Spectral Scaling Effects of ZY1-02D Satellite Imagery on Landscape Diversity Assessment in the Yellow River Delta: Using Harmonic Analysis and Interpretable XGBoost

Landscape diversity assessment is important for ecological monitoring, and hyperspectral remote sensing provides an effective means for regional-scale analysis. However, the effects of spectral resolution on landscape diversity representation remain unclear. Whether frequency-domain features provide complementary information also requires further investigation. This study focused on the Yellow River Delta, China. Landscape diversity was characterized using landscape richness, the Shannon-Wiener index, and the Simpson index derived from vegetation cover classification. Based on fused hyperspectral imagery, Gaussian spectral response functions were used to simulate different spectral resolutions. Spectral vegetation indices, spectral diversity indicators, and harmonic frequency-domain indicators were extracted. XGBoost was used to compare their performance across spectral resolutions, while SHAP was used to interpret feature contributions. Model performance generally declined as spectral resolution decreased, with R2 values ranging from 0.5148 to 0.6360 for the Shannon-Wiener index, 0.4947 to 0.5850 for the Simpson index, and 0.5037 to 0.6044 for landscape richness. Compared with models using spectral-domain features alone, adding harmonic frequency-domain indicators increased R2 by an average of 18.79% across all spectral resolutions and diversity metrics. The mean relative increase in R2 was 23.35% at the 70 nm, 80 nm, and multispectral resolutions, compared with 15.28% at the original, 20 nm, and 30 nm resolutions. SHAP analysis showed that spectral-domain indicators remained important across the evaluated resolutions, while frequency-domain indicators provided complementary information for landscape diversity characterization. These findings indicate that integrating spectral and frequency-domain information improves the characterization of landscape diversity across spectral resolutions, particularly under reduced spectral resolution.