Coupling machine learning with a biophysical model for maturity date prediction of apple fruit across China's apple planting regions.

Authors

Renwei Chen, Jing Wang, Zhihong Gong, Jianshuo Zhou, Yang Li, Rui Bai, Zhenjiang Qu, Zhangyan Le

Published in

Journal of the science of food and agriculture. Aug 01, 2026. Epub Aug 01, 2026.

Abstract

Accurate prediction of apple fruit maturity date is essential for optimizing harvest timing, fruit quality and market value under climate change. However, process-based crop models often show limited performance when extrapolated across large spatial scales, whereas machine learning models lack physiological interpretability. To address these limitations, this study has developed a hybrid framework integrating the process-based STICS model with machine learning approaches across China's apple planting regions.
Phenological observations from 24 sites and meteorological data from 250 stations during 1991-2020 were used to calibrate and evaluate six machine learning models. Among them, the random forest (RF) model achieved the best performance [coefficient of determination (R²) > 0.65, root mean square error (RMSE) < 8.1 days]. A hybrid approach was implemented by incorporating STICS-simulated maturity dates as input features into the machine learning models, enabling the capture of residual non-linear relationships between maturity dates of apple fruit and climatic and geographic variables. This integration further improved prediction accuracy (R² > 0.71, RMSE < 7.5 days), reducing errors by over 50% compared to the standalone STICS model. Spatially, the average maturity date was 282.5 ± 7.0 DOY (i.e. day of year), with the latest maturity in the Yellow River region and the earliest in the Southwest highlands. Temporally, maturity dates advanced slightly at 0.1 days decade^-1, with substantial regional variability. SHAP (i.e. Shapley Additive exPlanations) analysis identified chilling requirement, elevation and STICS-simulated maturity date as dominant drivers.
The hybrid STICS+RF framework effectively combines mechanistic understanding with data-driven learning, improving prediction accuracy and interpretability of apple fruit maturity date at regional scales. This approach provides a robust tool for optimizing harvest management and adapting apple production systems to climate change. © 2026 Society of Chemical Industry.

PMID:
42538841
Bibliographic data and abstract were imported from PubMed on 01 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.