Artificial Neural Networks for Risk Decision Support in Natural Hazards: A Case Study of Assessing the Probability of House Survival from Bushfires

Risk decision-making in natural hazards encompasses a plethora of environmental, socio-economic and management-related factors, and benefits greatly from exploring possible patterns and relations among these multivariate factors. Artificial neural networks, capable of general pattern classifications, are potentially well suited for risk decision support in natural hazards. This paper reports an example that assesses the risk patterns or probabilities of house survival from bushfires using artificial neural networks, with a simulation data set based on the empirical study by Wilson and Ferguson (Predicting the probability of house survival during bushfires, Journal of Environmental Management 23 (1986) 259–270). The aim of this study was to re-model and predict the relationship between risk patterns of house survival and a series of independent variables. Various configurations for input and output variables were tested using neural networks. An approach for converting linguistic terms into crisp numbers was used to incorporate linguistic variables into the quantitative neural network analysis. After a series of tests, results show that neural networks are capable of predicting risk patterns under all tested configurations of input and output variables, with a great deal of flexibility. Risk-based mathematical functions, be they linear or non-linear, can be re-modelled using neural networks. Finally, the paper concludes that the artificial neural networks serve as a promising risk decision support tool in natural hazards.     

Forecasting of cyanobacterial density in Torrão reservoir using artificial neural networks

The ability of general regression neural networks (GRNN) to forecast the density of cyanobacteria in the Torr?o reservoir (Tamega river, Portugal), in a period of 15 days, based on three years of collected physical and chemical data, was assessed. Several models were developed and 176 were selected based on their correlation values for the verification series. A time lag of 11 was used, equivalent to one sample (periods of 15 days in the summer and 30 days in the winter). Several combinations of the series were used. Input and output data collected from three depths of the reservoir were applied (surface, euphotic zone limit and bottom). The model that presented a higher average correlation value presented the correlations 0.991; 0.843; 0.978 for training, verification and test series. This model had the three series independent in time: first test series, then verification series and, finally, training series. Only six input variables were considered significant to the performance of this model: ammonia, phosphates, dissolved oxygen, water temperature, pH and water evaporation, physical and chemical parameters referring to the three depths of the reservoir. These variables are common to the next four best models produced and, although these included other input variables, their performance was not better than the selected best model.     

ANN modelling of sediment concentration in the dynamic glacial environment of Gangotri in Himalaya

The present study explores for the first time the possibility of modelling sediment concentration with artificial neural networks (ANNs) at Gangotri, the source of Bhagirathi River in the Himalaya. Discharge, rainfall and temperature have been considered as the main controlling factors of variations in sediment concentration in the dynamic glacial environment of Gangotri. Fourteen feed forward neural networks with error back propagation algorithm have been created, trained and tested for prediction of sediment concentration. Seven models (T1-T7) have been trained and tested in the non-updating mode whereas remaining seven models (T1a-T7a) have been trained in the updating mode. The non-updating mode refers to the scenario where antecedent time (previous time step) values are not used as input to the model. In case of the updating mode, antecedent time values are used as network inputs. The inputs applied in the models are either the variables mentioned above as individual factors (single input networks) or a combination of them (multi-input networks). The suitability of employing antecedent time-step values as network inputs has hence been checked by comparative analysis of model performance in the two modes. The simple feed forward network has been improvised with a series parallel non-linear autoregressive with exogenous input (NARX) architecture wherein true values of sediment concentration have been fed as input during training. In the glacial scenario of Gangotri, maximum sediment movement takes place during the melt period (May–October). Hence, daily data of discharge, rainfall, temperature and sediment concentration for five consecutive melt periods (May–October, 2000–2004) have been used for modelling. High Coefficient of determination values [0.77–0.88] have been obtained between observed and ANN-predicted values of sediment concentration. The study has brought out relationships between variables that are not reflected in normal statistical analysis. A strong rainfall: sediment concentration and temperature: sediment concentration relationship is shown by the models which are not reflected in statistical correlation. It has also been observed that usage of antecedent time-step values as network inputs does not necessarily lead to improvement in model performance.     

Modeling of nitrate concentration in groundwater using artificial intelligence approach—a case study of Gaza coastal aquifer

Nitrate concentration in groundwater is influenced by complex and interrelated variables, leading to great difficulty during the modeling process. The objectives of this study are (1) to evaluate the performance of two artificial intelligence (AI) techniques, namely artificial neural networks and support vector machine, in modeling groundwater nitrate concentration using scant input data, as well as (2) to assess the effect of data clustering as a pre-modeling technique on the developed models' performance. The AI models were developed using data from 22 municipal wells of the Gaza coastal aquifer in Palestine from 2000 to 2010. Results indicated high simulation performance, with the correlation coefficient and the mean average percentage error of the best model reaching 0.996 and 7 %, respectively. The variables that strongly influenced groundwater nitrate concentration were previous nitrate concentration, groundwater recharge, and on-ground nitrogen load of each land use land cover category in the well's vicinity. The results also demonstrated the merit of performing clustering of input data prior to the application of AI models. With their high performance and simplicity, the developed AI models can be effectively utilized to assess the effects of future management scenarios on groundwater nitrate concentration, leading to more reasonable groundwater resources management and decision-making     

Use of neural networks for monitoring surface water quality changes in a neotropical urban stream

This paper reports the using of neural networks for water quality analysis in a tropical urban stream before (2002) and after sewerage building and the completion of point-source control-based sanitation program (2003). Mathematical modeling divided water quality data in two categories: (a) input of some in situ water quality variables (temperature, pH, O2 concentration, O2 saturation and electrical conductivity) and (b) water chemical composition (N-NO2(-); N-NO3(-); N-NH4(+) Total-N; P-PO4(3-); K+; Ca2+; Mg+2; Cu2+; Zn2+ and Fe+3) as the output from tested models. Stream water data come from fortnightly sampling in five points along the Ipanema stream (Southeast Brazil, Minas Gerais state) plus two points downstream and upstream Ipanema discharge into Doce River. Once the best models are consistent with variables behavior we suggest that neural networking shows potential as a methodology to enhance guidelines for urban streams restoration, conservation and management.     

Forecasting of daily total atmospheric ozone in Isfahan

A neural network combined to an artificial neural network model is used to forecast daily total atmospheric ozone over Isfahan city in Iran. In this work, in order to forecast the total column ozone over Isfahan, we have examined several neural networks algorithms with different meteorological predictors based on the ozone-meteorological relationships with previous day's ozone value. The meteorological predictors consist of temperatures (dry and dew point) and geopotential heights at standard levels of 100, 50, 30, 20 and 10 hPa with their wind speed and direction. These data together with previous day total ozone forms the input matrix of the neural model that is based on the back propagation algorithm (BPA) structure. The output matrix is the daily total atmospheric ozone. The model was build based on daily data from 1997 to 2004 obtained from Isfahan ozonometric station data. After modeling these data we used 3 year (from 2001 to 2003) of daily total ozone for testing the accuracy of model. In this experiment, with the final neural network, the total ozone are fairly well predicted, with an Agreement Index 76%.     

Application Of Artificial Neural Network Models To Analyse The Relationships Between Gammarus pulex L. (Crustacea, Amphipoda) And River Characteristics

This study aimed at analysing the relationship between river characteristics and abundance of Gammarus pulex. To this end, four methods which can identify the relative contribution and/or the contribution profile of the input variables in neural networks describing the habitat preferences of this species were compared: (i) the ‘PaD‘ (‘Partial Derivatives‘) method consists of a calculation of the partial derivatives of the output in relation to the input variables; (ii) the ‘Weights‘method is a computation using the connection weights of the backpropagation Artificial Neural Networks; (iii) the ‘Perturb‘method analyses the effect of a perturbation of the input variables on the output variable; (iv) the ‘Profile‘ method is a successive variation of one input variable while the others are kept constant at a fixed set of values. The dataset consisted of 179 samples, collected over a three-year period in the Zwalm river basin in Flanders, Belgium. Twenty-four environmental variables as well as the log-transformed abundance of Gammarus pulex were used in this study. The different contribution methods gave similar results concerning the order of importance of the input variables. Moreover, the stability of the methods was confirmed by gradually removing variables. Only in a limited number of cases a shift in the relative importance of the remaining input variables could be observed. Nevertheless, differences in sensitivity and stability of the methods were detected, probably as a result of the different calculation procedures. In this respect, the ‘PaD‘method made a more severe discrimination between minor and major contributing environmental variables in comparison to the ‘Weights‘, ‘Profile‘ and ‘Perturb‘ methods. From an ecological point of view, the input variables ‘Ammonium‘ and to a smaller extent ‘COD‘, were selected by these methods as dominant river characteristics for the prediction of the abundance of Gammarus pulex in this study area.     

Application of ANN and ANFIS models for reconstructing missing flow data

Hydrological yearbooks, especially in developing countries, are full of gaps in flow data series. Filling missing records is needed to make feasibility studies, potential assessment, and real-time decision making. In this research project, it was tried to predict the missing data of gauging stations using data from neighboring sites and a relevant architecture of artificial neural networks (ANN) as well as adaptive neuro-fuzzy inference system (ANFIS). To be able to evaluate the results produced by these new techniques, two traditionally used methods including the normal ratio method and the correlation method were also employed. According to the results, although in some cases all four methods presented acceptable predictions, the ANFIS technique presented a superior ability to predict missing flow data especially in arid land stations with variable and heterogeneous data. Comparing the results, ANN was also found as an efficient method to predict the missing data in comparison to the traditional approaches.     

Estimation of soil specific surface area using some mechanical properties of soil by artificial neural networks

Soil specific surface area (SSA) is an important property of soil. Depending on the measurement techniques, determination of the SSA is costly and time consuming. Hence, a limited number of studies have been conducted to predict the SSA from the soil variables. In this study, the soil samples were taken from the literature. Fractal parameters (FP) were calculated by the model of Bird et al. (European Journal of Soil Science 51, 55–63, 2000) used as the input variables to predict the SSA. Some studies have been carried out on the prediction capability of the different parameters using the artificial neural networks (ANNs). The ANNs were further used and 20 models were developed to investigate the value of input variables to predict the SSA. The results showed that the PTF13 (RMSE?=?0.13) and PTF18 (RMSE?=?0.13) with the input variables of particle-size distribution and Atterberg limits revealed better performance than the other PTFs (in the training step). It is because of the fact that free swelling index (FSI) and Atterberg limits were closely correlated to the soil clay mineralogy as one of the important factors controlling the SSA. In general, this results demonstrated that the PTF9 with the variables of sand, clay, plastic limit (PL), liquid limit (LL), and FSI showed the best (RMSE?=?0.37) results in the estimation of the SSA. In conclusion, there was not a strong correlation between the soil mechanical properties and SSA but also ANNs were a suitable method to predict the SSA from the soil variables.     

Using GA-Ridge regression to select hydro-geological parameters influencing groundwater pollution vulnerability

For groundwater conservation and management, it is important to accurately assess groundwater pollution vulnerability. This study proposed an integrated model using ridge regression and a genetic algorithm (GA) to effectively select the major hydro-geological parameters influencing groundwater pollution vulnerability in an aquifer. The GA-Ridge regression method determined that depth to water, net recharge, topography, and the impact of vadose zone media were the hydro-geological parameters that influenced trichloroethene pollution vulnerability in a Korean aquifer. When using these selected hydro-geological parameters, the accuracy was improved for various statistical nonlinear and artificial intelligence (AI) techniques, such as multinomial logistic regression, decision trees, artificial neural networks, and case-based reasoning. These results provide a proof of concept that the GA-Ridge regression is effective at determining influential hydro-geological parameters for the pollution vulnerability of an aquifer, and in turn, improves the AI performance in assessing groundwater pollution vulnerability.     

downscaling procedures as a tool for integration of multiple air issues

In assessing the risks associated with climate change,downscaling has proven useful in linking surfacechanges, at scales relevant to decision making, tolarge-scale atmospheric circulation derived from GCMoutput. Stochastic downscaling is related to synopticclimatology, weather-typing approaches (classifyingcirculation patterns) such as the Lamb Weather Typesdeveloped for the United Kingdom (UK), the EuropeanGrosswetterlagen (Bardossy and Plate, 1992) and thePerfect Prognosis (Perfect Prog) method from numericalweather prediction. The large-scale atmosphericcirculation is linked with site-specific observationsof atmospheric variables, such as precipitation, windspeed or temperature, within a specified region. Classifying each day by circulation patterns isachieved by clustering algorithms, fuzzy rule bases,neural nets or decision trees. The linkages areextended to GCM output to account for climate change. Stochastic models are developed from the probabilitydistributions for extreme events. Objective analysiscan be used to interpolate values of these models toother locations. The concepts and some applicationsare reviewed to provide a basis for extending thedownscaling approach to assessing the integrated riskof the six air issues: climate change, UV-B radiation,acid rain, transport of hazardous air pollutants, smogand suspended particulates.     

Selecting Variables for Habitat Suitability of <Emphasis Type="Italic">Asellus</Emphasis> (Crustacea,Isopoda) by Applying Input Variable Contribution Methods to Artificial Neural Network Models

This study aimed to compare different methods to analyse the contribution of individual river characteristics to predict the abundance of Asellus (Crustacea, Isopoda). Six methods which provide the relative contribution and/or the contribution profile of the input variables of artificial neural network models were therefore compared: (1) the ‘partial derivatives’ method; (2) the ‘weights’ method; (3) the ‘perturb’ method; (4) the ‘profile’ method; (5) the ‘classical stepwise’ method; (6) the ‘improved stepwise’ method. Consequently, the key variables which affect the habitat preferences of Asellus could be identified. To evaluate the performance of the artificial neural network model, the model predictions were compared with the results of a multiple linear regression analysis. The dataset consisted of 179 samples, collected over a 3-year period in the Zwalm catchment in Flanders, Belgium. Twenty-four environmental variables as well as the log-transformed abundance of Asellus were used in this study. The different contribution methods seemed to give similar results concerning the order of importance of the input variables. Nevertheless, their diverse computation led to differences in sensitivity and stability of the methods and the derived outcomes on the habitat preferences. From an ecological point of view, the environmental variables describing the stream type (width, depth, stream order and distance to mouth) were the most significant variables for Asellus in the Zwalm catchment during the period 2000–2002 for all applied methods. Indirectly, one can conclude that the water quality is not the limiting factor for the survival of Asellus in the Zwalm catchment.     

Forest cover change and fragmentation using Landsat data in Maçka State Forest Enterprise in Turkey

In this study, Grey model (GM) and artificial neural network (ANN) were employed to predict suspended solids (SSeff) and chemical oxygen demand (CODeff) in the effluent from a wastewater treatment plant in industrial park of Taiwan. When constructing model or predicting, the influent quality or online monitoring parameters were adopted as the input variables. ANN was also adopted for comparison. The results indicated that the minimum MAPEs of 16.13 and 9.85% for SSeff and CODeff could be achieved using GMs when online monitoring parameters were taken as the input variables. Although a good fitness could be achieved using ANN, they required a large quantity of data. Contrarily, GM only required a small amount of data (at least four data) and the prediction results were even better than those of ANN. Therefore, GM could be applied successfully in predicting effluent when the information was not sufficient. The results also indicated that these simple online monitoring parameters could be applied on prediction of effluent quality well.     

Establishment of turbidity forecasting model and early-warning system for source water turbidity management using back-propagation artificial neural network algorithm and probability analysis

The purpose of this study is to establish a turbidity forecasting model as well as an early-warning system for turbidity management using rainfall records as the input variables. The Taipei Water Source Domain was employed as the study area, and ANOVA analysis showed that the accumulative rainfall records of 1-day Ping-lin, 2-day Ping-lin, 2-day Fei-tsui, 2-day Shi-san-gu, 2-day Tai-pin and 2-day Tong-hou were the six most significant parameters for downstream turbidity development. The artificial neural network model was developed and proven capable of predicting the turbidity concentration in the investigated catchment downstream area. The observed and model-calculated turbidity data were applied to developing the turbidity early-warning system. Using a previously determined turbidity as the threshold, the rainfall criterion, above which the downstream turbidity would possibly exceed this respective threshold turbidity, for the investigated rain gauge stations was determined. An exemplary illustration demonstrated the effectiveness of the proposed turbidity early-warning system as a precautionary alarm of possible significant increase of downstream turbidity. This study is the first report of the establishment of the turbidity early-warning system. Hopefully, this system can be applied to source water turbidity forecasting during storm events and provide a useful reference for subsequent adjustment of drinking water treatment operation.     

基于集合经验模态分解和支持向量机的溶解氧预测

应用集合经验模态分解(EEMD)和支持向量机(SVM)相结合的方法,建立一种天然水体溶解氧浓度预测模型。首先,利用EEMD方法将溶解氧时序分解成不同频段的分量,以降低序列的非平稳性;然后,根据各序列分量的自身特征建立合适的SVM预测模型,此过程通过相关分析确定各分量输入量;最后,将各子分量预测值合成得到最终的预测结果。使用该模型对嘉陵江北温泉段的溶解氧浓度进行预测,结果表明,与传统单一的SVM和BP神经网络模型相比,该模型能有效提高预测精密度,具有良好的应用前景。     

Monitoring diel dissolved oxygen dynamics through integrating wavelet denoising and temporal neural networks

Diel dissolved oxygen (DO) time series measured continuously using proximal sensors in situ for a temperate lake were denoised using discrete wavelet transform (DWT) with the orthogonal wavelet families of coiflet, daubechies, and symmlet with order of 10. Diel DO time series denoised were modeled using nine temporal artificial neural networks (ANNs) as a function of water level, water temperature, electrical conductivity, pH, day of year, and hour. Our results showed that time-lag recurrent network (TLRN) using denoised data emulated diel DO dynamics better than the best-performing TLRN using the original data, time-delay neural network (TDNN), and recurrent network (RNN). Daubechies basis dealt with diel DO data slightly better than the other bases given its coefficient of determination (r ²?=?87.1 %), while symmlet performed slightly better than the other bases in terms of root mean square error (RMSE?=?1.2 ppm) and mean absolute error (MAE?=?0.9 ppm).     

Land surface temperature estimating in urbanized landscapes using artificial neural networks

Scenario-based land surface temperature (LST) modeling is a powerful tool for adopting proper urban land use planning policies. In this study, using greater Isfahan as a case study, the artificial neural network (ANN) algorithm was utilized to explore the non-linear relationships between urban LST and green cover spatial patterns derived from Landsat 8 OLI imagery. The model was calibrated using two sets of variables: Normalized Difference Built Index (NDBI) and Normalized Difference Vegetation Index (NDVI). Furthermore, Compact Development Scenario (CDS) and Green Development Scenario (GDS) were defined. The results showed that GDS is more successful in mitigating urban LST (mean LST?=?40.93) compared to CDS (mean LST?=?44.88). In addition, urban LST retrieved from the CDS was more accurate in terms of ANOVA significance (sig?=?0.043) than the GDS (sig?=?0.010). The findings of this study suggest that developing green spaces is a key strategy to combat against the risk of LST concerns in urban areas.     

Traffic Parameters Estimation to Predict Road Side Pollutant Concentrations using Neural Networks

The analysis aims to evaluate which is the most important among traffic parameters (flows, queues length, occupancy degree, and travel time) to forecast CO and C₆H₆ concentrations. The study area was identified by Notarbartolo Road and bounded by Libertà Street and Sciuti Street in the urban area of Palermo in Southern Italy. In this area, various loop detectors and one pollution-monitoring site were located. Traffic data related to the pollution-monitoring site immediately near the road link were estimated by Simulation of Urban MObility (SUMO) traffic microsimulator software using as input the flows measured by loop detectors on other links of road network. Traffic and weather data were used as input variables to predict pollutant concentrations by using neural networks. Finally, after a sensitivity analysis, it was showed that queues length were the mostly correlated traffic parameters to pollutant concentrations. An erratum to this article can be found at     

Modelling geosmin concentrations in three sources of raw water in Quebec, Canada

The presence of off-flavour compounds such as geosmin, often found in raw water, significantly reduces the organoleptic quality of distributed water and diverts the consumer from its use. To adapt water treatment processes to eliminate these compounds, it is necessary to be able to identify them quickly. Routine analysis could be considered a solution, but it is expensive and delays associated with obtaining the results of analysis are often important, thereby constituting a serious disadvantage. The development of decision-making tools such as predictive models seems to be an economic and feasible solution to counterbalance the limitations of analytical methods. Among these tools, multi-linear regression and principal component regression are easy to implement. However, due to certain disadvantages inherent in these methods (multicollinearity or non-linearity of the processes), the use of emergent models involving artificial neurons networks such as multi-layer perceptron could prove to be an interesting alternative. In a previous paper (Parinet et al., Water Res 44: 5847-5856, 2010), the possible parameters that affect the variability of taste and odour compounds were investigated using principal component analysis. In the present study, we expand the research by comparing the performance of three tools using different modelling scenarios (multi-linear regression, principal component regression and multi-layer perceptron) to model geosmin in drinking water sources using 38 microbiological and physicochemical parameters. Three very different sources of water, in terms of quality, were selected for the study. These sources supply drinking water to the Québec City area (Canada) and its vicinity, and were monitored three times per month over a 1-year period. Seven different modelling methods were tested for predicting geosmin in these sources. The comparison of the seven different models showed that simple models based on multi-linear regression provide sufficient predictive capacity with performance levels comparable to those obtained with artificial neural networks. The multi-linear regression model (R ²?=?0.657, <0.001) used only four variables (phaeophytin, sum of green algae, chlorophyll-a and potential Redox) in comparison with ten variables (potassium, heterotrophic bacteria, organic nitrogen, total nitrogen, phaeophytin, total organic carbon, sum of green algae, potential Redox, UV absorbance at 254 nm and atypical bacteria) for the best model obtained with artificial neural networks (R ²?=?0.843).     

Formation and modeling of disinfection by-products in drinking water of six cities in China

Water quality parameters including TOC, UV(254), pH, chlorine dosage, bromide concentration and disinfection by-products were measured in water samples from 41 water treatment plants of six selected cities in China. Chloroform, bromodichloromethane, dibromochloromethane, dichloroacetic acid and trichloroacetic acid were the major disinfection by-products in the drinking water of China. Bromoform and dibromoacetic acid were also detected in many water samples. Higher concentrations of trihalomethanes and haloacetic acids were measured in summer compared to winter. The geographical variations in DBPs showed that TTHM levels were higher in Zhengzhou and Tianjin than other selected cities. And the HAA5 levels were highest in Changsha and Tianjin. The modeling procedure that predicts disinfection by-products formation was studied and developed using artificial neural networks. The performance of the artificial neural networks model was excellent (r > 0.84).