The influence of estimation window configuration on machine learning-based soybean yield estimation across black soil regions
文献类型：期刊论文
| 作者 | Huang, Shuyuan1,2; Liu, Yujie1,2,3; Chen, Jiahao1,2; Zhang, Ermei1,2; Pan, Tao1,2 |
| 刊名 | AGRICULTURAL AND FOREST METEOROLOGY
 |
| 出版日期 | 2026-03-01 |
| 卷号 | 378页码:110957 |
| 关键词 | Machine learning Phenological model Soybean yield estimation Sowing date Growing season length |
| ISSN号 | 0168-1923 |
| DOI | 10.1016/j.agrformet.2025.110957 |
| 产权排序 | 1 |
| 文献子类 | Article |
| 英文摘要 | The configuration of phenology-based time windows, which determines how environmental variables are temporally aggregated, plays a pivotal role in crop yield estimation. However, the quantitative effects of different window configurations on model performance and uncertainty require further investigation. This study systematically assesses the effects of four window-configuration strategies (fixed, observation, rule-based, and sliding) on soybean yield estimation across the black soil regions of China and the USA. Multi-source remote sensing and meteorological datasets were integrated with three machine learning algorithms: RF, XGBoost, and LSTM. Results show that dynamic windows (observation, rule-based, and sliding) can better align environmental fluctuations with crop phenological stages, resulting in modest yet consistent improvements in accuracy compared to fixed windows. The LSTM-sliding window combination achieves the largest RMSE decrease (48.456.6%), followed by LSTM-rule-based windows (32.9-38.2%) and LSTM-observation windows (11.8-22.0%). A trade-off is identified: while sliding windows (SWs) provide the highest accuracy, they also show greater interannual variability, higher computational cost, and lower interpretability. In comparison, rule-based windows (RBWs) exhibit a moderate decline in accuracy but demonstrate lower inter-group variability, with Delta R2 approximately one-third that of SW, offering more stable predictions. RBWs also exhibit better generalizability than observation windows, which rely on limited ground phenology data. Uncertainty decomposition reveals that, although the primary source of variation originates from input features and model structures, the configuration of the estimation window contributes approximately 11.9-13.7% to the total variation, indicating a secondary yet consistent factor influencing estimation stability. This study offers an analytical framework for quantifying the interactions among window design, algorithm type, and feature selection, thereby providing practical insights for future data-driven crop yield modeling. |
| URL标识 | 查看原文 |
| WOS关键词 | CHINA ; EARTH |
| WOS研究方向 | Agriculture ; Forestry ; Meteorology & Atmospheric Sciences |
| 语种 | 英语 |
| WOS记录号 | WOS:001636486500001 |
| 出版者 | ELSEVIER |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/219759]  |
| 专题 | 陆地表层格局与模拟院重点实验室_外文论文 |
| 通讯作者 | Liu, Yujie |
| 作者单位 | 1.Univ Chinese Acad Sci, Beijing 100049, Peoples R China; 2.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, Beijing 100101, Peoples R China; 3.Chinese Acad Sci, Key Lab Mt Hazards & Engn Resilience, Chengdu 610213, Peoples R China |
| 推荐引用方式 GB/T 7714 | Huang, Shuyuan,Liu, Yujie,Chen, Jiahao,et al. The influence of estimation window configuration on machine learning-based soybean yield estimation across black soil regions[J]. AGRICULTURAL AND FOREST METEOROLOGY,2026,378:110957. |
| APA | Huang, Shuyuan,Liu, Yujie,Chen, Jiahao,Zhang, Ermei,&Pan, Tao.(2026).The influence of estimation window configuration on machine learning-based soybean yield estimation across black soil regions.AGRICULTURAL AND FOREST METEOROLOGY,378,110957. |
| MLA | Huang, Shuyuan,et al."The influence of estimation window configuration on machine learning-based soybean yield estimation across black soil regions".AGRICULTURAL AND FOREST METEOROLOGY 378(2026):110957. |
入库方式： OAI收割
来源：地理科学与资源研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。