Classification Efficacy Using K-Fold Cross-Validation and Bootstrapping Resampling Techniques on the Example of Mapping Complex Gully Systems

The availability of aerial and satellite imageries has greatly reduced the costs time associated with gully mapping, especially in remote locations. Regardless, accurate identification gullies from images remains an open issue despite amount literature addressing this problem. main objective work was to investigate performance support vector machines (SVM) random forest (RF) algorithms extracting based on two resampling methods: bootstrapping k-fold cross-validation (CV). In order achieve objective, we used PlanetScope data, acquired during wet dry seasons. Using Normalized Difference Vegetation Index (NDVI) multispectral bands, also explored potential image discriminating surrounding land cover. Results revealed that had significantly different (p < 0.001) spectral profiles any other cover class regarding all bands image, both However, NDVI not efficient discrimination. Based overall accuracies, RF’s better CV, particularly season, where its up 4% than SVM’s. Nevertheless, level metrics (omission error: 11.8%; commission 19%) showed SVM combined CV more successful extraction season. On contrary, RF relatively low omission (16.4%) errors (10.4%), making it most algorithm estimated area 88 ± 14.4 ha season 57.2 18.8 standard error (8.2 ha), appropriate which a slightly higher (8.6 ha). For first time, study sheds light influence these techniques accuracy satellite-based mapping. More importantly, provides basis for further investigations into such techniques, when using data.