Estimation On Saline Alkali Soil Water And Salt Contents Based On Hyperspectral Technology In The Yellow River Delta

Coastal Saline is our precious land resources, as the major soil types in the Yellow RiverDelta, because of the low altitude, high water table and salinity, evaporation and water erosion,and soil salinization is serious. It has become the main factor of restricting the regionaleconomic development. Timely, precisely, dynamically for saline soil water and saltinformation, to govern saline soil, prevent further degradation and realize agriculturesustainable development is essential.Taking Kenli County of Dongying City in the Yellow River Delta region as the studyarea, on the basis of obtaining saline-alkali soil hyperspectral data, it studied the spectralcharacteristics variation due to the different saline-alkali soil water and salt content, andanalysised the relationship between the saline-alkali soil spectral and water and salt content.Extracted sensitive bands by the differential method, the envelope method and relevantanalysis to establish spectrum characteristic parameters, and constructed the quantitativerelationship model between the spectral data and the saline-alkali soil water and salt contentwith multiple linear regression analysis and BP neural network. It selected optimal estimationmodel through comparative analysis and accuracy verification.The main conclusions were as follows:(1)The spectral characteristics of the water and salt content were determinedIn wavelength of350nm-2500nm, saline soil spectral reflectance curve shape on thewhole of different water or salt content was consistent, and it decreased with saline soil wateror salt content increasing. The degree of decrease gradually increased along with wavelengthincreasing, especially in the infrared region.(2) The saline-alkali soil water content hyperspectral model was establishedAfter the first-order differential transformation of the reciprocal-logarithm, it chose552nm,862nm,1201nm,1430nm,2029nm,2218nm as the sensitive wavelengths,established estimate model by multiple linear regression analysis and BP neural network,through the precision verification analysis, and ultimately determined the best estimationmodel for the BP neural network of6-5-1structure, the prediction accuracy of R2=0.9586,RMSE=0.886, showing that the model with high accuracy can be used to estimate thesaline-alkali soil water content in the region studied. (3) The saline-alkali soil salt content hyperspectral model was establishedIn the analysis of the spectral characteristics of the saline-alkali soil salt content, it hadslight deviation by different data processing methods to select sensitive bands. it chose500nm-520nm,550nm-580nm,986nm-1000nm,1235nm-1245nm,1490nm-1510nm,2000nm-2030nm as the sensitive bands by means of the first-order differential of thereciprocal-logarithm transformation and chose490nm-510nm,825nm,855nm,1280nm-1300nm,1450nm-1480nm,1690nm-1720nm,1970nm-2000nm,2230nm-2250nmas the sensitive bands after envelope processing. Taking sensitive bands as the independentvariables, using multiple linear regression analysis and BP neural network to establish models,through the verification and analysis of accuracy, it ultimately determined the best estimationmodel that it was the BP neural network of the6-5-1network structure after the first-orderdifferential of the reciprocal-logarithm transformation. Through accuracy verification, themodel that the coefficient of determination was up to0.93had high accuracy for theprediction of saline-alkali soil salt content in the regions.