As a typical agroforestry ecosystem, farmland shelterbelts (FSs) play a pivotal role in delivering ecological and economic benefits and have been widely implemented worldwide. However, their fragmented distribution, irregular boundaries, and strong spectral similarity to adjacent land-cover types lead to severe spectral confusion in remote sensing imagery, posing significant challenges for accurate regional identification and monitoring. To address these issues, we propose a Shelterbelt Phenological Classification (SPC) framework that leverages the phenological characteristics of FSs. By integrating multi-temporal Sentinel-2 spectral, textural, and change information, the framework effectively captures the dynamic characteristics of FSs in Northeast China and, in combination with four mainstream classifiers, generates 10-m spatiotemporal distribution maps for 2019–2023. The results indicate a continuous increase in the total length of shelterbelts in Northeast China during the study period, with the proposed method achieving an overall accuracy of 90.2% and a Kappa coefficient of 0.88. This study provides a novel and effective solution for regional-scale mapping and monitoring of FSs in Northeast China and offers valuable methodological references for remote sensing-based monitoring of agroforestry ecosystems in the region. • A novel framework (SPC) suitable for large-scale shelterbelt classification was proposed. • Phenological window optimization and spectral-textural feature fusion markedly improved shelterbelt separability and classification accuracy. • The optimal results were achieved by the Random Forest algorithm among the four compared classification models. • A 10-m annual shelterbelt dataset for Northeast China from 2019 to 2023 was produced, revealing an overall increasing trend in shelterbelt length.
Chen et al. (2026) studied this question.