Diverse-Input Yield Prediction (DIYP) Model Applied Across Different Years With Variable Input Days
Journal
IEEE ACCESS
Journal Volume
14
Start Page
86221
End Page
86239
ISSN
2169-3536
Date Issued
2026-06
Author(s)
Abstract
The drone-collected multispectral data are useful in agriculture applications, particularly for crop yield prediction. However, the drone flying is often impacted by weather conditions, resulting in data miss or inaccurate information. In this paper, we propose a new rice yield prediction model, named Diverse-Input Yield Prediction (DIYP), which is based on the LSTM network and further enhanced with attention mechanisms and an encoder structure to address the varied input and cross-year prediction challenges. Our model can reduce the impact of missing data, and improve the yield prediction accuracy across different years. All rice datasets are provided by Taiwan Agricultural Research Institute, which consist of multispectral vegetation index data collected by drones. We built our DIYP model using only the 2020 dataset. The experimental results show that the proposed DIYP model outperforms state-of-the-art approaches across multiple metrics, achieving significant performance gains for both the 2021 and 2022 datasets. For the 2021 dataset, the RMSE decreases by 9-15%, while the ACC increases by 9-13%. Similarly, for the 2022 dataset, the RMSE is reduced by 5-23%, and the ACC is improved by 9%. Furthermore, our model achieves Kappa coefficients of 0.83 and 0.84 for 2021 and 2022, respectively, indicating strong agreement with ground truth measurements. In addition to conventional performance metrics, RMSE-based model selection criteria, including AIC, SBIC, HQIC, and AICc, were incorporated to provide a more rigorous comparison among the related models. The information-criterion-based results further confirm that DIYPv3 provides a favorable balance between prediction accuracy and model complexity.
Subjects
Crop
vegetation
missing data
time-series modeling
rice yield prediction
Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
Type
journal article