Gully erosion mapping with high resolution imagery and ALS data by using tree decision, hierarchical classification and OBIA

Abstract

The gully erosion presents spectral and spatial heterogeneity and altimetry variation. It is not a land use class, but an object and it can be mapped as a subclass, using OBIA. This study presents a methodology for delimitation of gullies in rural environments, based on image classification procedures. For such, two study areas were selected: one located in Minas Gerais, Brazil and another one located in Queensland, Australia. There were used high resolution images and ALS data. The objects were generated by multiresolution segmentation method. The most important attributes in the definition of gullies were selected using decision tree induction algorithms, being these attributes spectral, altimetry and texture. Classifications hierarchical and by decision trees were carried out. Using decision tree the classification is performed only by a factor of scale, not allowing the identification of all the constituent features of the gully system. In hierarchical classification, the procedure is performed at different scales and allowing to use of fuzzy logic. The classification obtained with hierarchical classification showed results more reliable with the field of reality, by allowing the use of different scales, fuzzy logic and integration of knowledge (the established rule base) compared to the automatic classification by decision tree. As different gullies erosion are similar when presents the same evolution stage and soil type, it is not possible to select attributes to classify all gully systems, being necessary to investigate attributes for each gully erosion, based on available data and existing land use classes in the area.

Gully Head-Cuts Inventory and Semi-Automatic Gully Extraction Using LiDAR and Topographic Openness—Case Study: Covurlui Plateau, Eastern Romania

“…Furthermore, geomorphometric derivates, including terrain roughness, terrain curvature [13], normalized slope, normalized elevation [14], and bidirectional shading [15], have been utilized in methodologies to identify gullies and extract gully shoulders. Machine learning techniques have also been utilized in various methodologies to identify and extract gullies, including Support Vector Machine [16,17], Neural Network [17,18], Maximum likelihood classification, Minimum Distance [16], Tree Decision Hierarchical Classification [19], and Random Forest [17]. In recent years, the emergence of techniques for creating high-resolution digital terrain models has led researchers to employ Lidar-derived digital elevation models for gully identification and extraction [20][21][22].…”