An Improved Method for the Evaluation and Local Multi-Scale Optimization of the Automatic Extraction of Slope Units in Complex Terrains

Abstract

Slope units (SUs) are sub-watersheds bounded by ridge and valley lines. A slope unit reflects the physical relationship between landslides and geomorphological features and is especially useful for landslide sensitivity modeling. There have been significant algorithmic advances in the automatic delineation of SUs. But the intrinsic difficulties of determining input parameters and correcting for unreasonable SUs have hindered their wide application. An improved method of the evaluation and local multi-scale optimization for the automatic extraction of SUs is proposed. The Sus’ groups more consistent with the topographic features were achieved through a stepwise approach from a global optimum to a local refining. First, the preliminary subdivisions of multiple SUs were obtained based on the r.slopeunit software. The optimal subdivision scale was obtained by a collaborative evaluation approach capable of simultaneously measuring objective minimum discrepancies and seeking a global optimum. Second, under the selected optimal scale, unreasonable SUs such as over-subdivided slope units (OSSUs) and under-subdivided slope units (USSUs) were further distinguished. The local average similarity (LS) metric for each SU was designed based on calculating the SU’s area, common boundary and neighborhood variability. The inflection points of the cumulative frequency curve of LS were calculated as the distinguishing intervals for those unrealistic SUs by maximum interclass variance threshold. Third, a new effective optimization mechanism containing the re-subdivision of USSUs and merging of OSSUs was put into effect. We thus obtained SUs composed of terrain subdivisions with multiple scales, which is currently one of the few available methods for non-single scales. The statistical distributions of density, size and shapes demonstrate the excellent performance of the refined SUs in capturing the variability of complex terrains. Benefiting from the sufficient integrating approach of diverse features for each object, it is a significant advantage that the processing object can be transferred from general entirety to each precise individual.

Automatic Extraction for Land Parcels Based on Multi-Scale Segmentation

“…We believe that the threshold range of internal heterogeneity of land parcels is determined by different spatial research orientations. Thus, land parcels extracted at different scale thresholds can be considered continuous, uniform, and closed spatial units [24].…”

“…Generally, in geological disaster identification research, a smaller spatial scale is required [4,19]; for soil erosion and slope units, a medium spatial scale is needed [20]; and for regional and marine units, a larger spatial scale is required [21]. Furthermore, the extracted land parcels, regardless of their spatial scale, can be defined as a representation of regional hydro-geomorphological conditions [22], with internal characteristics like similar slopes, aspect, surface undulation, and drainage intensity [23,24]. This study focuses on Xinghai County in the Qinghai-Tibet Plateau, Qinghai Province.…”