggregates leaf spectra, leaf chemical traits, water status, and root coumarin measurements by date, constructs real-date leaf-to-root pairs, and scores candidate lag centers using mechanism-informed evidence and pair support. The main analysis contains 580 sample records but only 10 independent date-level units. Under these sampling conditions, the model identifies a candidate 58–63 d lag window and uses 62 d as its representative center. Tolerance, candidate-grid, descriptor-role, and 2000-run weight-perturbation analyses retain dominant support near this region. Date-deletion analysis overlaps the window in 9 of 10 runs, although one removed date shifts the optimum, confirming that the result remains preliminary. At 62 d, the six linked pairs give an accumulation-proxy correlation of 0.928 with a 95% pair-bootstrap interval of 0.717–1.000. The model provides traceable temporal evidence for root coumarin analysis and harvest-timing support rather than a fixed biological period.

Qianhu, multi-source temporal data, near-infrared spectroscopy, leaf-to-root relationship, real-date pairing, lag-window mining, data processing model