Understanding the potentials of early-season crop type mapping by using Landsat-8, Sentinel-1/2, and GF-1/6 data

• Published in: Computers and electronics in agriculture Vol. 224; p. 109239
• Main Authors: Wang, Cong, Zhang, Xinyu, Wang, Wenjing, Wei, Haodong, Wang, Jiayue, Li, Zexuan, Li, Xiuni, Wu, Hao, Hu, Qiong
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2024
• Subjects: Crop mapping; Early identification; GF-1/6; Landsat-8; Sentinel-1/2; Early identification; Landsat-8; GF-1/6; Sentinel-1/2; Crop mapping
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2024.109239

•12 scenarios of using different satellite images for crop recognition were compared.•Combining GF-1, GF-6, Sentinel-1, Sentinel-2 and Landsat-8 identified crops earliest.•Early identification depends on feature properties, crop type, and climate condition. Early-season crop identification by remote sensing is challenging due to insufficient spectral and temporal information at early stage, especially when relying on single satellite data. Although multi-source datasets may offer additional useful information for classification, the specific significance of different satellite data and their combinations in early-season crop type mapping remains largely unclear. This study investigated the potential of integrating publicly available medium-and high-resolution image data (i.e., Landsat-8, Sentinel-1/2 and GF-1/6) on early-season crop type mapping, with Longjiang County in Heilongjiang Province, China serving as the study area. The results showed that among five single data sources, Sentinel-2 provided the earliest identification (F1-score exceeded 0.9) of rice, followed by GF-1 and Sentinel-1. In terms of corn identification, GF-6 demonstrated the earliest identifiable capability, followed by GF-1 and Sentinel-2. Combining multi-source datasets proves to be more effective for early-season crop classification compared to using single-source datasets. Among all 12 scenarios, the integration of GF-1, GF-6, Sentinel-1, Sentinel-2, and Landsat-8 yielded the best performance in early-season crop identification, achieving accurate identification of rice at the transplanting stage (4 months ahead of harvest) and corn at the heading stage (2 months before harvest). Feature separability analysis further revealed that the crucial spectral/temporal features for specific crop types and image availability related to climate conditions were the main factors affecting early-season crop identification. This work can provide valuable insights for selecting various satellite datasets to facilitate early crop identification and enhances our understanding of the possibilities in early-season crop type mapping by leveraging medium- and high-resolution satellite data.