1,721,102 research outputs found
Predicting gully occurrence at watershed scale: Comparing topographic indices and multivariate statistical models
Full text linkIn this study, the ability of five topographic indices to predict the gully trajectories observed in two adjacent watersheds located in Sicily (Italy) was evaluated. Two of these indices, named MSPI and MTWI, as far as we know, have never been employed to this aim. They were obtained by multiplying the stream power index (SPI) and the topographic wetness index (TWI), respectively, by the convergence index (CI). The predictive ability of the topographic indices was measured by using both cut-off independent (AUC: area under the receiver operating characteristic curve) and dependent statistics (Cohen's kappa index κ, sensitivity, specificity). These statistics were calculated also for 100 MARS (multivariate adaptive regression splines) and 100 LR (logistic regression) model runs, which used as predictors the topographic variables (i.e. contributing area, slope steepness, plan curvature and convergence index) combined into the five indices. Performance statistics of both topographic indices and statistical models were calculated using 100 random samples of 2 m grid cells, which were extracted only from flow concentration lines. This was done in order to focus the validation process on where gully erosion is more likely to occur. MSPI achieved the best predictive skill (AUC > 0.93; κ > 0.71) among the topographic indices and exhibited similar and better accuracy than local (i.e. trained and validated in the same watershed) and transferred (i.e. trained in one watershed and tested in the other one) LR models, respectively. On the other hand, MSPI performed similarly to transferred MARS runs (AUC > 0.92; κ > 0.71) but slightly worse than local MARS runs (AUC > 0.95; κ > 0.77). Based on the results of this experiment, it can be inferred that (i) including CI helps in detecting hollow areas where gullies are more likely to occur and (ii) MPSI can be a valid alternative to a data driven approach for mapping gully erosion susceptibility in areas where a gully inventory is not available, which is necessary to calibrate statistical models
Regional variability of terrain index and machine learning model applications for prediction of ephemeral gullies
No full textTerrain, or topographic, index models aggregate morphometric features of a landscape and use them to predict trajectories and initiation points of classical or ephemeral gullies. As an alternative to the index-based models, statistical machine learning approaches have been recently gaining an increased attention for identification of areas of gully susceptibility. Application of index-based or statistical models is normally restricted to the area of study, and regional model transferability is not well understood. In this study, seven terrain index models and two machine learning algorithms were applied to eight watersheds in Kansas and two watersheds in Sicily. The predictive ability of the nine models was measured by using both cut-off independent (area under the receiver operating characteristic curve) and dependent (Cohen's kappa index, sensitivity, specificity) statistics. The performance statistics of both terrain index and statistical models were cross-examined by finding an optimal threshold in one watershed and applying it to other watersheds. In Kansas, the model based on stream order (GORD) was found to perform similar to machine learning approaches, whereas the modified stream power index (MSPI) model overperformed other terrain index models in Sicily. Different landform characteristics in cultivated areas of the High Plains region in Kansas and in the steep hillslopes of Central Sicily caused models to have variable success rate when the indexes from one region were transferred to another region. The results also showed that a well-calibrated terrain index model, especially based on stream grid order, can be viewed as a valid alternative to a data driven approach for ephemeral gully mapping, but transferability of the optimal thresholds can be applied with caution
Improving the accuracy of rainfall prediction using a regionalization approach and neural networks
Full text linkSpatial and temporal analysis of precipitation patterns has become an intense research topic in contemporary climatology. Increasing the accuracy of precipitation prediction can have valuable results for decision-makers in a specific region. Hence, studies about precipitation prediction on a regional scale are of great importance. Artificial Neural Networks (ANN) have been widely used in climatological applications to predict different meteorological parameters. In this study, a method is presented to increase the accuracy of neural networks in precipitation prediction in Chaharmahal and Bakhtiari Province in Iran. For this purpose, monthly precipitation data recorded at 42 rain gauges during 1981-2012 were used. The stations were first clustered into well-defined groupings using Principal Component Analysis (PCA) and Cluster Analysis (CA), and then one separate neural network was applied to each group of stations. Another neural network model was also developed and applied to all the stations in order to measure the accuracy of the proposed model. Statistical results showed that the presented model produced better results in comparison to the second model
Assessment of LUNAR, iForest, LOF, and LSCP methodologies in delineating geochemical anomalies for mineral exploration
No full textGeochemical anomaly detection and delineation are crucial in mineral exploration, but they are challenged by high-dimensional data, complex inter-variable dependencies, and scarcity of ground truth labels for anomalies. Traditional outlier detection methods, including density-based and nearest-neighbor approaches, often misclassify anomalies close to the edges of the background data distribution, while ensemble methods face limitations in combining detectors effectively. Generic and global combination procedures frequently neglect local patterns in the data, leading to suboptimal detection of nuanced outlier characteristics, and the absence of robust selection processes can compromise ensemble performance due to underperforming detectors. To address these issues, this paper presents LUNAR (learnable unified neighborhood-based anomaly ranking), a novel outlier detection method that integrates graph neural networks with nearest neighbor analysis, and LSCP (locally selective combination in parallel outlier ensembles), which emphasizes local data structures and leverages pseudo-ground truth to optimize detector selection and improve score stability. This study also explored the efficacy of outlier detection methods, namely local outlier factor (LOF) and isolation forest (iForest) in detecting geochemical anomalies within the Varzaghan area, situated in the Ahar–Arasbaran zone of the Alborz–Azerbaijan Magmatic Belt. This region hosts diverse mineral occurrences, including porphyry Cu[sbnd]Mo deposits (e.g., Sungun), epithermal base metal veins (e.g., Zaylik), and Fe[sbnd]Cu skarn deposits (e.g., Sungun and Anjerd). Compared to the LOF and iForest, for the analysis of a trace element geochemical dataset from 1067 stream sediment samples, the LUNAR exhibited the highest relative percentage of delineated deposits along with superior AUC (area under curve) from ROC (receiver operating characteristic) analysis for both mineral occurrences and mineralized samples. The LOF-detected outliers for elements like As, Sb, and Ti, whereas the iForest-detected outliers for Ti, Pb, and Co, and the LUNAR-detected outliers for Au and pathfinder elements like As, Bi, and Sb. Employing a graph neural network, the LUNAR efficiently captured intricate outlier relationships within the multivariate geochemical dataset, surpassing the LOF. Spatial analysis uncovered a correlation between LSCP variants and the LUNAR in detecting geochemical anomalies and their association with known deposits. Based on AUC values, the LSCP_A (average) demonstrated relative superiority over the LSCP_AOM (average of maximum), LSCP_MOA (maximum of average), and LUNAR. Among the LSCP variants, the LSCP_A showcased superior performance, leveraging average scores, and detecting outliers of pathfinder elements for gold like As and Bi, along with lithologically-influenced elements like Cr and Ti, and the significant role of Cu. The mapping of clr-transformed Bi data aligned closely with mineral deposits, accentuating signatures typical of porphyry deposits in the Varzaghan district, including Cu, Au, Mo, and Bi. Compared to the iForest, the LSCP, particularly the LSCP_A, showcased proficiency in detecting geochemical anomalies through a localized approach and in comprehensively capturing diverse anomaly patterns, thus rendering it a promising method for handling complex datasets
Cloud-based interactive susceptibility modeling of gully erosion in Google Earth Engine
Full text linkThe gully erosion susceptibility literature is largely dominated by contributions focused on model comparison. This has led to prioritize certain aspects and leave others underdeveloped as compared to other natural hazard applications. For instance, in gully erosion data-driven modeling most studies use different platforms when it comes to data management, modeling and conversion into predictive maps. This in turn has limited the scope to catchment-scales. In this manuscript, we opt to propose a tool where the whole modeling procedure is unified within the same cloud computing system, allowing one to get rid of potential errors caused by input/output operations but also to extend the study areas indefinitely, as cloud data-management tools easily offer access to global data. Specifically, we present an interactive tool for susceptibility modeling in Google Earth Engine (GEE), the Susceptibility Tool for GEE (STGEE). Our tool requires few input data and makes use of the breadth of predictors’ information available in GEE. In this cloud computing environment, binary classifiers typical of susceptibility models can be called and fed with information related to mapping units and any natural hazards’ distribution over the geographic space. We tested our tool to generate susceptibility estimates for gully erosion occurrences in a study area located in Sicily (Italy). The tool we propose is equipped with a series of functions to aggregate the predictors’ information in space and time over a mapping unit of choice. Here we chose a Slope Unit partition but any polygonal structure can be chosen by the user. Once this information is derived, our tool calls for a Random Forest classifier to distinguish locations prone to gully erosion from locations where this process is not probabilistically expected to develop. This is done while providing a modeling performance overview, accessible via a separate panel. Such performance can be calculated on the basis of a exploratory analysis where all the information is used to fit a benchmark model as well as a spatial k-fold cross-validation scheme. Ultimately, the predictive function can be interactively used to generate susceptibility maps in real time, for the study area as well as any study area of interest. To promote the use of our tool, we are sharing it in a GitHub repository accessible at this link: https://github.com/giactitti/STGEE
Optimal slope units partitioning in landslide susceptibility mapping
Full text linkIn landslide susceptibility modeling, the selection of the mapping units is a very relevant topic both in terms of geomorphological adequacy and suitability of the models and final maps. In this paper, a test to integrate pixels and slope units is presented. MARS (Multivariate Adaptive Regression Splines) modeling was applied to assess landslide susceptibility based on a 12 predictors and a 1608 cases database. A pixel-based model was prepared and the scores zoned into 10 different types of slope units, obtained by differently combining two half-basin (HB) and four landform classification (LCL) coverages. The predictive performance of the 10 models were then compared to select the best performing one, whose prediction image was finally modified to consider also the propagation stage. The results attest integrating HB with LCL as more performing than using simple HB classification, with a very limited loss in predictive performance with respect to the pixel-based model
Landform classification: a high-performing mapping unit partitioning tool for landslide susceptibility assessment—a test in the Imera River basin (northern Sicily, Italy)
No full textIn landslide susceptibility studies, the type of mapping unit adopted affects the obtained models and maps in terms of accuracy, robustness, spatial resolution and geomorphological adequacy. To evaluate the optimal selection of these units, a test has been carried out in an important catchment of northern Sicily (the Imera River basin), where the spatial relationships between a set of predictors and an inventory of 1608 rotational/translational landslides were analysed using the multivariate adaptive regression splines (MARS) method. In particular, landslide susceptibility models were prepared and compared by adopting four different types of mapping units: the largely adopted grid cells (PX), the typical contributing area–controlled slope units (5000_SLU), the recently optimized parameter-free multiscale slope units (PF_SLU) and a new type (LCL_SLU) of slope unit obtained by crossing classic hydrological partitioning with landform classification. At the same time, once a pixel-based model was prepared, four different SLU modelling strategies were applied to each of the obtained slope unit layers, including two different types of pixel score zoning, a pixel score re-modelling and a factor-based SLU re-modelling. According to the achieved results, LCL_SLUs produced the highest performance and reliability, offering an optimal compromise between the high-performing but scattered and the smoothed but lower-performing prediction images that were obtained from pixel-based and hydrologic SLU–based modelling, respectively. Additionally, among the four adopted SLU modelling strategies, the new proposed procedure, which uses the zoned pixel–based score deciles as the LCL_SLU predictors for a new regression, resulted in the best outstanding performance (ROC_AUC = 0.95)
Predicting sediment deposition rate in check-dams using machine learning techniques and high-resolution DEMs
No full textSediments accumulated in check dams are a valuable measure to estimate soil erosion rates. Here, geographic information systems (GIS) and three machine learning techniques (MARS-multivariate adaptive regression splines, RF-random forest and SVM-support vector machine) were used, for the first time, to predict sediment deposition rate (SR) in check-dams located in six watersheds in SW Spain. There, 160 dry-stone check dams (~ 77.8 check-dams km−2), accumulated sediments during a period that varied from 11 to 23 years. The SR was estimated in former research using a topographical method and a high-resolution Digital Elevation Model (DEM) (average of 0.14 m3 ha−1 year−1). Nine environmental-topographic parameters were calculated and employed as predictors of the SR. The ability of MARS, RF and SVM was evaluated by using a five-fold cross-validation, considering the entire area (ALL), the check dams on the hillslope (HILL) and the valley-bottoms (VALLEY), as well as the three catchments (B, C and D) with the highest number of check dams. The accuracy of the models was assessed by the relative root mean square error (RRMSE) and the mean absolute error (MAE). The results revealed that RF and SVM are able to predict SR with higher and more stable accuracy than MARS. This is evident for the datasets ALL, VALLEY and D, where errors of prediction exhibited by MARS were from 44 to 77% (RRMSE) and from 37 to 62% (MAE) higher than those achieved by RF and SVM, but it also held for the datasets HILL and B where the difference of RRMSE and MAE was 7–10% and 12–17%, respectively
Monitoring of erosion on two calanchi fronts – Northern Sicily (Italy)
Full text linkIn the present research, two neighbouring calanchi fronts have been monitored by means of repeated
readings on erosion pins, that were carried out between November 2006 and October 2008. During the monitoring
period, a gauge station has been recording rainfalls, allowing us to compute the Rainfall-Runoff Erosivity Factor
of the USLE model. The research highlighted: i) a general correspondence between rainfalls temporal trends and
surface variation rhythms; ii) alternating erosion and deposition phases result in a retreat of the “calanchi” fronts
Detection of homogeneous precipitation regions at seasonal and annual time scales, northwest Iran
Full text linkDetection of homogeneous climate areas is a challenging issue, which can be affected by different criteria. One of the most prominent factors is choosing the time scale, which can lead to different spatial and temporal patterns. Total precipitation is a key factor in climatological studies, and studying its distribution is of utmost importance. The combination of principal components analysis and cluster analysis is used for homogeneous precipitation areas’ detection. Hence, the spatial pattern of total precipitation was investigated in northwestern Iran during the past two decades (1991–2010) on seasonal and annual time scales. The results of clustering on each time scale were validated, and well-defined clusters were investigated and compared with each other. Two homogeneous sub-regions were recognized in spring, the best period for depicting homogeneous precipitation clusters at seasonal resolution. The annual pattern of precipitation delineated three clusters in the study region. Finally, the characteristics of the well-clustered maps reveal the importance of time scale in detection of homogeneous precipitation sub-zones. © IWA Publishing 2017
- …
