Adopting any sampling scheme is subject to your objective and required accuracy. You can select reference points from Google Maps if you know the study area with random sampling. These reference points can then be used to train your supervised classifier for different LULC types. You don't need sampling for training in the case of a non-supervised classifier.

Hi Hossein,

I think these discussions might help you : https://www.researchgate.net/post/Accuracy_classification_assessment_using_ground_truth_points-any_thoughts or https://www.researchgate.net/post/Are_there_methods_of_validation_for_classification_of_satellite_images_change_detection_LST_maps_in_the_absence_of_landuse_and_national_topomaps

Regards,

RM

I suggest that you define the classes of land use and cover through Google Earth, or, if you already know the region of study well, define the classes. With the pre-defined classes, make a visit to the location (if possible) and collect points with a GPS, this can be done even with mobile applications. Collect different points in different areas representative of each class of land use and cover. Then, in the training step of the classifier algorithm in a GIS environment, plot the points collected in the field on the image to be classified, this will make you sure about what is each type of coverage contained in the image. These collected points can also be used to test the accuracy of the supervised classification in the corrections and post-classification stage.

I will highly recommend studying these articles which will answer your concern.

Article Land Use Change Detection Using Remote Sensing Technology

or,

Article Assessing the Impacts of Land Use and Land Cover Change on l...

If you mean how to collect field data for land use and cover then these readings will be appropriate:

http://www.fao.org/3/t0715e/t0715e0b.htm

or,

https://silo.tips/download/guide-to-gps-field-data-collection