Improved Soybean Mapping with Spectral Gaussian Mixture Modeling
Soybeans are a globally important crop, both commercially and nutritionally. Accurate mapping of soybean cultivation is essential for optimizing production and informing market strategies. However, traditional sample-driven soybean mapping algorithms often rely on extensive, representative datasets,...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
American Association for the Advancement of Science (AAAS)
2025-01-01
|
| Series: | Journal of Remote Sensing |
| Online Access: | https://spj.science.org/doi/10.34133/remotesensing.0473 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Soybeans are a globally important crop, both commercially and nutritionally. Accurate mapping of soybean cultivation is essential for optimizing production and informing market strategies. However, traditional sample-driven soybean mapping algorithms often rely on extensive, representative datasets, which can limit their applicability across different regions and periods. In contrast, existing sample-free soybean mapping methods have yet to fully exploit key physiological traits, such as chlorophyll content, canopy greenness, and water content, that are essential for distinguishing soybeans from other crops, particularly during peak growth stages when many crops share similar spectral characteristics. To address these limitations, this study introduces an innovative approach: the spectral Gaussian mixture model (SGMM) for global-scale soybean mapping. Specifically, the SGMM develops a novel Bhattacharyya coefficient weighting method to optimize spectral probabilistic separability between soybeans and other crops. Moreover, it identifies an accurate soybean mapping timeframe, named the optimal time window, to refine spectral feature extraction across varying environmental conditions and crop calendars. Unlike previous methods that rely on fixed thresholds or a limited set of spectral indices, our SGMM offers a probabilistic mapping framework that dynamically adapts to regional variations in soybean growth. The SGMM was validated across multiple soybean-producing regions, showing high accuracy with average overall accuracies of 0.875 in China, 0.907 in the United States, 0.895 in Argentina, and 0.884 in Brazil. Furthermore, the provincial-level estimates of soybean areas correlated strongly with official statistics, highlighting the model’s reliability and scalability for global soybean mapping. By leveraging key physiological insights and optimizing spectral feature extraction, the SGMM provides an efficient, scalable solution for global agricultural monitoring and can serve as a reference for mapping other crops. |
|---|---|
| ISSN: | 2694-1589 |