ASSESSING THEHYPERSPECTRAL REMOTE SENSING DATA TO DIAGNOSIS CROP VARIABLES MODEL IN TROPICAL IRRIGATED WETLAND RICE

Abstract

Canopy spectral measureme~t using ground-based hyperspectral r;levice and rice crop variables such as leaf area index (LAI), leaf dry weight (LDW) and SPAD values were done periodically during growth season with involving three rice cultivars (Pandanwangi, Ciherang and IR Jumbo) ahd four nitrogen application levels (NO,N80, N92 and NI03 kg/ha). Thestudy is directed to explore all possible waveband combinations tested in reflectance of vegetation indices (VIs) and to develop a predictive model of relation between hyperspectral-based vegetation indiceswith rice crop variables. . Analysis of all possible two-pair waveband combinations used in VIs was investigated with 6,786 combinations to gain optimal waveband attributed to crop variables. To develop.efficient and accurate model, various multivariate regression models were examined with ten-fold cross validations. Accuracy validation of predicted model was performed using reflectance and FDR, NDVI, RVI, RDVI and SA VI data. Validation of predictive model using flJR implied better accuracy to estimate LAI using whole season data (R2=0.856). Meanwhile, the model using SA VI denoted highest values (R2=0.852)for predicting LAI While the validation of predictive model using RVI implied the highest values (K=O. 797) for predicting LDW. Moreover, the test of predictive model using SAVI indicated the highest value (R2=0.658) for predicting SPAD values. According to overall validation using VIs, it seems that RVI has the best accuracy to validate the predictive model of LAI than those of LDW or SPAD values. Meanwhile, the most significant of K to validate the predictive model was obtained on FDR data with R2=0.859for LAl