Assessing the Consistency of Five Remote Sensing-Based Land Cover Products for Monitoring Cropland Changes in China
Fuliang Deng,Xinqin Peng,Jiale Cai,Lanhui Li,Fangzhou Li,Chen Liang,Wei Liu,Ying Yuan,Mei Sun
DOI: https://doi.org/10.3390/rs16234498
IF: 5
2024-12-01
Remote Sensing
Abstract:The accuracy assessment of cropland products is a critical prerequisite for agricultural planning and food security evaluations. Current accuracy assessments of remote sensing-based cropland products focused on the consistency of spatial patterns for specific years, yet the reliability of these cropland products in time-series analysis remains unclear. Using cropland area data from the second and third national land surveys of China (referred to as NLSCD) as a benchmark, we evaluate the area-based and spatial-based consistency of cropland changes in five 30 m time-series land cover products covering 2010 and 2020, including the annual cropland dataset of China (CACD), the annual China Land Cover Dataset (CLCD), China's Land-use/cover dataset (CLUD), the Global Land-Cover product with Fine Classification System (GLC_FCS30), and GlobeLand30. We also employed the GeoDetector model to explore the relationships between the consistency in cropland change and the environmental factors (e.g., cropland fragmentation, topographic features, frequency of cloud cover, and management practices). The area-based consistency analysis showed that all five cropland products indicate a declining trend in cropland areas in China over the past decade, while the amount of cropland loss ranges from 5.59% to 57.85% of that reported by the NLSCD. At the prefecture-level city scale, the correlation coefficients between the cropland area changes detected by five cropland products and the NLSCD are low, with GlobeLand30 having the highest coefficient at 0.67. The proportion of prefecture-level cities where the change direction of cropland area in each cropland product is inconsistent with the NLSCD ranges from 13.27% to 39.23%, with CLCD showing the highest proportion and CLUD the lowest. At the pixel scale, the spatial-based consistency analysis reveals that 79.51% of cropland expansion pixels and 77.79% of cropland loss pixels are completely inconsistent across five cropland products, with the southern part of China exhibiting greater inconsistency compared to Northwest China. Besides, the frequency of cloud cover and management practices (e.g., irrigation) are the primary environmental factors influencing consistency in cropland expansion and loss, respectively. These results suggest low consistency in cropland change across five cropland products, emphasizing the need to address these inconsistencies when generating time-series cropland datasets via remote sensing.
environmental sciences,imaging science & photographic technology,remote sensing,geosciences, multidisciplinary