Deriving the thematic accuracy of models is a fundamental part of image classification analyses. K-fold cross-validation (KCV), as an accuracy assessment technique, can be biased because existing built-in algorithms of software solutions do not handle the high autocorrelation of remotely sensed images, leading to overestimation of accuracies. We aimed to quantify the magnitude of the overestimation of KCV-based accuracies and propose a method to overcome this problem with the example of rooftops using a WorldView-2 (WV2) satellite image, and two orthophotos. Random split to training/testing subsets, independent testing and different types of repeated KCV sampling strategies were used to generate input datasets for classification. Results revealed that applying the random splitting of reference data to training/testing subsets and KCV methods had significantly biased the accuracies by up to 17%; overall accuracies (OAs) can incorrectly reach >99%. We found that repeated KCV can provide similar results to independent testing when spatial sampling is applied with a sufficiently large distance threshold (in our case 10 m). Coarser resolution of WV2 ensured more reliable results (up to a 5-9% increase in OA) than orthophotos. Object-based pixel purity of buildings showed that when using a majority filter for at least of 50% of objects the final accuracy approached 100% with each sampling method. The final conclusion is that KCV-based modelling ensures better accuracy than single models (with better pixel purity on the object level), but the accuracy metrics without spatially filtered sampling are not reliable.

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

Gullies reduce both the quality and quantity of productive land, posing a serious threat to sustainable agriculture, hence, food security. Machine Learning (ML) algorithms are essential tools in the identification of gullies and can assist in strategic decision-making relevant to soil conservation. Nevertheless, accurate identification of gullies is a function of the selected ML algorithms, the image and number of classes used, i.e., binary (two classes) and multiclass. We applied Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Random Forest (RF) on a Systeme Pour l’Observation de la Terre (SPOT-7) image to extract gullies and investigated whether the multiclass (m) approach can offer better classification accuracy than the binary (b) approach. Using repeated k-fold cross-validation, we generated 36 models. Our findings revealed that, of these models, both RFb (98.70%) and SVMm (98.01%) outperformed the LDA in terms of overall accuracy (OA). However, the LDAb (99.51%) recorded the highest producer’s accuracy (PA) but had low corresponding user’s accuracy (UA) with 18.5%. The binary approach was generally better than the multiclass approach; however, on class level, the multiclass approach outperformed the binary approach in gully identification. Despite low spectral resolution, the pan-sharpened SPOT-7 product successfully identified gullies. The proposed methodology is relatively simple, but practically sound, and can be used to monitor gullies within and beyond the study region.

Building extraction from remote sensing data is an important topic in urban studies and the deep learning methods have an increasing role due to their minimal requirements in training data to reach outstanding performance. We aimed to investigate the original U-Net architecture’s efficiency in building segmentation with different number of training images and the role of data augmentation based on multisource remote sensing data with varying spatial and spectral resolutions (WorldView-2 [WV2], WorldView-3 [WV3] images and an aerial orthophoto [ORTHO]). When the trainings and predictions were conducted on the same image, U-Net provided good results with very few training images (validation accuracies: 94-97%; 192 images). Combining the ORTHO’s and WV2’s training data for prediction on WV3 provided poor results with low F1-score (0.184). However, the inclusion of only 48 WV3 training images significantly improved the F1-score (0.693), thus, most buildings were correctly identified. Accordingly, using only independent reference data (other than the target image) is not enough to train an accurate model. In our case, the reference from WW2 and ORTHO images did not provide an acceptable basis to train a good model, but a minimal number of training images from the targeted WV3 improved the accuracy (F1-score: 69%).