Distribution map of the different lithologic units in loess plateau of eastern Golestan by using remote sensing technique; Aghband research area

Introduction: Along with the climate, Soil is an essential natural resource. Although soil studies in Iran have been started more than 50 years ago, the soil map of the country has not been fully prepared yet, and only 20-25% of the lands have been mapped already. Many soil maps of Iran need to be updated, but the common methods in soil mapping are costly and time-consuming. Hence, using data obtained from remote sensing is trending nowadays. These data possess a great importance as a proper approach for preparation of various maps due to timely presentation of information, variety in type, being digitized and possibility of digital processing. Classification is one of the useful methods of obtaining data from remote sensing images which allows the user to generate various types of information such as different maps. Different regions of Golestan province are covered with loessial sediments. Recent studies in loess plateau of Golestan province show that these loesses possess useful information about climate change and landscape evolution in quaternary period. The aims of the present study was mapping different lithologic units in loess plateau of eastern Golestan using remote sensing technique and evaluating the efficiency of using remote sensing images to identify and prepare the distribution map of different lithology units in the study area. Material and methods: Loess plateau of Golestan province which is located at the north of Alborz Mountains and western region of Qopeh-dagh Mountains possesses an area equal to 250000 hectares; the height of the mentioned loess plateau is in the range of 45 m to 700 m from sea level. In this study, about 40 km2 of western part of Golestan loess plateau is selected as the study area located near Aghband village, 60 km from Gonbad city. In the following study, distribution map of different lithology units using the data from Landsat 8 and Sentinel 2 satellites was performed via classification of support vector machine, maximum likelihood and photomorphic image analysis methods in GIS software interface by Sentinel-2A (RGB band  No. 2,3, 4) and Landsat8-OLI (band  No. 2,3, 4). After preparing distribution map of different lithology units, their accuracy was evaluated via confusion matrix, The produced maps were evaluated by 607 ground truth points obtained by field survey and Google Earth and their overall accuracies and kappa coefficients were calculated.   Result and discussion: Field observations revealed four types of lithology units including loess, loess-paleosol of early Pleistocene, marl and limestone. Each of these lithology types had specific colors and could be identified by color difference in true color composite bands of satellite data and Google Earth images. After the lithology units were digitally classified via the maximum likelihood and support vector machine methods, their distribution maps were prepared and evaluated. Classification of Sentinel 2 images via support vector machine and maximum likelihood methods led to a accuracy of 89.45% and 87.7%, respectively, whereas these values for Landsat were 85.17% and 81.87%.   Conclusion: Each lithology of Golestan loess plateau had a specific color of its own; color of loesses are yellow and yellowish, marls have different colors including blue and green and colors between them, color of lime-stone are white and milky and color of loess-paleosols are red.The results associated with the comparison of overall accuracy and Kappa coefficient of the classified images indicate that although both algorithms of support vector machine and maximum likelihood which are utilized for digital classification had high overall accuracy and Kappa coefficient, superiority of support vector machine compared to maximum likelihood method is obvious in both images utilized in the present study. However, on the other hand, a higher accuracy may be observed in the images taken by Sentinel 2 due to smaller pixel size. Thus, it may be concluded that support vector machine method not only has performed more capably in detection of different lithologies, but also changing the source of the satellite images has no impact on the efficiency of this method. Noting that one of the objectives of the study was evaluating the efficiency of using remote sensing technique to identify and prepare the distribution map of different lithology units in the study area, it was concluded that utilizing remote sensing techniques in the studied region may be feasible along with the other mapping methods.