Pixel-based and object-based image analysis approaches for classifying broad land cover classes over agricultural landscapes are compared using three supervised machine learning algorithms: decision tree (DT), random forest (RF), and the support vector machine (SVM).
We used the following broad land cover classes: 1) dense and secondary forest or old fallow fields with trees; 2) areas with shrubs and grasses such as in recently fallowed fields; 3) agricultural land; and 4) built-up and barren land.
Hence, the target habitats that showed stability (persistence) between the two dates were excluded from the analysis, as well as the land cover classes not suitable for restoration (broad-leaved forests, arable land, artificial and other agricultural areas).
Land cover classes from VGT-TEA were aggregated to broad, general class types, and then compared and validated with classifications derived from fine-resolution (Landsat) data.
Table 3 Forest cover classes S/No.
The NOAA land cover classes are changed into SLAMM land cover classes for modeling purposes.
The resulting raster land cover map of our study area represented 15 total land cover classes.
On average, such a land cover mosaic was made up of three different land cover classes.
Pattern metrics were correlated with proportion of land in a land cover class, especially for proportion agricultural and proportion urban land, which exhibited the broadest land cover gradients in the study area.
Finally, SOC stocks were estimated for each land cover class.
