Spatial modeling of floods using Artificial Neural Network (ANN) and analyst functions of GIS

With the accelerating rate for the development of human communities in form of urban and rural residential areas, necessity of forecasting and modeling various aspects of natural and human hazards, ensuring associated with controlling the risk of various hazards and other management measures in order to reducing their harmful effects among urban and regional planners and managers has been increased. The flood phenomenon is the one of the hazardous disaster that dictates mortal and economic losses on many people every year all around the world within rural and urban habitats. With regarding intensifies and frequencies of flood events, the global and scientific associations concerns have been increased about its consequences and damages.So far in current study an efficient and different method for analyzing this natural hazard have been developed and proposed. First, based on a literature review and consultation with experts and scholars aware of the particular circumstances of the under studied area, the required data and information were gathered from various resources such as Iran National Cartographic Center, Tehran province water and wastewater company, the satellite remotely sensed imageries and Geological Survey & Mineral Explorations of Iran (GSI) and after preparing and carrying out the required reforms, they were entered into environment of Geographic information system (GIS). The approach of this study is based on using the spatial analyst functions and tools within GIS for manipulating the data of hydrometric and rain gauges stations, the remotely sensed satellite imagery and thematic layers in the bed of Artificial Neural Network (ANN) algorithm for spatial modeling of flood in the basin of Kan river located in the west northern part of Tehran province. In the next step for achieve the goal of study, the optimal architecture of neural network has been designed and developed based on trial and error and some of the recommended relationships developed in the earlier studies. Using seven inputs layers including the thematic layers of land slope, curvature, accumulated flow, Normalized Differential Vegetation Index (NDVI), geology units, soil classes and the daily precipitation data and one target layer including recharge values of stations within under studied area along with eight and sixteen neurons dedicated to the first and second hidden layers respectively and at last one neuron for the desired output that shows the daily discharge of stations, the spatial modeling of flood incidence were performed. The designed methodology were applied for forecasting and simulating the daily recharge of hydrometric stations then based on the highest recorded accuracy and using the inter-neurons weights between the first two layers of the accomplished network that were applied for weighting overlay of thematic layers, the map of flooding potential was created. The produced map indicates the susceptibility of lands within Kan basin to flood event that generally the northern areas and lands around Kan river have the highest potential for flood hazard. The resulted relative importance for input layers from running Artificial Neural Network were indicate input layer of the daily precipitation data has the highest weight that followed by vegetation layer input for weighting overlay and assessing the flooding potential in form of the continuous surface within the studied area. The main reason of that is the more dynamic and variability of the above mentioned inputs and so their more importance in incidence of flood in comparison with the other input layers. The parameters of R2, RMSE and MAE were computed for the evaluation of the accomplished method efficiency and accuracy. The algorithm of artificial neural network in forecasting daily discharge with values of 0.82, 0.13 and 0.18 and maximum discharge with values of 0.84, 0.12 and 0.16 for R2, RMSE and MAE respectively reached acceptable outputs. Generally the applied ANN method in assessing the maximum discharge that has a very high correlation with flood occurrence had the more accurate results. The results of this study can be applied in two type for spatial modeling of flooding consist of discrete and continuous forms for the future environmental planning in regional scale as a possibility for improving crisis and environment hazards management. Generally the current study demonstrated that the joint application of the GIS spatial analyst functions and tools with artificial network have the high capability for predicting the potential of natural hazards occurrence like flooding.