ESTIMATION OF BIOPHYSICAL PARAMETERS USING HYPERSPECTRAL REMOTE SENSING

Abstract

In this research, two different types of nitrogen fertilizer application (uniformly 30 kg/ha and variably from 0 kgN/ha to 70 kgN/ha with the difference of 10 kgN/ha) are supplied at basal dressing in order to nitrogen contents of rice plant already possessed at panicle initiation stage using hyperspectral sensor (AISA+, 400nm~1000nm, 68bands) to calculate the optimum amount of nitrogen fertilizer at topdressing. The hyperspectral reflectance at panicle initiation stage was compared with the various crop physiological parameters, such as amount of nitrogen fertilizer at basal dressing, dry weight, leaf area index (LAI), and nitrogen contents. The coefficient of correlation between the hyperspectral reflectance of each band and the field data were more than 0.56 negatively at 420 nm ~ 500 nm region, 0.64 negatively at 575 nm ~ 682 nm regions and 0.60 at 726 nm ~ 991 nm regions, except the SPAD value and the nitrogen contents percentage of leaf and stem. It is possible to make nitrogen contents estimation model to predict the nitrogen contents of 15 plots, which is supplied 8 degrees of nitrogen fertilizer doses using the multi regression analysis with forward stepwise regression method. The multi regression models resulted in R2 values of 0.85 with 5% of significant level. As a result, it is possible to estimate nitrogen contents of rice plant using the nitrogen contents estimation model, which is calculated by multi regression analysis using the 15 data with 8 degrees of nitrogen fertilizer doses, even if there are low spatial variability of nitrogen contents, which is supplied with 30 kg/ha nitrogen fertilizer at basal dressing. The maximum error is 0.971g/m2.