Common Research And Near-infrared Analysis Models Of Organic Matter For Many Types Of Soils

Soil fertility index detection is an important basic work of modern agriculture, the methods of existing detection require biochemical reagents, process complex and not easy to scale applications. Near-infrared(NIR) spectroscopy can be directly detected samples without reagent, having the advantage of quick and easy. It have confirmed study that Feasibility of Near Infrared Spectroscopy for soil fertility indicators(organic matter), but due to the soils have many types, further establish personalized model and joint model for multi-type soil, it is important to the near-infrared spectroscopy to scale application in the field of soil. We collected three zones typical soil samples of Dongbei, Nanjin and Guangdong. We build stabilized NIR model and test framework base on multiple scaling prediction set that randomly divided. Using equidistant combination partial least squares(EC-PLS) to prefer wavelength, establishing the equivalence model set:(1) further propose repetition rate priority combination partial least squares(RRPC-PLS), establish near-infrared personalized analysis model for three types of soils organic matter;(2) establish near-infrared joint analysis model for three types of soils organic matter to conduct common research.Using EC-PLS method, about the soil of Dongbei, the obtained optimal initial wavelength(I), number of wavelengths(N) and number of wavelengths gaps(G), were 1090 nm, 63 and 10, respectively; about the soil of Nanjing, the obtained optimal I, N and G were 1310 nm, 48 and 4, respectively; about the soil of Guangdong, the obtained optimal I, N and G were 1172 nm, 72 and 9, respectively. Establishing the equivalence model set base on these, further using RRPC-PLS method to establish joint analysis model for three types of soils organic matter, the number of wavelengths used were 8, 9 and 56; on the other hand, for the equivalence model set to further establish joint analysis model for three types of soils organic matter, the corresponding equidistant wavelength combination I, N and G were 1632 nm, 18 and 8, respectively.Testing the preferred model by the validation sample which was not involved in the modeling process set. For obtained optimal RRPC-PLS model, the prediction root mean square deviation(V-SEP) and prediction correlation coefficient(V-RP) of Dongbei soil organic matter were 0.188wt% and 0.903, respectively; the V-SEP and V-RP of Nanjing soil organic matter were 0.507wt% and 0.945, respectively; the V-SEP and V-RP of Guangdong soil organic matter were 0.197 wt% and 0.917, respectively. For obtained optimal joint model, the V-SEP and V-RP of Dongbei soil organic matter were 0.182wt% and 0.918, respectively; the V-SEP and V-RP of Nanjing soil organic matter were 0.573wt% and 0.926, respectively; the V-SEP and V-RP of Guangdong soil organic matter were 0.249 wt% and 0.847. The results show that the near infrared predicted values and measured values has good congruency and high correlation, and infrared predicted possesses the advantages of no reagent, quick and easy.In this paper, for different types of soils, proposed the new ways of personalized and joint model to detect organic soil organic matter. It is important for Near-infrared(NIR) spectroscopy large-scale applications in the field of soils, and also provided valuable reference for development of small and dedicated spectrometer.