You can try the unsupervised classification and then pixel extraction in arcgis or erdas software, and if you have already got the information from ground checking about soil type and other land uses, you can use the supervised classification method for better result.
Dear sir ... there are many ways to classify the image by using unsupervised which is used only to recognized the classes in the area of interest , and supervised image which is the best way to classify any image with different methods. you can use ERDAS imagine or ENVI software and you will find many methods there. and there area technique to classified the image by using Indices and you will get best result.
1. There is a tool under spatial analyst that you can use. however, review the basics of unsupervised and supervised classification methods, and see which is applicable in your study based on the limitations you have. ArcGIS can do the basic classification but other specialist aoftwares like ENVI, ERDAS imagine and IDRISI are options. You can also use qgis if you want. Whats important if for you to learn the basics first.
I am interested to know information about different band ratios which can help in delineating and extracting different features, lithology, dry/wet soil, NDVI from Landsat Images using Arc GIS.
As I know, it is possible, and easy, to extract geomophological data from a Landsat Image, as the other comments said. But lithology is so much harder. You would need data about geochemistry, structure and geophysics that are not usually included into the images. The other parameters (dry soil and NDVI) I do not know how it can be extracted, but it could be easy too.
1. The purpose is to develop a map of land use / land cover, lithology map, geomorphological map.
The best thing would be to have a field radiometer and obtain with this device a spectral signature of each type of coverage. Or have a wide knowledge of the area, to select the ideal areas of extremadura for each class.
2. Study the pattern of land use and changes in the coastline with time.
The morphology, it is difficult without a contour map, at least detailed, but if you can do with these products, the best would be the aerial photogallery.
*Remember that the accuracy depends on the geometry of the pixels of the satellite image and its size.
I think that you must make a supervised classification.
You can make your own spectral signatures from the Landsat images, and then make the classification of the all images.
But I think that you must considers the temporal difference between the images that you are going to use. If this is the case, probably you must repeat this procedure to each image. Other alternative is to make your spectral signatures with a radiometer or ask o someone that could have it, but taking into account this temporal difference.
I think that the supervised classification could give you all the information that you need