Self-Supervised Vision Transformers Improve Crop Yield Prediction from Satellite Imagery
American Journal Of Big Data · 25.03.2026 · 7 (1) · 252-265
DOI: 10.71465/ajbd3662 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal Of Big Data, online erschienen am 25.03.2026. Verfasst von Haoran Meng und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Accurate and timely crop yield prediction is critical for global food security, agricultural resource allocation, and early warning systems for potential supply disruptions. Conventional approaches based on process-based simulation models and handcrafted spectral features struggle to scale across diverse geographies and are heavily reliant on large volumes of labeled ground-truth data, which are expensive and laborious to collect. Recent advances in deep learning, particularly vision transformers (ViTs) and self-supervised learning (SSL), offer a promising avenue to overcome these limitations by enabling models to learn powerful spatiotemporal representations from abundant unlabeled satellite imagery. This paper proposes a framework that leverages SSL-pretrained ViTs fine-tuned on multi-temporal Sentinel-2 satellite imagery for county-level crop yield prediction across the U.S. Corn Belt. We systematically evaluate the proposed approach against supervised CNN baselines and LSTM networks. Experimental results demonstrate that SSL pre-training significantly reduces prediction error and improves data efficiency, achieving an RMSE of 4.71 bushels per acre and an R² of 0.89 for maize yield prediction with only 20% of available labeled data. This study underscores the potential of self-supervised ViT-based models to advance large-scale precision agriculture through scalable, label-efficient learning from satellite observations.
Zitieren
APA
Meng, H., Pan, Z., & Lindström, S. (2026). Self-Supervised Vision Transformers Improve Crop Yield Prediction from Satellite Imagery. American Journal Of Big Data, 7(1), 252-265. https://doi.org/10.71465/ajbd3662
Vancouver
Meng H, Pan Z, Lindström S. Self-Supervised Vision Transformers Improve Crop Yield Prediction from Satellite Imagery. American Journal Of Big Data. 2026;7(1):252-265. doi:10.71465/ajbd3662
BibTeX
@article{Meng2026,
author = {Meng, Haoran and Pan, Zixu and Lindström, Sofia},
title = {Self-Supervised Vision Transformers Improve Crop Yield Prediction from Satellite Imagery},
journal = {American Journal Of Big Data},
year = {2026},
volume = {7},
number = {1},
pages = {252--265},
doi = {10.71465/ajbd3662}
}
EUVIMED