Incorporating legacy soil data to minimize errors in salinity change detection: a case study of Darab Plain, Iran

The results of a 1990 soil survey of a salinized region in Darab Plain, southern Iran, were combined with soil sampling data taken in 2002 from the same locations and employed as a basis for salinity change detection in the region. New preprocessing of satellite imagery was used, along with statistical analysis of the digital number (DN)-salinity relationship, in order to determine salinization of the area. Removal of outliers on the basis of interfering land uses improved the correlations. Nonlinear regression (NLR) in the form y = a + bx(alpha) provided a suitable predictor of salinity (y, dS m(-1)) for both 1990 and 2002 based on DNs (x). Among the 12 tested methods of salinity classification in this study, the six salinity class method with intervals 0-4, 4-10, 10-32, 32-64, 64-80 and > 80 dS m(-1) was selected. A series of accuracy assessments through a trial-and-error procedure was the basis of the selection of the best method and led to a final accuracy of 91%. About 42% of the lands located on 'no saline' and 'low salinity' classes in 1990 had changed to the 'medium', 'very high' and 'new agricultural land' classes in 2002.