Recurrent floods are one of the major global threats among people, particularly in developing countries like India, as this nation has a tropical monsoon type of climate. Therefore, flood susceptibility (FS) mapping is indeed necessary to overcome this type of natural hazard phenomena. With this in mind, we evaluated the prediction performance of FS mapping in the Koiya River basin, Eastern India. The present research work was done through preparation of a sophisticated flood inventory map; eight flood conditioning variables were selected based on the topography and hydro-climatological condition, and by applying the novel ensemble approach of hyperpipes (HP) and support vector regression (SVR) machine learning (ML) algorithms. The ensemble approach of HP-SVR was also compared with the stand-alone ML algorithms of HP and SVR. In relative importance of variables, distance to river was the most dominant factor for flood occurrences followed by rainfall, land use land cover (LULC), and normalized difference vegetation index (NDVI). The validation and accuracy assessment of FS maps was done through five popular statistical methods. The result of accuracy evaluation showed that the ensemble approach is the most optimal model (AUC = 0.915, sensitivity = 0.932, specificity = 0.902, accuracy = 0.928 and Kappa = 0.835) in FS assessment, followed by HP (AUC = 0.885) and SVR (AUC = 0.871).

This work focuses on comparing results of flood susceptibility modelling in the part of Middle Ganga Plain, Ganga foreland basin. Following inclusivity rule, 12 major flood explanatory factors including a new one, geomorphology, have been utilized. Out of 1000 randomly generated flood-points from 2008 Landsat 5 TM image derived flood polygon, 70% have been utilized for the training purpose of Shannon’s entropy (SE) model and 30% for area under receiver operating characteristic (AUROC) method of validation of both, SE and frequency ratio (FR), models. Result from FR shows that the contributions of specific classes of different explanatory factors to flooding susceptibility vary whereas the SE model suggests that geomorphology is the most contributing factor. The AUROC curve for SE (0.90) was better than that for FR (0.85). Hence, the SE model predicts flood-susceptible areas more accurately than the multivariate statistical model FR in the study area.

There is an evident increase in the importance that remote sensing sensors play in the monitoring and evaluation of natural hazards susceptibility and risk. The present study aims to assess the flash-flood potential values, in a small catchment from Romania, using information provided remote sensing sensors and Geographic Informational Systems (GIS) databases which were involved as input data into a number of four ensemble models. In a first phase, with the help of high-resolution satellite images from the Google Earth application, 481 points affected by torrential processes were acquired, another 481 points being randomly positioned in areas without torrential processes. Seventy percent of the dataset was kept as training data, while the other 30% was assigned to validating sample. Further, in order to train the machine learning models, information regarding the 10 flash-flood predictors was extracted in the training sample locations. Finally, the following four ensembles were used to calculate the Flash-Flood Potential Index across the Bâsca Chiojdului river basin: Deep Learning Neural Network–Frequency Ratio (DLNN-FR), Deep Learning Neural Network–Weights of Evidence (DLNN-WOE), Alternating Decision Trees–Frequency Ratio (ADT-FR) and Alternating Decision Trees–Weights of Evidence (ADT-WOE). The model’s performances were assessed using several statistical metrics. Thus, in terms of Sensitivity, the highest value of 0.985 was achieved by the DLNN-FR model, meanwhile the lowest one (0.866) was assigned to ADT-FR ensemble. Moreover, the specificity analysis shows that the highest value (0.991) was attributed to DLNN-WOE algorithm, while the lowest value (0.892) was achieved by ADT-FR. During the training procedure, the models achieved overall accuracies between 0.878 (ADT-FR) and 0.985 (DLNN-WOE). K-index shows again that the most performant model was DLNN-WOE (0.97). The Flash-Flood Potential Index (FFPI) values revealed that the surfaces with high and very high flash-flood susceptibility cover between 46.57% (DLNN-FR) and 59.38% (ADT-FR) of the study zone. The use of the Receiver Operating Characteristic (ROC) curve for results validation highlights the fact that FFPI_DLNN-WOE is characterized by the most precise results with an Area Under Curve of 0.96.

This study has developed a new ensemble model and tested another ensemble model for flood susceptibility mapping in the Middle Ganga Plain (MGP). The results of these two models have been quantitatively compared for performance analysis in zoning flood susceptible areas of low altitudinal range, humid subtropical fluvial floodplain environment of the Middle Ganga Plain (MGP). This part of the MGP, which is in the central Ganga River Basin (GRB), is experiencing worse floods in the changing climatic scenario causing an increased level of loss of life and property. The MGP experiencing monsoonal subtropical humid climate, active tectonics induced ground subsidence, increasing population, and shifting landuse/landcover trends and pattern, is the best natural laboratory to test all the susceptibility prediction genre of models to achieve the choice of best performing model with the constant number of input parameters for this type of topoclimatic environmental setting. This will help in achieving the goal of model universality, i.e., finding out the best performing susceptibility prediction model for this type of topoclimatic setting with the similar number and type of input variables. Based on the highly accurate flood inventory and using 12 flood predictors (FPs) (selected using field experience of the study area and literature survey), two machine learning (ML) ensemble models developed by bagging frequency ratio (FR) and evidential belief function (EBF) with classification and regression tree (CART), CART-FR and CART-EBF, were applied for flood susceptibility zonation mapping. Flood and non-flood points randomly generated using flood inventory have been apportioned in 70:30 ratio for training and validation of the ensembles. Based on the evaluation performance using threshold-independent evaluation statistic, area under receiver operating characteristic (AUROC) curve, 14 threshold-dependent evaluation metrices, and seed cell area index (SCAI) meant for assessing different aspects of ensembles, the study suggests that CART-EBF (AUC_SR= 0.843; AUC_PR= 0.819) was a better performant than CART-FR (AUC_SR= 0.828; AUC_PR= 0.802). The variability in performances of these novel-advanced ensembles and their comparison with results of other published models espouse the need of testing these as well as other genres of susceptibility models in other topoclimatic environments also. Results of this study are important for natural hazard managers and can be used to compute the damages through risk analysis.