Abstract: This study explores the feasibility of estimating soil cadmium (Cd) content in farmland using GF-1 remote sensing satellite imagery. Correlation and different regression analyses have been separately conducted with sample Cd content and logarithmic, square root, and inverse square root transformation of image spectral image, which is filtered out vegetation information in the pre-processed remote sensing images. The linear regression model, with an accuracy above 0.95, is selected using competitive adaptive reweighted resampling based on the inverse square root transformation. The remote sensing estimates results, however, revealed a significant number of anomalous values in areas such as ponds, flooded rice fields, rooftops, hardened road surfaces and so on. An interpolation is employed with neighboring normal estimates to replace these anomalies, resulting in the final estimated values. Correlation analysis and modeling accuracy assessments suggest that this method is feasible and holds promise for practical applications in soil quality monitoring and land management.

Key words: cultivated soil, cadmium content, GF-1, spectral characteristics, inversion model