A geospatial graph neural network framework for analyzing climate variability impacts on crop yields in northeast Thailand
บทความในวารสาร
ผู้เขียน/บรรณาธิการ
กลุ่มสาขาการวิจัยเชิงกลยุทธ์
รายละเอียดสำหรับงานพิมพ์
รายชื่อผู้แต่ง: Waqas, M., Humphries, U. W., Wangwongchai, A., Ahmad, S., Dechpichai, P., Hlaing, P. T.
ผู้เผยแพร่: Springer
ปีที่เผยแพร่ (ค.ศ.): 2026
วารสาร: Theoretical and Applied Climatology (0177-798X)
Volume number: 157
หน้าแรก: 1
หน้าสุดท้าย: 19
จำนวนหน้า: 19
นอก: 0177-798X
eISSN: 1434-4483
URL: https://link.springer.com/article/10.1007/s00704-026-06166-2
ภาษา: English-Great Britain (EN-GB)
บทคัดย่อ
This study aims to quantify and predict the impacts of El Niño–Southern Oscillation (ENSO) phases on rainfed rice yields in Northeastern (NE) Thailand using a deep learning (DL) framework. It introduces the Geospatial-Aware Graph Neural Network for Yield Prediction (GeoGNN-Yield), a spatiotemporal model that captures complex climate–yield interactions. The model was trained using a 30-year dataset (1993–2022) of provincial rice yields and climate variables, including precipitation, maximum and minimum temperature (Tmax and Tmin), relative humidity (RH), and ENSO indices. Mann–Kendall (MK) trend analysis detected significant trends (p < 0.05) in 48.80% of the 1,308 records, with Tmin increasing at 0.020 °C year⁻¹ and RH declining at 0.051% year⁻¹. Canonical correlation analysis (CCA) showed ENSO–yield correlations ranging from 0.001 to 0.38, with Niño 4 being most influential. GeoGNN-Yield achieved strong predictive performance (R² = 0.94 training; 0.86 testing) and revealed spatially opposite but symmetric El Niño and La Niña yield anomalies (mean absolute change = 78.90%), demonstrating its value for climate-resilient agricultural planning in NE Thailand.
คำสำคัญ
Deep learning, El Niño–Southern Oscillation (ENSO), Graph Neural Network (GNN), Northeastern Thailand, Rainfed rice yield






