Digital Mapping of Different forms of Soil Iron in the Eastern Shore of Urmia Lake by using Landsat-8 OLI Imagery

In this study, digital mapping of the most important forms of soil Iron were done using two data mining techniques namely Decision Tree (DT) and Cubist (Cu) models. The study area includes 500 km2 of lands from two different sites located in the eastern shore of dried bed of Urmia Lake, northwest of Iran. 131 surface soil samples were taken from depth of 0-10 cm and three different forms of Iron including i): total iron (Fet); ii) pedogenic iron (Fed); and iii) amorphous iron (Feo) were measured. A total of 19 environmental covariates (auxiliary variables) derived from the Landsat-8 OLI imagery related to July 2017 were used in this study. It was found that Cu model has a higher precision than that of the DT model for predicting all three forms of  soil iron with the values R2=0.89 and RMSE= 2.25 g/kg , R2=0.85 and RMSE=0.57 g/kg and R2=0.88 and RMSE=0.09 g/kg for predicting Fet, Fed and Feo, respectively. In addition, the results of the importance and percentage of contribution of environmental covariates in both models indicated the high importance of some spectral indices such as Normalized Difference Moisture Index (NDMI) and Modified Soil Adjusted Vegetation Index (MSAVI) in the prediction of Fet, Fed and Feo. Generally, the Cu model has a higher ability and performance in modeling and predicting the spatial distribution of different forms of soil iron in the study area compared to the DT model.