Wheat yield forecasting using deep learning: A comparison between unstructured and structured data

D. Al-Shammari; P. Filippi; S. Poole; S. Han; T.F.A. Bishop

Wheat yield forecasting using deep learning: A comparison between unstructured and structured data

In: Precision agriculture '25

Authors:

D. Al-Shammari

D. Al-Shammari dhahi.al-shammari@sydney.edu.au

Search for other papers by D. Al-Shammari in
Current site
Google Scholar
PubMed

P. Filippi

P. Filippi
Search for other papers by P. Filippi in
Current site
Google Scholar
PubMed

S. Poole

S. Poole
Search for other papers by S. Poole in
Current site
Google Scholar
PubMed

S. Han

S. Han
Search for other papers by S. Han in
Current site
Google Scholar
PubMed

, and

T.F.A. Bishop

T.F.A. Bishop
Search for other papers by T.F.A. Bishop in
Current site
Google Scholar
PubMed

Type:: Chapter

Pages:: 1365–1371

DOI:: https://doi.org/10.1163/9789004725232_178

Access via:

Dar Hadith al Hassania

Download PDF

Deep learning (DL) models can utilise unstructured (pixel-based) data rather than exclusively depending on structured (organised in rows and columns) data, which may enhance the model’s capacity to capture spatial dependencies and non-linear patterns. Two DL models (a fully connected layer model and TabNet) were employed for within-field yield forecasting using structured data, while a 3D-convolutional neural network (3D-CNN) leveraged unstructured data. An extreme gradient boosting (XGBoost) model, a more traditional decision-tree-based ML model, was also utilised to benchmark the DL models. A leave-one-season-out cross-validation (LOSOCV) approach was implemented, considering each field individually where all but one season was used for training, with the final season reserved for forecasting. The results indicated that the 3D-CNN, using limited yet representative unstructured data, surpassed all other models evaluated in this study. By capturing local spatio-temporal patterns, the 3D-CNN achieved the highest Lin’s concordance correlation coefficient (CCC) in three out of four fields. Future work should aim to integrate ancillary data (e.g., management practices) to better capture the underlying complexity at finer scales.

Citation Info

Save
Cite
Email this content

Share link with colleague or librarian

You can email a link to this page to a colleague or librarian:
Email this content
or copy the link directly:

https://brill.edhh.ma/edcollchap-oa/book/9789004725232/BP000178.xml
The link was not copied. Your current browser may not support copying via this button.

Link copied successfully

Collapse
Expand

Precision agriculture '25

E-Book ISBN:: 9789004725232

Publisher:: Wageningen Academic

Print Publication Date:: 18 Jun 2025

Subjects
- Life Sciences
  - General
  - Agriculture & Environment

Metrics

	All Time	Past 365 days	Past 30 Days
Abstract Views	0	0	0
Full Text Views	95	86	9
PDF Views & Downloads	145	126	5

African Studies	Education	Media Studies
American Studies	History	Middle East and Islamic Studies
Ancient Near East and Egypt	Human Rights and Humanitarian Law	Musicology
Art History	International Law	Philosophy
Asian Studies	International Relations	Religious Studies
Biblical Studies	Jewish Studies	Slavic and Eurasian Studies
Biology	Languages and Linguistics	Social Sciences
Book History and Cartography	Life Sciences	Theology and World Christianity
Classical Studies	Literature and Cultural Studies

Subjects

Authors

Open Access

Product Information

Company

Contact

Wheat yield forecasting using deep learning: A comparison between unstructured and structured data

Citation Info

Share link with colleague or librarian

Precision agriculture '25

Table of Contents

Metrics

Metrics