The purpose of the present study is to compare the prediction performances of three different approaches such as decision tree (DT), support vector machine (SVM) and adaptive neuro-fuzzy inference ...system (ANFIS) for landslide susceptibility mapping at Penang Hill area, Malaysia. The necessary input parameters for the landslide susceptibility assessments were obtained from various sources. At first, landslide locations were identified by aerial photographs and field surveys and a total of 113 landslide locations were constructed. The study area contains 340,608pixels while total 8403pixels include landslides. The landslide inventory was randomly partitioned into two subsets: (1) part 1 that contains 50% (4000 landslide grid cells) was used in the training phase of the models; (2) part 2 is a validation dataset 50% (4000 landslide grid cells) for validation of three models and to confirm its accuracy. The digitally processed images of input parameters were combined in GIS. Finally, landslide susceptibility maps were produced, and the performances were assessed and discussed. Total fifteen landslide susceptibility maps were produced using DT, SVM and ANFIS based models, and the resultant maps were validated using the landslide locations. Prediction performances of these maps were checked by receiver operating characteristics (ROC) by using both success rate curve and prediction rate curve. The validation results showed that, area under the ROC curve for the fifteen models produced using DT, SVM and ANFIS varied from 0.8204 to 0.9421 for success rate curve and 0.7580 to 0.8307 for prediction rate curves, respectively. Moreover, the prediction curves revealed that model 5 of DT has slightly higher prediction performance (83.07), whereas the success rate showed that model 5 of ANFIS has better prediction (94.21) capability among all models. The results of this study showed that landslide susceptibility mapping in the Penang Hill area using the three approaches (e.g., DT, SVM and ANFIS) is viable. As far as the performance of the models are concerned, the results appeared to be quite satisfactory, i.e., the zones determined on the map being zones of relative susceptibility.
► I used three soft computing models i.e., DT, SVM and ANFIS for landslide mapping. ► Total 15 susceptibility maps were produced and their performances were assessed. ► Among all models, model 5 using DT showed highest prediction performance. ► Results showed all 3 models are efficient for landslide susceptibility mapping.
Landslides are one of the most frequent and common natural hazards in Malaysia. Preparation of landslide susceptibility maps is one of the first and most important steps in the landslide hazard ...mitigation. However, due to complex nature of landslides, producing a reliable susceptibility map is not easy. For this reason, a number of different approaches have been used, including direct and indirect heuristic approaches, deterministic, probabilistic, statistical, and data mining approaches. Moreover, these landslides can be systematically assessed and mapped through a traditional mapping framework using geoinformation technologies. Since the early 1990s, several mathematical models have been developed and applied to landslide hazard mapping using geographic information system (GIS). Among various approaches, fuzzy logic relation for mapping landslide susceptibility is one of the techniques that allows to describe the role of each predisposing factor (landslide-conditioning parameters) and their optimal combination. This paper presents a new attempt at landslide susceptibility mapping using fuzzy logic relations and their cross application of membership values to three study areas in Malaysia using a GIS. The possibility of capturing the judgment and the modeling of conditioning factors are the main advantages of using fuzzy logic. These models are capable to capture the conditioning factors directly affecting the landslides and also the inter-relationship among them. In the first stage of the study, a landslide inventory was complied for each of the three study areas using both field surveys and airphoto studies. Using total 12 topographic and lithological variables, landslide susceptibility models were developed using the fuzzy logic approach. Then the landslide inventory and the parameter maps were analyzed together using the fuzzy relations and the landslide susceptibility maps produced. Finally, the prediction performance of the susceptibility maps was checked by considering field-verified landslide locations in the studied areas. Further, the susceptibility maps were validated using the receiver-operating characteristics (ROC) success rate curves. The ROC curve technique is based on plotting model sensitivity—true positive fraction values calculated for different threshold values versus model specificity—true negative fraction values on a graph. The ROC curves were calculated for the landslide susceptibility maps obtained from the application and cross application of fuzzy logic relations. Qualitatively, the produced landslide susceptibility maps showed greater than 82% landslide susceptibility in all nine cases. The results indicated that, when compared with the landslide susceptibility maps, the landslides identified in the study areas were found to be located in the very high and high susceptibility zones. This shows that as far as the performance of the fuzzy logic relation approach is concerned, the results appeared to be quite satisfactory, the zones determined on the map being zones of relative susceptibility.
This paper presents the results of the cross-validation of a multivariate logistic regression model using remote sensing data and GIS for landslide hazard analysis on the Penang, Cameron, and ...Selangor areas in Malaysia. Landslide locations in the study areas were identified by interpreting aerial photographs and satellite images, supported by field surveys. SPOT 5 and Landsat TM satellite imagery were used to map landcover and vegetation index, respectively. Maps of topography, soil type, lineaments and land cover were constructed from the spatial datasets. Ten factors which influence landslide occurrence, i.e., slope, aspect, curvature, distance from drainage, lithology, distance from lineaments, soil type, landcover, rainfall precipitation, and normalized difference vegetation index (ndvi), were extracted from the spatial database and the logistic regression coefficient of each factor was computed. Then the landslide hazard was analysed using the multivariate logistic regression coefficients derived not only from the data for the respective area but also using the logistic regression coefficients calculated from each of the other two areas (nine hazard maps in all) as a cross-validation of the model. For verification of the model, the results of the analyses were then compared with the field-verified landslide locations. Among the three cases of the application of logistic regression coefficient in the same study area, the case of Selangor based on the Selangor logistic regression coefficients showed the highest accuracy (94%), where as Penang based on the Penang coefficients showed the lowest accuracy (86%). Similarly, among the six cases from the cross application of logistic regression coefficient in other two areas, the case of Selangor based on logistic coefficient of Cameron showed highest (90%) prediction accuracy where as the case of Penang based on the Selangor logistic regression coefficients showed the lowest accuracy (79%). Qualitatively, the cross application model yields reasonable results which can be used for preliminary landslide hazard mapping.
This study developed a deep learning based technique for the assessment of landslide susceptibility through a one-dimensional convolutional network (1D-CNN) and Bayesian optimisation in Southern ...Yangyang Province, South Korea. A total of 219 slide inventories and 17 slide conditioning variables were obtained for modelling. The data showed a complex scenario. Some past slides have spread over steep lands, while others have spread through flat terrain. Random forest (RF) served to keep only important factors for further analysis as a pre-processing measure. To select CNN hyperparameters, Bayesian optimization was used. Three methods contributed to overcoming the overfitting issue owing to small training data in our research. The selection of key factors by RF helped first of all to reduce information dimensionality. Second, the CNN model with 1D convolutions was intended to considerably decrease the number of its parameters. Third, a high rate of drop-out (0.66) helped reduce the CNN parameters. Overall accuracy, area under the receiver operating characteristics curve (AUROC) and 5-fold cross-validation were used to evaluate the models. CNN performance was compared to ANN and SVM. CNN achieved the highest accuracy on testing dataset (83.11%) and AUROC (0.880, 0.893, using testing and 5-fold CV, respectively). Bayesian optimization enhanced CNN accuracy by~3% (compared with default configuration). CNN could outperform ANN and SVM owing to its complicated architecture and handling of spatial correlations through convolution and pooling operations. In complex situations where some variables make a non-linear contribution to the occurrence of landslides, the method suggested could thus help develop landslide susceptibility maps.
•1D-CNN was used for landslide susceptibility mapping.•The proposed model was compared with feedforward neural networks and SVM.•Random Forest was used to identify significant landslide conditioning factors.•Bayesian optimization was applied to improve the model accuracy.•Model validation was accomplished using AUROC and 5-fold cross validation.
Urban vegetation mapping is critical in many applications, i.e., preserving biodiversity, maintaining ecological balance, and minimizing the urban heat island effect. It is still challenging to ...extract accurate vegetation covers from aerial imagery using traditional classification approaches, because urban vegetation categories have complex spatial structures and similar spectral properties. Deep neural networks (DNNs) have shown a significant improvement in remote sensing image classification outcomes during the last few years. These methods are promising in this domain, yet unreliable for various reasons, such as the use of irrelevant descriptor features in the building of the models and lack of quality in the labeled image. Explainable AI (XAI) can help us gain insight into these limits and, as a result, adjust the training dataset and model as needed. Thus, in this work, we explain how an explanation model called Shapley additive explanations (SHAP) can be utilized for interpreting the output of the DNN model that is designed for classifying vegetation covers. We want to not only produce high-quality vegetation maps, but also rank the input parameters and select appropriate features for classification. Therefore, we test our method on vegetation mapping from aerial imagery based on spectral and textural features. Texture features can help overcome the limitations of poor spectral resolution in aerial imagery for vegetation mapping. The model was capable of obtaining an overall accuracy (OA) of 94.44% for vegetation cover mapping. The conclusions derived from SHAP plots demonstrate the high contribution of features, such as Hue, Brightness, GLCM_Dissimilarity, GLCM_Homogeneity, and GLCM_Mean to the output of the proposed model for vegetation mapping. Therefore, the study indicates that existing vegetation mapping strategies based only on spectral characteristics are insufficient to appropriately classify vegetation covers.
•Decision tree (DT) machine learning algorithm was used to map the flood susceptible areas in Kelantan, Malaysia.•We used an ensemble frequency ratio (FR) and logistic regression (LR) model in order ...to overcome weak points of the LR.•Combined method of FR and LR was used to map the susceptible areas in Kelantan, Malaysia.•Results of both methods were compared and their efficiency was assessed.•Most influencing conditioning factors on flooding were recognized.
Flood is one of the natural hazards which occur all over the world and it is critical to be controlled through proper management. Severe flood events in Kelantan, Malaysia cause damage to both life and property every year, and therefore the development of flood model to recognize the susceptible areas in watersheds is important for decision makers. Remote sensing (RS) and geographic information system (GIS) techniques could be useful in hydrological studies while they are able to fulfill all the requirements for comprehensive, rapid and accurate analysis. The aim of the current research is to compare the prediction performances of two different approaches such as rule-based decision tree (DT) and combination of frequency ratio (FR) and logistic regression (LR) statistical methods for flood susceptibility mapping at Kelantan, Malaysia. DT is based on the rules which are created precisely and strongly by considering all the characteristics of the variables which can enhance the performance of the flood susceptibility mapping. On the other hand, LR as multivariate statistical analysis (MSA) has some weak points. For that reason, FR was used to analyze the impact of classes of each variable on flood occurrence and overcome the weakness of LR. At first, flood inventory map with a total of 155 flood locations was extracted from various sources over the part of the Kelantan. Then the flood inventory data was randomly divided into a testing dataset 70% (115 flood locations) for training the models and the remaining 30% (40 flood locations) was used for validation purpose. The spatial database includes digital elevation model (DEM), curvature, geology, river, stream power index (SPI), rainfall, land use/cover (LULC), soil type, topographic wetness index (TWI) and slope. For validation both success and prediction rate curves were performed. The validation results showed that, area under the curve for the results of DT and integrated method of FR and LR was 87% and 90% for success rate and 82% and 83% for prediction rate respectively.
Landslide inventories are in high demand for risk assessment of this natural hazard, particularly in tropical mountainous regions. This research designed residual networks for landslide detection ...using spectral (RGB bands) and topographic information (altitude, slope, aspect, curvature). Recent studies indicate that deep learning methods such as convolutional neural networks (CNN) improve landslide mapping results compared to traditional machine learning. But the effects of network architecture designs and data fusion remain largely underexplored in landslide detection. We compared a one-layer CNN with two of its deeper counterparts and residual networks with two fusion strategies (layer stacking and feature-level fusion) to detect landslides in Cameron Highlands, Malaysia. Sixteen different maps were created using proposed methods and evaluated in separate training and testing sub-areas based on overall accuracy, F1-score, and mean intersection over union (mIOU) metrics. When layer stacking is used as a fusion approach, none of the network designs improved landslide detection results. However, our findings showed that when using feature-level fusion, results could be enhanced with the same network designs. Residual networks performed best improving F1-score and mIOU by 0.13 and 12.96%, respectively, using feature-level fusion rather than layer stacking. CNN models also enhanced the detection outcome with the same fusion approach. On single modality datasets, models' performance varies according to input data, highlighting the effects of input data on network architecture selection. In general, residual networks found to converge faster and generalize better to test areas than other models tested in this research.
This paper summarizes findings of landslide hazard analysis on Penang Island, Malaysia, using frequency ratio, logistic regression, and artificial neural network models with the aid of GIS tools and ...remote sensing data. Landslide locations were identified and an inventory map was constructed by trained geomorphologists using photo-interpretation from archived aerial photographs supported by field surveys. A SPOT 5 satellite pan sharpened image acquired in January 2005 was used for land-cover classification supported by a topographic map. The above digitally processed images were subsequently combined in a GIS with ancillary data, for example topographical (slope, aspect, curvature, drainage), geological (litho types and lineaments), soil types, and normalized difference vegetation index (NDVI) data, and used to construct a spatial database using GIS and image processing. Three landslide hazard maps were constructed on the basis of landslide inventories and thematic layers, using frequency ratio, logistic regression, and artificial neural network models. Further, each thematic layer’s weight was determined by the back-propagation training method and landslide hazard indices were calculated using the trained back-propagation weights. The results of the analysis were verified and compared using the landslide location data and the accuracy observed was 86.41, 89.59, and 83.55% for frequency ratio, logistic regression, and artificial neural network models, respectively. On the basis of the higher percentages of landslide bodies predicted in very highly hazardous and highly hazardous zones, the results obtained by use of the logistic regression model were slightly more accurate than those from the other models used for landslide hazard analysis. The results from the neural network model suggest the effect of topographic slope is the highest and most important factor with weightage value (1.0), which is more than twice that of the other factors, followed by the NDVI (0.52), and then precipitation (0.42). Further, the results revealed that distance from lineament has the lowest weightage, with a value of 0. This shows that in the study area, fault lines and structural features do not contribute much to landslide triggering.
One of the most challenging research subjects in remote sensing is feature extraction, such as road features, from remote sensing images. Such an extraction influences multiple scenes, including map ...updating, traffic management, emergency tasks, road monitoring, and others. Therefore, a systematic review of deep learning techniques applied to common remote sensing benchmarks for road extraction is conducted in this study. The research is conducted based on four main types of deep learning methods, namely, the GANs model, deconvolutional networks, FCNs, and patch-based CNNs models. We also compare these various deep learning models applied to remote sensing datasets to show which method performs well in extracting road parts from high-resolution remote sensing images. Moreover, we describe future research directions and research gaps. Results indicate that the largest reported performance record is related to the deconvolutional nets applied to remote sensing images, and the F1 score metric of the generative adversarial network model, DenseNet method, and FCN-32 applied to UAV and Google Earth images are high: 96.08%, 95.72%, and 94.59%, respectively.
PDF
