Abstract:。 The tide salt clay in Zhejiang Province was selected as research object, and the soil N and soil P were analyzed with NIR spectroscopic techniques. Six group samples were collected from a rice farm. Several kinds of nutritional water with different concentrations were added to the six groups, and then the samples were dried and rubbed. At last, 120 samples were got from six groups equably. Standard normal variate (SNV), multivariate scatter correction (MSC) and smoothing of moving average were used to process the spectral data. Different calibration models were established and the performance of these models was compared with different pretreatment methods. After comparison, smoothing of moving average was found to be the best spectral pretreatment method. 96 samples were randomly selected from 120 samples as the calibration set, and the other 24 samples were used as the validation samples. Two discriminating analysis models were developed using partial least squares (PLS) method and least squares-support vector machine (LS-SVM) method respectively. The correlation coefficients (r) between the measured data and the predicted data from PLS were 0.9454(N), 0.9327(P) respectively, and 0.0321(N),0.9547(P) from LS-SVM, respectively. The root mean standard error of prediction (RMSEP) were 0.0321(N),0.0089(P) from PLS, and 0.0378(N),0.0101(P) from LS-SVM. The results showed that NIRS could be used to evaluate the soil N and soil P.
