Abstract: Four independent field experiments with 6 wheat varieties and 5 nitrogen application levels were conducted, and time-course measurements were taken on the canopy hyperspectral reflectance and leaf N accumulation per unit soil area (LNA, g N·m^-2). By adopting reduced precise sampling method, all possible normalized difference spectral indices ［NDSI（i, j）］ within the spectral range of 350-2500 nm were constructed, and the relationships of LNA to the NDSI（i, j）were quantified, aimed to explore the new sensitive spectral bands and key index from precise analysis of groundbased hyperspectral information, and to develop prediction models for wheat LNA. The results showed that the sensitive spectral bands for LNA were located in visible light and near infrared regions, especially at 860 nm and 720 nm for wheat LNA. The monitoring model based on the NDSI(860,720) was formulated as LNA=26.34×［NDSI(860,720)］^1.887, with R²=0.900 and SE=1.327. The fitness test of the derived equations with independent datasets showed that for wheat LNA, the model gave the estimation accuracy of 0.823 and the RMSE of 0.991 g N·m^-2, indicating a good fitness between the measured and estimated LNA. The present normalized hyperspectral parameter of NDSI(860,720) and its derived regression model could be reliably used for the estimation of winter wheat LNA.

Key words: hyperspectral reflectance, normalized difference spectral index, monitoring, wheat, leaf nitrogen accumulation, Ebinur Lake region, landscape pattern, ecological risk, spatio-temporal variation.