Key points are not available for this paper at this time.
High-resolution irrigated cropland maps are essential for optimizing agricultural productivity and managing freshwater resources. China has the world's largest irrigated cropland area, but its rapid irrigation expansion in recent decades has exacerbated regional water stress. This study develops an automated machine learning framework to map irrigated croplands over the past three decades (1990 to 2020, per decade) at 30-m resolution across northern China, a region that accounts for more than 70% of China's irrigated cropland. The main contributions of our framework include: (1) delineating irrigated areas by constructing a composite irrigation performance index that integrates greenness, humidity, and temperature time-series features, refined with samples of two specific irrigated crop types; (2) incorporating synthetic phenological features and a 30 m surface temperature dataset to improve the classification performance of irrigated farmland by utilizing phenological changes and cooling effects caused by irrigation; (3) leveraging 98,697 Landsat scenes to train 673 localized random forest models for each time period to enhance classification accuracy across diverse regions. Validation results indicate that the overall accuracy of the generated irrigated cropland maps ranges from 0.78 to 0.85. Produced maps show strong agreement with statistical data and outperform three existing irrigation products. The 30 m resolution maps reveal that ~34,000 km² of irrigated cropland expansion has occurred predominantly in relatively arid and semi-arid zones. By minimizing reliance on statistical data and integrating irrigation-induced cooling effects and phenological variations, our approach achieves satisfactory mapping and provides a framework with broad application potential to deliver valuable data for regional irrigation and water resource management.
Li et al. (Wed,) studied this question.