Work

A Multi-Temporal Phenology-Driven Framework for Long-Term Multi-Crop Mapping in Hubei, China

Year: 2026
Type: conference-paper
Abstract: Reliable, long-term multi-crop maps are critical for sustainable agricultural management and food security. However, existing multiple crop mapping approaches are often limited by severe inter-class confusion caused by the spectral similarities among different crops in complex agricultural landscapes. To address this challenge, we propose a multi-temporal, phenology-driven framework (MPDNet) for long-term multi-crop mapping. Our framework explicitly incorporates intra-annual phenological dynamics, thereby exploiting the distinct temporal-spectral trajectories across growth stages within a year. First, we developed a phenology-aware module that leverages the positional encoding and global context modeling capabilities of Transformer architectures to sequentially fuse multi-temporal satellite observations at monthly time steps. Then, the multi-level backbone and lightweight decoder utilize these phenological distinctions to generate refined crop-type maps. Finally, we applied this framework to Hubei Province, China, generating an annual, multi-season dataset from 2001-2025 covering five dominant crops (rice, wheat, maize, soybean, and rapeseed). The mapping results show strong agreement with official statistics and achieve excellent field segmentation quality and spatial continuity compared with existing multi-crop products. Overall, this framework significantly improves multi-crop mapping performance and provides a consistent, long-term dataset to support downstream agricultural applications such as yield estimation and policy analysis.
Institution: Wuhan University
Language: English
Subfield: Ecology
Open Access status: hybrid
OpenAlex ID: W7204898703
FWCI:
Cites: 18
Cited by:
Related to: 10