Effect Of Freeze Thawing On Accuracy Of Determination Soil Phosphorus And Potassium Content Using Near Infrared Spectroscopy

Because of the low accuracy and large prediction deviation using NIRS to measure soilP and K content in present study, the soil samples were carried on ultra-lowtemperature freeze thawing treatment using liquid nitrogen in this study. The effect ofultra-low temperature freeze thawing treatment on soil nutrients, near-infrared and infraredspectroscopy were analyzed. The wavelength and spectroscopy related to soil nutrients wereestablished and the effect of NIR quantitative calibration model and predication accuracywere analyzed using the soil which carried on ultra-low temperature freeze thawingtreatment, the conclusions are as follows:The soil nutrient and NIR of235soil samples treated ultra-low temperature freezethawing treatment using liquid nitrogen were measured in this study, the result indicated thattotal P and total K contents of treated soil did not change, while available P and K contents oftreated soil increased significantly, effective P content did not change significantly and theNIR absorbance also changed too. The long-term positioning brown soil and greenhouse soilspectrum were also measured, it can be identified the sensitive bands affected by P and Kwere1000nm～1890nm,1980nm～2200nm, and2200nm～2380nm. The effected spectrumof P, K contents to NIR absorption bands were determined through studying the correlationcoefficient between soil contents with NIR and quantitative calibration model were alsoestablished to identify the spectrum band. The result showed that the model established usingthe range of these bands was much better than that using full wavelength range.Compared to the calibration model, the mean centralization and first derivative were thebest pretreatment methods. The calibration models were established by PLSR, ANN and PCRtechniques to relate NIR spectral data to the total P content. In PLSR model, the coefficient ofdetermination and prediction accuracy are improved after treatment,(before freeze thawingtreatment, Rc~2=0.949, SEC=0.224, Rp~2=0.942, SEP=0.299, RPD=3.064; after freeze thawingtreatment, Rc~2=0.953, SEC=0.214, Rp~2=0.957, SEP=0.273, RPD=3.355). The calibrationmodel established by ANN had low prediction accuracy for unhanded soil, but had higherprediction accuracy for treated soil (before treatment, Rc~2=0.967, SEC=0.341, Rp~2=0.952,SEP=0.470, RPD=1.949; after treatment: Rc~2=0.979, SEC=0.153, Rp~2=0.985, SEP=0.168, RPD=5.452). The calibration model established by PCR had low prediction accuracy, so itcan not get the non-destructive quantitative requirements.The calibration models were established by PLSR, ANN and PCR techniques to relateNIR spectral data to content of unhandled soil available P, the result showed that all themodels can not reach non-destructive quantitative requirements. However, after treatment, themodels were better.(PLSR: RC~2=0.893, SEC=34.007, RP~2=0.918, SEP=30.845, RPD=3.167;ANN: RC~2=0.967, SEC=30.970, RP~2=0.952, SEP=31.325, RPD=3.539; PCR: RC~2=0.862,SEC=36.588, RP~2=0.892, SEP=43.276, RPD=2.562). The model established by PLSR andANN can non-destructively detect available P content rapidly.The effect of calibration models established using PLSR, ANN, and PCR for unhandledand treated soils were not very good and had great prediction deviation. These calibrationmodels can not reach our requirements that detect soil available K content non-destructively.The calibration models were established by PLS, ANN and PCR techniques to relateNIR spectral data to content of unhandled soil available K. In the three calibration models,PLSR was the best to detect available K content. In PLSR model, the coefficient ofdetermination and prediction accuracy were improved after treatment,(before freeze thawingtreatment, Rc~2=0.928, SEC=52.038, Rp~2=0.922, SEP=58.701, RPD=2.911; after freezethawing treatment, Rc~2=0.942, SEC=47.422, Rp~2=0.938, SEP=42.340, RPD=4.035). Theprediction accuracy was low for unhandled soil using ANN calibration model, but it washigher for treated soil.(before freeze thawing treatment, Rc~2=0.878, SEC=71.239, Rp~2=0.867,SEP=93.338, RPD=1.831; after freeze thawing treatment, Rc~2=0.896, SEC=30.556,Rp~2=0.893, SEP=53.814, RPD=3.175). It can be concluded that models using ANN andPLSR can detect available K content non-destructively. Model established using PCR hadlow prediction accuracy and could not reach non-destructive quantitative requirements.In PLSR, ANN, and PCR models for measuring effective K content of unhandled soil,The results were not very good, but the model established by ANN for treated soil was better,RC~2=0.912, SEC=62.047, RP~2=0.917, SEP=42.810, RPD=5.192, so it can be used to detecteffective K content non-destructively.