Yes, this is pretty broad, but let's see if this doesn't help.

You may be interested to know that the NASS (http://www.nass.usda.gov/) has land cover and cropland data. Now I understand if you want to do it with your own data over some extent or for greater control. Actually, this is easier than it might seem, especially if using ArcGIS.

You say that you have already determined control areas, or training samples--this is good. You can load in the raster datasets that you are planning on using, especially satellite imagery or satellite infrared bands. You can create a supervised classification using the image classification toolbar (here is a link to the process description: http://resources.arcgis.com/en/help/main/10.1/index.html#/Creating_training_samples/00nv0000000n000000/). After delineating the control points, run the classification tool, which should output a raster (see how here: http://resources.arcgis.com/en/help/main/10.1/index.html#/Executing_the_Interactive_Supervised_Classification_tool/00nv0000000t000000/). Note that they say to use the "Interactive Supervised Classification" selection; you can also use the "Maximum Likelihood Classification" as well.

Please see check the link below. Hope it helps.

http://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-analyst-toolbox/understanding-segmentation-and-classification.htm

https://www.wrc.umn.edu/sites/wrc.umn.edu/files/classification_of_the_wetlands.pdf

https://www.youtube.com/watch?v=TpuV7DT6seI

Thank you for your answers, Jackson and Mustafa!

The main software I will be using is eCognition (I should have mentioned this too). Nevertheless I need ground truth data for accuracy assessment and I don't have such data for all my classes. What would be the way to compensate this problem, apart from additional field surveys?

Hi Patrick, ...yes, broad subject ! 

In short, nothing replaces proper botanical inventory,  identify the (main) proper vegetation types (and ecological group) and field survey (especially soil) and a broad ecological knowledge of similar environment. You may have  the best supervised or unsupervised classif .... and simply get irrelevant "rubbish" on your screen. Nothing replaces preliminary detailed field work. 

As example, you may have a look at:

1) A crude approach to vegetation survey using aerial photos supported by field surveys (before Sat imagery and sophisticated software were available:

 https://www.researchgate.net/publication/304283957_DEST5925_Map_Bir-El-Ghanem_2                   - Vegetation map Bir El Ghanem - Libya 

and 

https://www.researchgate.net/publication/37885658_Range_study_on_Jeffara_plain_Libya_I_Sheep_project_-_Bir_El_Ghanem      Technical report 

2) Using Remote sensed data and field work :

2.1 ) Rangelands of the Ravnina region in the Karakum desert of Turkmenistan: Current condition and utilisation, Technical Report · August 2009; in DESERTIFICATION AND REGENERATION: MODELING THE IMPACT OF MARKET REFORM ON CENTRAL ASIAN RANGELAND, Affiliation: DARCA (INCO-COPERNICUS / RTD Project: ICA2-CT-2000-10015)

at https://www.researchgate.net/publication/302562916_Rangelands_of_the_Ravnina_region_in_the_Karakum_desert_of_Turkmenistan_Current_condition_and_utilisation 

2.2) vegetation survey linked with desertification 

 https://www.researchgate.net/publication/303057047_A_spatial_desertification_indicator_for_Mediterranean_arid_rangelands_a_case_study_in_Algeria

3) and you may be interested to read also : 

https://www.researchgate.net/publication/303997109_Diagram_of_Rangeland_inventory_and_monitoring_referring_to_different_scientific_schools_and_communities 

and 

https://www.researchgate.net/publication/303996439_From_Inventory_To_Monitoring_In_Semi-Arid_And_Arid_Rangelands

I hope this may help you. 

Regards 

Gus 

Data DEST5925 Map Bir-El-Ghanem 2

Technical Report Range study on Jeffara plain (Libya). I. Sheep project - Bir El Ghanem

Technical Report Rangelands of the Ravnina region in the Karakum desert of Tu...

Article A spatial desertification indicator for Mediterranean arid r...

Data Diagram of Rangeland inventory and monitoring referring to d...

Chapter From Inventory To Monitoring In Semi-Arid And Arid Rangelands

Hi Gus,

thank you very much for your answer and extensive literature introduction!

Best whishes, Patrick

Hi Patrick,

additionally to your set of variables (orthofotos, CIR, etc.) you could consider to use ariborne laser scanning data too, which is available in Austria.

Maybe this article is helpful:

https://www.researchgate.net/publication/307522357_BIODIVERSITY_MAPPING_VIA_NATURA_2000_CONSERVATION_STATUS_AND_EBV_ASSESSMENT_USING_AIRBORNE_LASER_SCANNING_IN_ALKALI_GRASSLANDS

Best wishes,

Philipp

Article BIODIVERSITY MAPPING VIA NATURA 2000 CONSERVATION STATUS AND...

Hi Patrick,

as an initial introduction, I think "Remote Sensing and GIS for Ecologists using Open Source Software" is a good textbook with also a nice online guide. Please find it here. http://book.ecosens.org/

Note that eCognition also works with LIDAR data or LIDAR-based raster data products, this might be a good way of maximizing the information content of your input data. However, I also warn that unless your classes are very basic and can be recognized by aerial photo interpretation, it will be essential that you collect field data for all the categories you use. Field data has to be well distributed in space across your site, and balanced across categories (approximately the same amount of reference pixels for each class). Maybe if it is a protected area, the already existing local management plan or some biodiversity monitoring or Natura 2000 monitoring, or an existing vegetation map can also be a source of reference data.

If you revert to using manual image interpretation to generate reference data from the same images that you want to classify, you lose the independence of your training and validation data. Any issues specific to your image dataset (camera problems, timing with respect to phenology, atmospheric issues) would also influence your training and validation data then. Reviewers often reject such studies and ask for independent training and validation data, whether from interpretation of remote sensing or from the field.

http://book.ecosens.org/

Hi Philipp, hi András, thank you for your answers!