Multi-Source Transformer-Based Crop Yield Forecasting Across Canadian Agro-Ecological Zones: A Comparative Evaluation of CanadaYieldNet Against Deep Learning and Operational Baselines with Systematic Data Modality Ablation
International Journal of Emerging Technologies and Innovative Research · 27.07.2026 · 53
DOI: 10.48175/ijetir-6656 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Emerging Technologies and Innovative Research, online erschienen am 27.07.2026. Verfasst von Seema Sharma and Dr. Praveen Gupta. Die Arbeit nennt 35 Literaturangaben.
Abstract (Originalsprache)
Accurate forecasting of crop yields at national scales remains a cornerstone of food security planning, agricultural market stability, and climate adaptation policy, yet existing deep learning approaches seldom exploit the full breadth of remote sensing modalities or undergo systematic evaluation across Canada’s heterogeneous agro-ecological zones. To address this gap, we introduce CanadaYieldNet, a transformer-based framework that fuses five complementary data sources—optical imagery from Sentinel-2 and MODIS, synthetic aperture radar from the RADARSAT Constellation Mission, soil moisture from NASA SMAP, weather reanalysis from ERA5-Land, and static soil and terrain attributes from CanSIS and CDEM—through a carefully designed architecture combining a Temporal Fusion Transformer, Vision Transformer, static feature encoder, and an Adaptive Gated Fusion layer with multi-head cross-attention. Evaluated across six provinces and five major crops between 2014 and 2024, the system consistently outperformed thirteen established baselines, achieving mean \(R^2 = 0.94\), RMSE = 4.2 bu/acre, and MAPE = 6.8%, with optical remote sensing and weather data emerging as the most influential modalities in ablation studies. Crucially, the model demonstrated resilience under climatic extremes, maintaining predictive accuracy during droughts and excess moisture events, while SHAP and attention visualisation confirmed agronomically coherent drivers such as peak NDVI and growing-degree days. By uniting methodological innovation with interpretability, CanadaYieldNet establishes a new benchmark for national-scale yield forecasting in Canada, offering both robust predictive capacity and actionable insights for agricultural decision-makers, with all code and processed datasets released under open licences to ensure transparency and reproducibility.
Zitieren
APA
Gupta, S. S. A. D. P. (2026). Multi-Source Transformer-Based Crop Yield Forecasting Across Canadian Agro-Ecological Zones: A Comparative Evaluation of CanadaYieldNet Against Deep Learning and Operational Baselines with Systematic Data Modality Ablation. International Journal of Emerging Technologies and Innovative Research, 53. https://doi.org/10.48175/ijetir-6656
Vancouver
Gupta SSADP. Multi-Source Transformer-Based Crop Yield Forecasting Across Canadian Agro-Ecological Zones: A Comparative Evaluation of CanadaYieldNet Against Deep Learning and Operational Baselines with Systematic Data Modality Ablation. International Journal of Emerging Technologies and Innovative Research. 2026:53. doi:10.48175/ijetir-6656
BibTeX
@article{Gupta2026,
author = {Gupta, Seema Sharma and Dr. Praveen},
title = {Multi-Source Transformer-Based Crop Yield Forecasting Across Canadian Agro-Ecological Zones: A Comparative Evaluation of CanadaYieldNet Against Deep Learning and Operational Baselines with Systematic Data Modality Ablation},
journal = {International Journal of Emerging Technologies and Innovative Research},
year = {2026},
pages = {53},
doi = {10.48175/ijetir-6656}
}
EUVIMED