Performance trade-offs between single-task and multi-task in multi-source prediction of yield and grain protein content in winter wheat

2026-08-10

PRE

Linghan Huang

Meiqi Zhang

Syed Tahir Ata-Ul-Karim

Kang Yu

Krzysztof Kuśnierek

Wei Li

Xiaojun Liu

Yongchao Tian

Yan Zhu

Weixing Cao

Qiang Cao *

DOI：10.1016/j.compag.2026.112295

Original

Original request for help

Save

Abstract

En 中文

• MTL outperformed STL with sparse features, while STL excelled with rich inputs. • MLP outperformed Transformer and RF with stage-aggregated tabular inputs. • The agronomic + sensor subset achieved the best yield and GPC accuracy. • Biomass, N status, water, and light were key at erecting and grain filling. • Weighted_Top8 subset retained high accuracy with fewer features and lower cost.

Keywords:

Feature integration

Multi-trait prediction

Agronomic-sensor trait synergies

Model interpretability

Efficient subset design

Performance trade-offs between single-task and multi-task in multi-source prediction of yield and grain protein content in winter wheat

Abstract

Journal

Organization

Cited Papers

Citing Papers