2.4 Field data collection for parameterization of the
classifier and accuracy assessment
Field data, which was later utilized for training and accuracy
assessment, was collected following several steps. Field campaigns were
conducted and information about land-cover types was recorded with a
non-differential GPS during May and August 2016. Notes were also taken
on land-cover and land-use types. For a better interpretation of the
results, photos were taken in the field as well. We also interviewed the
farmers about their activities to reconstruct land use by recalling
memories about land use, back to 1985. Field-collected data were then
used to parameterize and correct the random forest classifier, and
assess the classification accuracy of produced maps. We showed satellite
imagery available via Google Earth mapping services during the
interviews with farmers if we were not sure about the type of LCLU.