Document Type : Research Paper

Authors

• Mahsa Jahanbakhsh ¹
• Ali Esmaeily ²

¹ M.Sc. in Remote Sensing Engineering, Faculty of Civil and Surveying Engineering, Graduate University of Advanced Technology

² Assistant Professor, Faculty of Civil and Surveying Engineering, Graduate University of Advanced Technology

Abstract

Extended Abstract
Introduction
    In recent years, we have seen the importance and high demand of lithium (Li) due to its many applications, for example in the production of rechargeable lithium batteries, which are mainly related to the global markets of electric vehicle manufacturing to achieve a healthy environment and more suitable transportation. Due to this high demand, the identification of new lithium reserves is very important and the investigation of its identification and zoning methods has been the focus of many researchers, and the use of remote sensing data and image processing techniques in the detection of lithium due to cost reduction of earth exploration has increased, greatly.In this research, using modern methods, a general and intelligent approach was presented, so that with the least time and cost, after selecting the bands of the desired satellite images and zoning the area of Degh Ptergan, in Zirkoh city, South Khorasan province, as a possible area for the existence of lithium reserves, modeling was done by the supervised machine learning method, and the relative importance of the variables was determined using the trained model.Also, the relative importance of the variables was determined by the trained model, and the ability of each of the remote sensing techniques to achieve this goal has been challenged.
Materials&  Methods
    Here, 13 bands of Sentinel-2 images and the region of 12 known lithium mines around the world were used as lithium presence areas, so that, by going through steps, suitable data for modeling were produced. In this way, by using the boundaries of these mines, samples were produced that can be used as input for modeling algorithms. The maximum entropy algorithm was used to model the distribution of lithium samples. Since the correlation between the input variables reduces the performance of the model and makes it difficult to interpret the results of the modeling, first, the correlation between the input variables was calculated and those with a high correlation were discarded. So that, 16 variables were used as input in the maximum entropy algorithm and finally a suitable model was obtained with the AUC (Area Under the Curve) criterion of 0.706 and by it, the study area of Degh Patregan, located in the province South Khorasan, Iran was zoned and two possible areas containing lithium resources were identified.To determine the relative importance and contribution of the input variables in the prediction map of lithium minerals, the Jacknife method was implemented. According to this method, the variables B10, B06/B08, B06/B07 and B01/B10 have a high relative importance, which shows that they have more information than the other variables. Then classic remote sensing techniques including color composition, band ratio, principal component analysis and SAM was done to zone the study area, too. The results of maximum entropy modeling were compared with these techniques and the high ability of the maximum entropy algorithm was determined.
Results & Discussion
   According to the results and prediction maps related to the classical methods, it showed that although some of these methods approximately identified the areas specified by the maximum entropy algorithm, but they had problems that is emphasized on the development of more suitable remote sensing algorithms to describe the changes associated with lithium minerals. The maximum entropy algorithm with its unique options is a powerful tool for extracting the features of satellite images and expresses their hidden information more clearly. The accuracy of this method was compared with classical techniques and it was able to provide a more appropriate classification with a low noise and with a Kappa coefficient of 0.8775 and an overall accuracy of 0.9435, and identified two areas with the possibility of the presence of lithium minerals in the study area.
Conclusion & Suggestions
In the present research, the study area of Degh Patergan, located in South Khorasan province, Iran, was zoned, whereby two possible areas containing lithium resources were identified and the ability of classical remote sensing methods and maximum entropy algorithm was challenged. The method discussed in the research may be used as a cost-effective and technological solution with priority over field mapping for mineral exploration in remote border areas with difficult access, also an automatic approach with the maximum entropy algorithm was presented for the exploration of different mineral resources, which can be used for other exploration as well. Therefore, it is suggested to be used in different areas and to explore different sources.

Keywords

• Lithium
• Maximum entropy
• Remote sensing
• SAM
• Sentinel-2

Main Subjects

• Remote Sensing (RS)