Long history paddy rice mapping across Northeast China with deep learning and annual result enhancement method

• DOI: 10.5194/essd-2024-516
• Publication Date: 2025-01-29T12:58:25Z
• Authors: Zihui Zhang; Lang Xia; Fen Zhao; Yue Gu; Jing Yang; Yan Zha; Shangrong Wu; Peng Yang
• Keywords: Paddy field

Abstract. Northeast China, a significant production base for paddy rice, has received lots of attention in crop mapping. However, understanding the spatiotemporal dynamics rice expansion this region remains limited, making it difficult to track changes planting over time. For first time, study utilized multi-sensor Landsat data and deep learning model, full resolution network (FR-Net), explore annual mapping China from 1985 2023 (available at https://doi.org/10.6084/m9.figshare.27604839.v1, Zhang et al., 2024). First, cross-sensor training dataset comprising 155 images was created map rice. Then, we developed result enhancement (ARE) method, which considers differences category probability FR-Net different stages diminish impact limited sample large-scale across-sensors The accuracy evaluated using 107954 ground truth samples. In comparison traditional methods, results obtained ARE method showed 6 % increase F1 score. overall model methods achieved high user (UA), producer (PA), score, Matthews correlation coefficient (MCC) values 0.92, 0.95, 0.93, 0.81, respectively. revealed that area used cultivation increased 1.11×104 km2 6.45×104 km2. Between 2023, there an 5.34×104 area, with highest growth (4.33×104 km2) occurring Heilongjiang province. This shows long-history could be learning, will beneficial timely adjustments patterns ensuring food security.