Prediction of soil salinity using multivariable regression on the basis of extracted indices from Landsat 8 satellite (Case study: Urmia)

Managing and monitoring of salinity is one of the most important affair in agriculture, especially in arid and semi-arid area. For this purpose we have to use new technology like remote sensing and GIS. The relationship between soil parameters with satellite data is an effective step for predicting soil salinity. In this study we use multivariable regression based on relationship between topographical properties and extracted indices from Landsat 8 satellite for predicting soil salinity in Urmia. For predicting soil salinity, samples of 40 points from 0-30 cm soil depth were taken. Electrical conductivity from soil saturation extract (ECe) was measured. After performing the necessary processing on satellite images, pixel values in the different bands were extracted. The data was divided into two series: Training data (80%), validation data (20%). The relationship between satellite data and results of multivariable linear regression methods predicted, accuracy of the model by using factors such as R- squared, standard error of the mean, adjusted R-squared and Durbin Watson statistic evaluated. Results showed that model predicted with correlation coefficient, standard error of the mean, adjusted R-squared and Durbin Watson statistic were 70.3, 10.03, 61.8 and 1.709 respectively. Finally, the model evaluated by statistical indices. The indices values of Geometric Mean Error Ratio (GMER), R- squared (R2) and Root Mean Square Error (RMSE) measured 0.867, 0.638 and 0.354 respectively. The results showed that the model has a better estimation of soil salinity.