Modeling Gotvand-Aghili irrigation and drainage network soil salinity by usingLandsat-8 image and OLS and Ridge regressions

Soil salinization is one of the main obstacles in the agriculture sector and one of the main factors of crop yield reduction. In this study, a Landsat-8 image and extracted data of 121 soil samples from network surface were utilized and soil salinity indices, multiple linear regression with ordinary least square and Ridge regression were applied so that Gotvand-Aghili irrigation and drainage network soil salinity was modeled and soil salinity map was finally prepared. The results indicated that there is a significant correlation between bands of Landsat-8 Operational Land Imager (OLI) sensor and electrical conductivity of soil samples and the red band had the highest correlation. Furthermore, the findings revealed a significant correlation between salinity indices and electrical conductivity of soil samples. The results of OLS regression model further disclosed that this model was able to represent 48% of the network soil salinity variation, however, according to the variance inflation factor, independent variables of OLS model enjoyed multicollinearity. Conversely, the results of Ridge regression illuminated that this approach has a high ability for removing multicollinearity of independent variables. By mapping OLS and Ridge regression models, the average soil salinity of the network assessed 5.8 and 4.5 ds/m, respectively. Moreover, according to Ridge regression model, 33% of the network area was classified as moderately saline while 15% of it would fall into a strongly saline class. Also, based on the salinity map of OLS regression model, 6% of the network area was categorized as very strongly saline class.