Landslide recognition using SVM, Random Forest, and Maximum Likelihood classifiers on high-resolution satellite images: A case study of Itaóca, southeastern Brazil

Abstract

Landslide identification is important for understanding their conditioning factors, and for constructing susceptibility, risk, and vulnerability maps. In remote sensing this can be accomplished manually or through classifiers. This study compares three image classifiers (Maximum Likelihood, Random Forest, and Support Vector Machines (SVM)) used in identifying landslides in Itaóca (São Paulo, Brazil). Two datasets were used: a RapidEye-5 (5 m) image and a Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) (12.5 m). Seven pixel-based classifications were produced, two for each classifier and a binary class that identified only landslides and non-landslides. One classification contained five spectral bands (5B), while the other contained six bands (6B) and included the slope derived from the DEM. The results were validated using Kappa index and F1 score. The SVM 6B classification achieved the best results among the validation indices used herein. It identified a landslide area of 399,325 m². The results contribute to landslide mapping in tropical environments using pixel-based classifiers. However, although the SVM classification was successful, only landslides with larger areas were captured by the algorithms, confirming the importance of conducting further analyses using images with finer spatial resolution.

Application of Object-Based Image Analysis for Detecting and Differentiating between Shallow Landslides and Debris Flows

“…The accuracy metrics produced results similar to those reported in the literature Dias et al, 2021c;Soares et al, 2022), namely for shallow landslides an F1 Score of 78% and an OA of 87; and for debris flows an F1 score of 86% and an OA of 89%. Satisfactory results were achieved for the spatial accuracy metrics.…”

“…TP represents objects correctly classified; FP represents objects wrongly classified; TN represents objects correctly not classified as shallow landslide or debris flow; FN represents objects incorrectly not classified as shallow landslide or debris flow. PA indicates the probability that a given object has been correctly classified, and UA indicates the probability that a classified object actually represents this class (Dias et al, 2021c). In addition, spatial accuracy metrics were applied (Eisank et al 2014b;Hölbling et al, 2017).…”

“…These findings are similar to those of other remote sensingbased landslide mapping studies in Brazil. These studies include Dias et al (2021c), who applied three pixel-based algorithms in São Paulo state and found that the most common errors were related to non-vegetated areas, such as pasture classified as landslides, or clouds and stream channels. The same authors also reported difficulties in classifying small landslide scars.…”

“…Creating a landslide inventory is the first and main step for susceptibility, vulnerability and risk analysis, and it should be noted that the accuracy of any inventory has a direct effect on the efficiency of subsequent studies. Manual, semi-automated and automated methods have been used to identify and map this type of hazard (Carou et al, 2017;Hölbling et al, 2017;Barella et al, 2019;Comert et al, 2019;Canavesi et al, 2020;Karantanellis et al, 2021;Dias et al, 2021c;Soares et al, 2022;Liang et al, 2022). Object-based image analysis (OBIA) has a high potential for landslide recognition in satellite imagery (Martha et al, 2010;Hölbling et al, 2017;Comert et al, 2019;Hölbling, 2022).…”

Remote sensing landslide recognition method based on LinkNet and attention mechanism

“…The object-oriented landslide recognition method uses objects to reduce spectral variance correlation in each information class, and combines the features related to objects, such as shape and texture, to make the results more reasonable [5,6]. But the disadvantage is that noise information will be generated in the process of segmentation, and these units will be ignored in classification, so there will be some limitations.…”