Study of an on-line measurement method for the salt parameters of soda-saline soils based on the texture features of cracks

The Songnen plain in China has a typical soda-saline soil that frequently shrinks and cracks under natural conditions during water evaporation. Contrast (CON) is a typical gray level co-occurrence matrix (GLCM) texture feature, which can well represent the cracking degree of dried soil surface and is highly related with the salt content of soda-saline soils. In order to make a quantitative analysis of the effects of salt content on the cracking behavior of soda-saline soils, laboratory controlled and field verification experiments on the soda-saline soil of the Songnen plain were conducted in the present study. The laboratory experiment included 56 samples and was conducted to establish the relationships between the CON values of laboratory-generated crack patterns of soil samples and some primary salt parameters, such as the electrical conductivity (EC), Na⁺ content, Cl⁻ content, CO3²⁻ content, HCO3⁻ content and total salinity. The field verification experiment included 48 samples and was conducted to establish the relationships between the CON values of field-generated crack patterns of soil samples and the salt parameters. For both laboratory experimental samples and field samples, the CON values were computed from the binary images of the cracked patterns and were significantly linearly correlated to the EC, Na⁺ content and salinity (coefficient of determination [R²] >0.87); the relationship between the CON values of soil samples and the Cl⁻ content was also quite linear (R² >0.73). An online measurement method for rapid and accurate determination of salt contents of soda-saline soils was then proposed based on the results of the laboratory experiment. After that, the salt parameters of 48 field samples were estimated using the online method and fitted against the actual measured values. The fitting results relating the CON values and Na⁺ content, Cl⁻ content, EC and salinity showed high prediction accuracy (R² >0.91, RPD >4.49); the measurement method for CO3²⁻ content also had reasonable prediction accuracy (R²=0.86 and RPD=2.42) but with relatively poor stability and poor reliability. However, the measurement method for HCO3⁻ content showed very poor prediction accuracy (R²=0.51 and RPD=1.31).