GIS and ANN model for landslide susceptibility mapping

1 IntroductionThe population growth and the expansion of settlements and life-lines over hazardous areas exert increasingly great impact of natural disasters both in the developed and developing countries. In many countries, the economic losses and casualties due to landslides are greater than commonly recognized and generate a yearly loss of property larger than that from any other natural disasters, including earthquakes, floods and windstorms. Landslides in mountainous terrain often occur a…     

GIS techniques for regional-scale landslide susceptibility assessment: the Sicily (Italy) case study

This study describes the assessment of landslide susceptibility in Sicily (Italy) at a 1:100,000 scale using a multivariate logistic regression model. The model was implemented in a GIS environment by using the ArcSDM (Arc Spatial Data Modeller) module, modified to develop spatial prediction through regional data sets. A newly developed algorithm was used to automatically extract the detachment area from mapped landslide polygons. The following factors were selected as independent variables of the logistic regression model: slope gradient, lithology, land cover, a curve number derived index and a pluviometric anomaly index. The above-described configuration has been verified to be the best one among others employing from three to eight factors. All the regression coefficients and parameters were calculated using selected landslide training data sets. The results of the analysis were validated using an independent landslide data set. On an average, 82% of the area affected by instability and 79% of the not affected area were correctly classified by the model, which proved to be a useful tool for planners and decision-makers.     

Uncertainty due to DEM error in landslide susceptibility mapping

Terrain attributes such as slope gradient and slope shape, computed from a gridded digital elevation model (DEM), are important input data for landslide susceptibility mapping. Errors in DEM can cause uncertainty in terrain attributes and thus influence landslide susceptibility mapping. Monte Carlo simulations have been used in this article to compare uncertainties due to DEM error in two representative landslide susceptibility mapping approaches: a recently developed expert knowledge and fuzzy logic-based approach to landslide susceptibility mapping (efLandslides), and a logistic regression approach that is representative of multivariate statistical approaches to landslide susceptibility mapping. The study area is located in the middle and upper reaches of the Yangtze River, China, and includes two adjacent areas with similar environmental conditions – one for efLandslides model development (approximately 250 km²) and the other for model extrapolation (approximately 4600 km²). Sequential Gaussian simulation was used to simulate DEM error fields at 25-m resolution with different magnitudes and spatial autocorrelation levels. Nine sets of simulations were generated. Each set included 100 realizations derived from a DEM error field specified by possible combinations of three standard deviation values (1, 7.5, and 15 m) for error magnitude and three range values (0, 60, and 120 m) for spatial autocorrelation. The overall uncertainties of both efLandslides and the logistic regression approach attributable to each model-simulated DEM error were evaluated based on a map of standard deviations of landslide susceptibility realizations. The uncertainty assessment showed that the overall uncertainty in efLandslides was less sensitive to DEM error than that in the logistic regression approach and that the overall uncertainties in both efLandslides and the logistic regression approach for the model-extrapolation area were generally lower than in the model-development area used in this study. Boxplots were produced by associating an independent validation set of 205 observed landslides in the model-extrapolation area with the resulting landslide susceptibility realizations. These boxplots showed that for all simulations, efLandslides produced more reasonable results than logistic regression.     

An uncertainty and sensitivity analysis approach for GIS-based multicriteria landslide susceptibility mapping

GIS-based multicriteria decision analysis (MCDA) methods are increasingly being used in landslide susceptibility mapping. However, the uncertainties that are associated with MCDA techniques may significantly impact the results. This may sometimes lead to inaccurate outcomes and undesirable consequences. This article introduces a new GIS-based MCDA approach. We illustrate the consequences of applying different MCDA methods within a decision-making process through uncertainty analysis. Three GIS-MCDA methods in conjunction with Monte Carlo simulation (MCS) and Dempster–Shafer theory are analyzed for landslide susceptibility mapping (LSM) in the Urmia lake basin in Iran, which is highly susceptible to landslide hazards. The methodology comprises three stages. First, the LSM criteria are ranked and a sensitivity analysis is implemented to simulate error propagation based on the MCS. The resulting weights are expressed through probability density functions. Accordingly, within the second stage, three MCDA methods, namely analytical hierarchy process (AHP), weighted linear combination (WLC) and ordered weighted average (OWA), are used to produce the landslide susceptibility maps. In the third stage, accuracy assessments are carried out and the uncertainties of the different results are measured. We compare the accuracies of the three MCDA methods based on (1) the Dempster–Shafer theory and (2) a validation of the results using an inventory of known landslides and their respective coverage based on object-based image analysis of IRS-ID satellite images. The results of this study reveal that through the integration of GIS and MCDA models, it is possible to identify strategies for choosing an appropriate method for LSM. Furthermore, our findings indicate that the integration of MCDA and MCS can significantly improve the accuracy of the results. In LSM, the AHP method performed best, while the OWA reveals better performance in the reliability assessment. The WLC operation yielded poor results.     

Landslide susceptibility evaluation based on BPNN and GIS: a case of Guojiaba in the Three Gorges Reservoir Area

A landslide susceptibility evaluation is vital for disaster management and development planning in the Yangtze River Three Gorges Reservoir Area. In this study, with the support of remote sensing and Geographic Information System, 4 factor groups comprising 10 separate subfactors of landslide-related data layers were selected to establish a susceptibility evaluation model based on the back-propagation neural network including slope, aspect, plan curvature, strata and lithology, distance to faults, land use/land cover, Normalized Difference Vegetation Index, Normalized Difference Water Index, distance from roads, and effect of rivers. During model development, a three-layered interconnected neural network structure of 10 (input layer) × 20 (hidden layer) × 1 (output layer) was used for evaluating the landslide susceptibility in Guojiaba. At the same time, a back-propagation algorithm was applied to calculate the weights between the input layer and the hidden layer and between the hidden layer and the output layer. The results showed that the effect of slope has the highest weight value (0.2051), which is more than two times that of the other factors, followed by strata and lithology (0.1213) and then the effect of rivers (0.1201). At the end of the susceptibility evaluation, the area was divided into four zones such as very high, high, moderate and low susceptibility. For verification, the receiver operating characteristic curve for the back-propagation neural network-derived landslide susceptibility evaluation model was drawn, and the results showed that the area under the receiver operating characteristic curve was 0.8790 and the prediction accuracy was 88%. Furthermore, the results obtained from this article were then verified by comparing with the existing landslide historical data and multiple field-verified results. Lastly, the landslide susceptibility map will help decision makers in risk management, site selection, site planning, and the design of control engineering.     

GIS-based landslide susceptibility mapping for the 2005 Kashmir earthquake region

The M_w 7.6 October 8, 2005 Kashmir earthquake triggered several thousand landslides throughout the Himalaya of northern Pakistan and India. These were concentrated in six different geomorphic–geologic–anthropogenic settings. A spatial database, which included 2252 landslides, was developed and analyzed using ASTER satellite imagery and geographical information system (GIS) technology. A multi-criterion evaluation was applied to determine the significance of event-controlling parameters in triggering the landslides. The parameters included lithology, faults, slope gradient, slope aspect, elevation, land cover, rivers and roads. The results showed four classes of landslide susceptibility. Furthermore, they indicated that lithology had the strongest influence on landsliding, particularly when the rock is highly fractured, such as in shale, slate, clastic sediments, and limestone and dolomite. Moreover, the proximity of the landslides to faults, rivers, and roads was also an important factor in helping to initiate failures. In addition, landslides occurred particularly in moderate elevations on south facing slopes. Shrub land, grassland, and also agricultural land were highly susceptible to failures, while forested slopes had few landslides. One-third of the study area was highly or very highly susceptible to future landsliding and requires immediate mitigation action. The rest of the region had a low or moderate susceptibility to landsliding and remains relatively stable. This study supports the view that (1) earthquake-triggered landslides are concentrated in specific zones associated with event-controlling parameters; and (2) in the western Himalaya deforestation and road construction contributed significantly to landsliding during and shortly after earthquakes.     

Predictive landslide susceptibility assessment using the weight-of-evidence method in north-central Garhwal Himalaya,India

Landslides are frequent natural disasters in mountainous regions, particularly in the Himalayas in India during the southwest monsoon season. Although scientific study of landslides has been in progress for years, no significant achievement has been made to preclude landsliding and allay disasters. This research was undertaken to understand the areal distribution of landslides based on geological formations and geomorphological processes, and to provide more precise information regarding slope instability and mechanisms of failure. After completing a landslide inventory, prepared through field investigation and satellite image analysis, 493 landslides, comprising 131 investigated in the field and 362 identified from satellite imagery, were identified and mapped. The areal distribution of these landslides shows that sites more prone to landsliding have moderate to steep slopes, the lithology of the Lesser Himalayan formations, and excavations for road corridors. Landslide susceptibility zones were delineated for the area using the weight-of-evidence method on the basis of the frequency and distribution of landslides. Weights of selected variables were computed on the basis of severity of triggering factors. The accuracy of landslide susceptibility zones, calculated statistically (R² = .93), suggests high accuracy of the model for predicting landsliding in the area.     

Spatially and temporally distributed modeling of landslide susceptibility

Mapping of landslide susceptibility in forested watersheds is important for management decisions. In forested watersheds, especially in mountainous areas, the spatial distribution of relevant parameters for landslide prediction is often unavailable. This paper presents a GIS-based modeling approach that includes representation of the uncertainty and variability inherent in parameters. In this approach, grid-based tools are used to integrate the Soil Moisture Routing (SMR) model and infinite slope model with probabilistic analysis. The SMR model is a daily water balance model that simulates the hydrology of forested watersheds by combining climate data, a digital elevation model, soil, and land use data. The infinite slope model is used for slope stability analysis and determining the factor of safety for a slope. Monte Carlo simulation is used to incorporate the variability of input parameters and account for uncertainties associated with the evaluation of landslide susceptibility. This integrated approach of dynamic slope stability analysis was applied to the 72-km² Pete King watershed located in the Clearwater National Forest in north-central Idaho, USA, where landslides have occurred. A 30-year simulation was performed beginning with the existing vegetation covers that represented the watershed during the landslide year. Comparison of the GIS-based approach with existing models (FSmet and SHALSTAB) showed better precision of landslides based on the ratio of correctly identified landslides to susceptible areas. Analysis of landslide susceptibility showed that (1) the proportion of susceptible and non-susceptible cells changes spatially and temporally, (2) changed cells were a function of effective precipitation and soil storage amount, and (3) cell stability increased over time especially for clear-cut areas as root strength increased and vegetation transitioned to regenerated forest. Our modeling results showed that landslide susceptibility is strongly influenced by natural processes and human activities in space and time; while results from simulated outputs show the potential for decision-making in effective forest planning by using various management scenarios and controlling factors that influence landslide susceptibility. Such a process-based tool could be used to deal with real-dynamic systems to help decision-makers to answer complex landslide susceptibility questions.     

A procedure for landslide susceptibility zonation by the conditional analysis method

Numerous methods have been proposed for landslide probability zonation of the landscape by means of a Geographic Information System (GIS). Among the multivariate methods, i.e. those methods which simultaneously take into account all the factors contributing to instability, the Conditional Analysis method applied to a subdivision of the territory into Unique Condition Units is particularly straightforward from a conceptual point of view and particularly suited to the use of a GIS. In fact, working on the principle that future landslides are more likely to occur under those conditions which led to past instability, landslide susceptibility is defined by computing the landslide density in correspondence with different combinations of instability factors. The conceptual simplicity of this method, however, does not necessarily imply that it is simple to implement, especially as it requires rather complex operations and a high number of GIS commands. Moreover, there is the possibility that, in order to achieve satisfactory results, the procedure has to be repeated a few times changing the factors or modifying the class subdivision. To solve this problem, we created a shell program which, by combining the shell commands, the GIS Geographical Research Analysis Support System (GRASS) commands and the gawk language commands, carries out the whole procedure automatically. This makes the construction of a Landslide Susceptibility Map easy and fast for large areas too, and even when a high spatial resolution is adopted, as shown by application of the procedure to the Parma River basin, in the Italian Northern Apennines.     

基于GIS与BP神经网络的采空塌陷易发性预测

以湖北省鄂州程潮铁矿和黄石大冶铁矿为例,利用GIS空间分析功能对研究区数据进行提取分级、赋值统计及归一化等处理,构建了包括高程、坡度、地层、地下开采点的分布密度、相距最近地下开采点的距离、开采厚度与深度比值、蚀变接触带缓冲区、地下水深度以及地表地物类型的矿区采空塌陷易发性评价指标数据集;借助IDL语言调用Matlab神经网络工具箱,将研究区2011和2012年的指标数据集作为输入数据,塌陷易发性作为期望输出,建立基于BP神经网络的矿区采空塌陷易发性预测模型;通过选取并优化训练样本,实现对2013年矿山塌陷易发性的预测。结果表明,高易发区及以上的区域包含89.91%的采空塌陷,随着易发等级的提高,采空塌陷面积占易发等级面积比也随之增大;采空塌陷的分布具有明显的地带性,高易发区基本沿着岩体与围岩的接触带分布。模型解决了塌陷预测中的非线性映射问题,预测结果与实际调查情况基本吻合。BP神经网络模型与GIS技术相结合预测矿区采空塌陷的易发性具有可行性。     

一种基于地理环境相似度的滑坡负样本可信度度量方法

滑坡负样本在统计型滑坡危险度制图中具有重要作用,能抑制统计模型对滑坡危险度的高估。当前滑坡负样本采样方法采集的负样本可信度未知,在负样本采样过程中,极有可能将那些潜在滑坡点错选为负样本,这些假的负样本会降低负样本集的质量和训练样本集的质量,进而影响统计模型的精度。本文基于“地理环境越相似、地理特征越相似”的地理学常识,认为与正样本有着相似地理环境的点极有可能是未来发生滑坡的点;与正样本的地理环境越不相似的点,则越有可能是负样本。基于此假设提出一种基于地理环境相似度的负样本可信度度量方法,将该方法应用于滑坡灾害频发的陇南山区油房沟流域,对油房沟进行滑坡负样本可信度评价制图;使用油房沟流域的滑坡发生初始面来验证该方法的有效性。结果发现：滑坡发生初始面上所有栅格点的负样本可信度平均值为0.26,超过95%的栅格点的负样本可信度都小于0.5,说明本文提出的负样本可信度度量方法合理。     

Comparison of landslide susceptibility based on a decision-tree model and actual landslide occurrence: The Akaishi Mountains, Japan

This paper proposes a statistical decision-tree model to analyze landslide susceptibility in a wide area of the Akaishi Mountains, Japan. The objectives of this study were to validate the decision-tree model by comparing landslide susceptibility and actual landslide occurrence, and to reveal the relationships among landslide occurrence, topography, and geology. Landslide susceptibility was examined through ensemble learning with a decision tree. Decision trees are advantageous in that estimation processes and order of important explanatory variables are explicitly represented by the tree structures. Topographic characteristics (elevation, slope angle, profile curvature, plan curvature, and dissection and undissection height) and geological data were used as the explanatory variables. These topographic characteristics were calculated from digital elevation models (DEMs). The objective variables were landslide occurrence and reactivation data between 1992 and 2002 that were depicted by satellite image analysis. Landslide susceptibility was validated by comparing actual data on landslides that occurred and reactivated after the model was constructed (between 2002 and 2004).This study revealed that, from 2002 to 2004, landslides tended to occur and reactivate in catchments with high landslide susceptibility. The landslide susceptibility map thus depicts the actual landslide occurrence and reactivation in the Akaishi Mountains. This result indicates that the decision-tree model has appropriate accuracy for estimating the probabilities of future landslides. The tree structure indicates that landslides occurred and reactivated frequently in the catchments that had an average slope angle exceeding ca. 29° and a mode of slope angle exceeding 33°, which agree well with previous studies. A decision tree also quantitatively expresses important explanatory variables at the higher order of the tree structure.     

Landslide susceptibility mapping using GIS and the weight-of-evidence model

The weights-of-evidence model (a Bayesian probability model) was applied to the task of evaluating landslide susceptibility using GIS. Using landslide location and a spatial database containing information such as topography, soil, forest, geology, land cover and lineament, the weights-of-evidence model was applied to calculate each relevant factor's rating for the Boun area in Korea, which had suffered substantial landslide damage following heavy rain in 1998. In the topographic database, the factors were slope, aspect and curvature; in the soil database, they were soil texture, soil material, soil drainage, soil effective thickness and topographic type; in the forest map, they were forest type, timber diameter, timber age and forest density; lithology was derived from the geological database; land-use information came from Landsat TM satellite imagery; and lineament data from IRS satellite imagery. Tests of conditional independence were performed for the selection of factors, allowing 43 combinations of factors to be analysed. For the analysis of mapping landslide susceptibility, the contrast values, W ⁺ and W ^-, of each factor's rating were overlaid spatially. The results of the analysis were validated using the previous landslide locations. The combination of slope, curvature, topography, timber diameter, geology and lineament showed the best results. The results can be used for hazard prevention and land-use planning.     

Validation of a region-wide model of landslide susceptibility in the Manawatu–Wanganui region of New Zealand

Since European settlement 160 years ago, much of the indigenous forest in New Zealand hill country has been cleared for pastoral agriculture, resulting in increased erosion and sedimentation. To prioritise soil conservation work in the Manawatu–Wanganui region, we developed a model of landslide susceptibility. It assigns high susceptibility to steep land not protected by woody vegetation and low susceptibility everywhere else, following the commonly used approach for identifying inappropriate land use. A major storm on 15–16 February 2004 that produced many landslides was used to validate the model. The model predicted hills at risk to landsliding with moderate accuracy: 58% of erosion scars in the February storm occurred on hillsides considered to be susceptible. The model concept of slope thresholds, above which the probability of landsliding is high and below which the probability is low, is not adequate because below 30° the probability of landsliding is approximately linearly related to slope. Thus, reforestation of steep slopes will need to be combined with improved vegetation management for soil conservation on moderate slopes to significantly reduce future landsliding.     

GIS analysis to assess landslide susceptibility in a fluvial basin of NW Sicily (Italy)

Landslide hazard assessment, effected by means of geostatistical methods, is based on the analysis of the relationships between landslides and the spatial distributions of some instability factors. Frequently such analyses are based on landslide inventories in which each record represents the entire unstable area and is managed as a single instability landform. In this research, landslide susceptibility is evaluated through the study of a variety of instability landforms: landslides, scarps and areas uphill from crown. The instability factors selected were: bedrock lithology, steepness, topographic wetness index and stream power index. The instability landform densities computed for all the factors, which were arranged in Unique Condition Unit, allowed us to derive a total of three prediction images for each landslide typology. The role of the instability factors and the effects generated by the use of different landforms were analyzed by means of: a) bivariate analysis of the relationships between factors and landslide density; b) predictive power validations of the prediction images, based on a random partition strategy.The test area was the Iato River Basin (North-Western Sicily), whose slopes are moderately involved in flow and rotational slide landslides (219 and 28, respectively). The area is mainly made up of the following complexes: Numidian Flysch clays (19%, 1%), Terravecchia sandy clays (5%, 1%), Terravecchia clayey sands (3%, 0.3%) and San Cipirello marly clays (9%, 0%). The steepness parameter shows the highest landslide density in the [11–19°] class for both the typologies (8%, 1%), even if the density distributions for rotational slides are right-asymmetric and right-shifted. We obtained significant differences in shape when we used different instability landforms. Unlike scarps and areas uphill from crowns, landslide areas produce left-asymmetric and left-shifted density distributions for both the typologies. As far as the topographic wetness index is concerned, much more pronounced differences were detected among the instability landforms of rotational slides. In contrast, the flow landslides produce normal-like density distributions. The latter and the rotational slide landslide areas produce the highest density values in the class [5.5–6.7], despite an abrupt decreasing trend starting from the first class [3.2–4.4], which is generated by the density values of the rotational slide scarps and areas uphill from crowns. The stream power index at the foot of the slopes, which was automatically derived using a GIS-procedure, shows a positive correlation with the landslide densities marked by the maximum classes: [4.8–6.0] for flows, and [6.0–7.2] for rotational slides. The validation procedure results confirmed that the choice of instability landform influences the results of the susceptibility analysis. Furthermore, the validation procedure indicates that: a) the predictive models are generally satisfactory; b) scarps and zones uphill from crown areas are the most diagnostically unstable landforms, for flow and rotational slide landslides respectively.     

A landslide susceptibility model using the Analytical Hierarchy Process method and multivariate statistics in perialpine Slovenia

Landslides cause damage to property and unfortunately pose a threat even to human lives. Good landslide susceptibility, hazard, and risk models could help mitigate or even avoid the unwanted consequences resulted from such hillslope mass movements. For the purpose of landslide susceptibility assessment the study area in the central Slovenia was divided to 78 365 slope units, for which 24 statistical variables were calculated. For the land-use and vegetation data, multi-spectral high-resolution images were merged using Principal Component Analysis method and classified with an unsupervised classification. Using multivariate statistical analysis (factor analysis), the interactions between factors and landslide distribution were tested, and the importance of individual factors for landslide occurrence was defined. The results show that the slope, the lithology, the terrain roughness, and the cover type play important roles in landslide susceptibility. The importance of other spatial factors varies depending on the landslide type. Based on the statistical results several landslide susceptibility models were developed using the Analytical Hierarchy Process method. These models gave very different results, with a prediction error ranging from 4.3% to 73%. As a final result of the research, the weights of important spatial factors from the best models were derived with the AHP method. Using probability measures, potentially hazardous areas were located in relation to population and road distribution, and hazard classes were assessed.     

Urban expansion in Auckland,New Zealand: a GIS simulation via an intelligent self-adapting multiscale agent-based model

ABSTRACT

Abstract: When modelling urban expansion dynamics, cellular automata models focus mostly on the physical environments and cell neighbours, but ignore the ‘human’ aspect of the allocation of urban expansion cells. This limitation is overcome here using an intelligent self-adapting multiscale agent-based model. To simulate the urban expansion of Auckland, New Zealand, a total of 15 urban expansion drivers/constraints were considered over two periods (2000–2005, 2005–2010). The modelling takes into consideration both a macro-scale agent (government) and micro-scale agents (residents of three income levels), and their multi-level interactions. In order to achieve reliable simulation results, ABM was coupled with an artificial neural network to reveal the learning process and heterogeneity of the multi-sub-residential agents. The ANN-ABM accurately simulated the urban expansion of Auckland at both the global and local scales, with kappa simulation value at 0.48 and 0.55, respectively. The validated simulation result shows that the intelligent and self-adapting ANN-ABM approach is more accurate than an ABM with a general type of agent model (kappa simulation = 0.42) at the global scale, and more accurate than an ANN-based CA model (kappa simulation = 0.47) at the local scale. Simulation inaccuracy stems mostly from the outdated master land use plan.     

Qualitative landslide susceptibility assessment by multicriteria analysis: A case study from San Antonio del Sur, Guantánamo, Cuba

Geomorphological information can be combined with decision-support tools to assess landslide hazard and risk. A heuristic model was applied to a rural municipality in eastern Cuba. The study is based on a terrain mapping units (TMU) map, generated at 1:50,000 scale by interpretation of aerial photos, satellite images and field data. Information describing 603 terrain units was collected in a database. Landslide areas were mapped in detail to classify the different failure types and parts. Three major landslide regions are recognized in the study area: coastal hills with rockfalls, shallow debris flows and old rotational rockslides denudational slopes in limestone, with very large deep-seated rockslides related to tectonic activity and the Sierra de Caujerí scarp, with large rockslides. The Caujerí scarp presents the highest hazard, with recent landslides and various signs of active processes. The different landforms and the causative factors for landslides were analyzed and used to develop the heuristic model. The model is based on weights assigned by expert judgment and organized in a number of components such as slope angle, internal relief, slope shape, geological formation, active faults, distance to drainage, distance to springs, geomorphological subunits and existing landslide zones. From these variables a hierarchical heuristic model was applied in which three levels of weights were designed for classes, variables, and criteria. The model combines all weights into a single hazard value for each pixel of the landslide hazard map. The hazard map was then divided by two scales, one with three classes for disaster managers and one with 10 detailed hazard classes for technical staff. The range of weight values and the number of existing landslides is registered for each class. The resulting increasing landslide density with higher hazard classes indicates that the output map is reliable. The landslide hazard map was used in combination with existing information on buildings and infrastructure to prepare a qualitative risk map. The complete lack of historical landslide information and geotechnical data precludes the development of quantitative deterministic or probabilistic models.     

基于GIS的区域群发性降雨型滑坡时空预报研究

以滑坡灾害突出的雅安市雨城区为例,综合考虑降雨强度、前期降雨量及下垫面(地形、岩性、植被覆盖等)构建了基于GIS分析获取的易发指数+BP型神经网络时空预报模型。首先通过试验确定了模型的网络参数和网络结构,然后通过危险性区划图获取降雨型滑坡易发指数,并利用GIS的空间插值功能和雨量站数据获取相应降雨型滑坡的雨量数据,将量化后下垫面的易发指数和降雨数据作为神经元输入层数据。将模型应用于研究区,其中46个降雨型滑坡数据作为训练样本,10个降雨型滑坡数据作为检验样本,预测精度达到90%,显示该模型对于降雨型滑坡的时空预报精度较高。