Construction of a computer model of watershed hydrology may be driven by the need to solve a particular problem or the scientific pursuit of model building, which may or may not be used for problem solving. Frequently, the type of a model to be built is dictated by the availability of data. Different models have different data needs. In general, distributed models require much more data than do lumped models. In most cases, needed data either do not exist or are not available in full. That is why regionalization and synthetic techniques are useful. Even if the needed data are available, problems remain with regard to incompleteness, inaccuracy, and in-homogeneity of data. Then, of course, storage, handling, retrieval, analysis, and manipulation of data have to be dealt with. If the volume of data required is large, data processing can be quite a sophisticated undertaken. A watershed hydrology model is an assemblage of component models corresponding to different components of hydrologic cycle. The time-scale of model output (e.g. streamflow) greatly influences the type of the model or details to be included in the model. For example, a monthly watershed model is quite different in its architecture and construction from, say, an hourly model. Thus, the elements of simulation differ with the model type. It remains an unresolved question as to hydrologic laws operating at different times-scales for different components of the hydrologic cycle. Solution of this question will greatly facilitate model construction and more clearly define the data needs.
Among various soft computing methods, Artificial NeuralNetworks (ANNs) are promising tools based on their ability in the modelling of nonlinear processes . Essentially, ANN is a massively parallel-distributed information processing system that has certain performance characteristics resembling biological neuralnetworks of the human brain . It has a flexible mathematical structure that is capable of mapping complex nonlinear relationships between input and output data sets and deriving general trends without describing physical relationships. However, it should not be considered as a mere blackbox . ANNs are able to provide a mapping from one multivariate space to another, given a set of data representing that mapping. Even if the data is noisy and contaminated with errors, ANNs have been known to identify the underlying rule . Thus, ANN has a strong input-output structure and well suited for hydrological models in terms of system estimation and prediction.
the water resources management and water quality problem. In this article, artificial neuralnetworks (ANN), M5tree (M5T) approaches and statistical approaches such as Multiple Linear Regression (MLR), Sediment Rating Curves (SRC) are used for estimation daily suspended sediment concentration from daily temperature of water and streamflow in river. These daily datas were measured at Iowa station in US. These prediction aproaches are compared to each other according to three statistical criteria, namely, mean square errors (MSE), mean absolute relative error (MAE) and correlation coefficient (R). When the results are compared ANN approach have better forecasts suspended sediment than the other estimation methods.
Dracup, 1991; Basson and van Rooyen, 2001). Therefore, considering that the proposed model is not a forecasting approach, the performance of the NN-based model must not be evaluated by using the classical procedure of splitting all available data into training and validation (and/or) test sets. The statistics were calculated over the monthly and annual series, and then verification and validation processes were done. As Stedinger and Taylor (1982) report, these are two important stages of the development and use of a stochastic streamflow model. Verification consists of demonstrating that the statistics explicitly involved in the model formulation are statistically the same for both generated and historical flows; validation of a streamflow model is the demonstration that such a model is capable of reproducing statistics which are not explicitly included in its formulation. In this study, verification statistics (means, standard deviations, lag-1 correlations and lag-2 correlations) were included in addition to some relevant validation statistics related to droughts and storage of the series. The computed statistics were grouped into four different categories:
Support Vector Machines (SVMs) are a relatively new form of machine learning that was developed by Vapnik [ 27 ]. The term SVM is used to refer to both classification and regression methods as well as the terms Support Vector Classification (SVC) and Support Vector Regression (SVR), which refer to the problems of classification and regression, respectively [ 28 ]. There are several studies where SVRs were used in hydrological forecasting. Khan and Coulibaly [ 29 ] found that an SVR model was more e ﬀective at predicting 3– 12 month lake water levels than ANN models. Rajasekaran et al. [ 30 ] used SVR successfully for storm surge predictions, and Kisi and Cimen [ 31 , 32 ] used SVR to estimate daily evaporation and daily streamflow, respectively. Finally, SVR have been successfully used to predict hourly streamflow by Asefa et al. [ 33 ] and were shown to perform better than ANN and ARIMA models for monthly streamflow prediction by Wang et al. [ 34 ] and Maity et al. [ 35 ], respectively. Yuan and Tan [ 36 ] used SVRs as a screening tool to test for drought resistance of rice. However, to date SVRs have not been applied to forecast a given drought index.
precipitation depths is extremely valuable for the identifi- cation of the streamflow evolution: in the period immedi- ately following the forecast, a rising limb, for example, will keep increasing or will reach the peak and begin to decrease depending if the rainfall is continuing or if it has already stopped. Jain and Srinivasulu (2006) used both rainfall and flow values for decomposing the flow hydrograph and then forecasting one-step ahead daily streamflow with a multi- network approach: the decomposition was performed with methods based on physical concepts and with a small SOM network, which classified the flows in low, medium and high ranges. Corzo and Solomatine (2007) applied a modular ar- chitecture based on the distinction of baseflow and excess flow obtained with i) a K-means clustering algorithm, ii) a semi-empirical constant slope method or iii) filtering algo- rithms of the hydrographs (where i) and ii) are again based on past flows only).
Abstract. We present an experiment on fifty multilayer per- ceptrons trained for streamflowforecasting on three water- sheds using bootstrapped input series. This type of neural network is common in hydrology and using multiple train- ing repetitions (ensembling) is a popular practice: the infor- mation issued by the ensemble is then aggregated and con- sidered to be the final output. Some authors proposed that the ensemble could serve the calculation of confidence in- tervals around the ensemble mean. In the following, we are interested in the reliability of confidence intervals ob- tained in such fashion and in tracking the evolution of the ensemble of neuralnetworks during the training process. For each iteration of this process, the mean of the ensem- ble is computed along with various confidence intervals. The performance of the ensemble mean is evaluated based on the mean absolute error. Since the ensemble of neural net- works resemble an ensemble streamflow forecast, we also use ensemble-specific quality assessment tools such as the Continuous Ranked Probability Score to quantify the fore- casting performance of the ensemble formed by the neuralnetworks repetitions. We show that while the performance of the single predictor formed by the ensemble mean improves throughout the training process, the reliability of the associ- ated confidence intervals starts to decrease shortly after the initiation of this process. While there is no moment during the training where the reliability of the confidence intervals is perfect, we show that it is best after approximately 5 to 10 it- erations, depending on the basin. We also show that the Con- tinuous Ranked Probability Score and the logarithmic score do not evolve in the same fashion during the training, due to a particularity of the logarithmic score.
Additionally, the authors in (Chaouachi, A.; Kamel, R.M.; Ichikawa, R.; Hayashi, H.; Nagasaka, K., 2009) studied the applicability of artificial neuralnetworks for 1 day ahead solar power generation forecasting. Different types of networks were tested and a neural network ensemble is more precise than conventional networks (multi-layered perceptron, radial basis function, recurrent network), albeit all models demonstrate acceptable forecasting accuracy. Likewise, in (Yona, Senjyu, Saber, & Funabashi, 2007) a comparative study between different artificial neuralnetworks models was conducted to predict insolation one day ahead, in which the Recurrent neural network outperforms the Feedforward neural network. Furthermore, the authors in (Paoli, Voyant, Muselli, & Nivet, 2010) presented a MLP neural network prediction approach to determine the global radiation at a daily horizon. They assumed an ad hoc time series preprocessing that reduces the error forecasts of about 5% compared to classical predictors. Additionally, in (Mantzari & Mantzaris, 2013) the researchers implemented a MLP neural network for half hour cloudiness forecasting and considered it an important tool for the estimation of cloudiness affecting solar radiation.
ANNs have been commonly implemented for predicting stock prices (White (1988); Kamijo and Tanigawa (1990); Khan (2011)), while there has been little effort on forecasting volatil- ity through neuralnetworks. Moreover, neuralnetworks have been mostly employed in combination with GARCH models (Hajizadeh et al. (2012); Maciel et al. (2016)). For in- stance, Donaldson and Kamstra (1997) investigated the usefulness of a semi-nonparametric GARCH model to capture nonlinear relationships, proving that the ANN model performs better than all competing models. Hu and Tsoukalas (1999), instead, combined the forecasts from four conditional volatility models within a neuralnetworks architecture, showing that the ANNs predict accurately well the targeted variable during crisis periods. Recently, Arneri´c et al. (2014) based their neuralnetworks on the squared innovations deriving from a GARCH model. They relied on a Jordan neural network (JNN) and showed that a NN model provides superior forecasting accuracy in comparison with other linear and nonlin- ear models.
Abstract: Over the last few recent years, there has been much research directed at predicting the future and making better decisions. This research has led to many developments in forecasting methods. Most of these methodological advances have been based on statistical techniques. Statistical methods and neuralnetworks are commonly used for time series prediction. Empirical results have shown that NeuralNetworks outperform linear regression specially in the case of more complex behaviour of dependent variables like nonlinear, dynamic and chaotic behaviours. Neuralnetworks are reliable for modeling nonlinear, dynamic market predictions. Neural Network makes very few assumptions as opposed to normality assumptions commonly found in statistical methods. Neural network can perform prediction after learning the underlying relationship between the input variables and outputs. From a statistician’s point of view, neuralnetworks are analogous to nonparametric, nonlinear regression models.
181 by availability of wide-coverage, high temporal satellite data. NDVI time series data has been employed to predict a NDVI variable beyond the time span. Univariate autoregressive integrated moving average (ARIMA) models are widely used for a univariate time series forecasting, also for the NDVI time series . However, these models are parametric and are based on the assumption that the time series been forecasted are linear and stationary and have a limited ability to capture nonlinearity and nonstationary in time series. The difficulty of forecasting arises from the inherent non-linearity and non-stationarity in the NDVI index. Many previous studies propose that non-linear machine learning approaches such as neural network models perform better than traditional time series linear models with minimum initial assumptions and high forecasting accuracy. Therefore, neuralnetworks are used as an alternative to traditional statistical forecasting methods.
Literature review shows that the artificial neuralnetworks (ANN) models have been used in different fields such as psychology (Levine, 2006; Quek and Moskowitz, 2007), medicine (Gueli et al, 2005; Lisboa and Taktak, 2006), mathematics (Hernandez and Salinas, 2004), engineering (Pierre et al, 2001; Dharia and Adeli, 2003), finance (Wong and Selvi, 1998; Kumar and Walia, 2006; Angelini et al, 2008), and economics. As the literature shows, the applications of the ANN models in economics include sectors such as water resources (Hamed et al, 2004; Singh et al, 2009; Turan and Yurdusev, 2009; Firat et al, 2010), tourism (Palmer et al, 2006) and energy sector (Islam et al, 1995; Nizami and Al-Garni, 1995; Nasr et al, 2003; Murat and Ceylan, 2006; Sozen and Arcaklioglu, 2007; Azadeh et al, 2007 and 2008; Fadare, 2009; Geem and Roper, 2009; Sozen, 2009; Ekonomou, 2010; Geem, 2011; Kanka et al, 2011; Limanond et al, 2011).
Prediction of daily need product is an important issue in collaborative planning & replenishment. Nowadays artificial neuralnetworks (ANNs) have been popularly applied to supply demand chain problems such as prediction of collaborative supply chain planning, collaborative demand planning. An ANN model is a computer model whose architec ture essentially mimics the learning capability of the human brain. The processing elements of an ANN resemble the biological structure of neurons and the internal operation of a human brain. Many simple interconnected linear or nonlinear computational elements are operating in parallel processing at multiple layers. In some applications it has been specified that ANNs have limitations for learning the data patterns. They may perform inconsistently and unpredictable because of the complex daily need data used. Sometimes data is so voluminous that learning patterns may not work. Continuous and large volume of data needs to be checked for redundancy and the data size should be decreased for the algorithm to work in a shorter time and give more generalized solutions.
 Reinoso, N. (2017). Forecasting Harmful Algal Blooms for Western Lake Erie using Data Driven Machine Learning Techniques. (Electronic Thesis or Dissertation). Retrieved from https://etd.ohiolink.edu/  Segre, D., Vitkup, D., & Church, G. M. (2002).
This paper surveys recent literature on last two decades in the domain of ANN which used to forecast foreign exchange. This paper classifies the literatures based on forecasting model, input data type, forecasting intervals, and evaluation method. This paper aims to provide a comprehensive overview of recent research in foreign exchange forecasting and to identify possible opportunities for future research works.
uses numerically predicted weather data will not take into account the effect of cloud cover and cloud formation when initializing, therefore sky imaging and satellite data methods used to predict the PV power output with higher accuracy. The article also outlined some key factors affecting the accuracy of prediction, such as forecast horizon, forecasting interval width, system size and PV panels mounting method (fixed or tracking). The aim of the work published in  was to study the effect of forecast horizon on the accuracy of the method used to predict the PV power production, which was Support Vector Regression (SVR) using numerically predicted weather data. Two forecast horizons studied: up to 2 and 25 hours ahead. As expected, the forecasting of up to 2 hours ahead was more accurate with RMSE and MAE increased 13% and 17%, respectively, when the forecast horizon was up to 25 hours ahead.
Artificial NeuralNetworks (ANN): NN are computing systems or technique that mimics the studying methods of the mind to find out the family members among the variables of a machine. They method enter information records to examine and achieve know-how for forecasting or classifying patterns and so forth. type of paintings. ANN consists of range of simple processing factors called neurons. All data processing is achieved inside this neuron best. A network of related artificial neurons may be designed, and a getting to know algorithm may be carried out to educate it . indicators (input records) are exceeded between neurons over connection links and every connection link has an related weight, which in a neural community, multiplies the signal transmitted. The weights constitute facts being utilized by the network to solve a trouble. Then the weighted sum is operated upon through an activation feature (normally nonlinear), and output information are conveyed to other neurons. The weights are continuously altered at the same time as training to improve accuracy and generalize skills  .
During the 2010 budget speech Finance Minister Pravin Gordhan suggested that the current inflation targeting framework needs to be reviewed - especially with a view of improving inflation forecasts. Forecasting is a key activity for economic policy makers McNelis and McAdam (2005), and as such, the simple recourse has been to rely on simple forecasting models given the complexity of underlying policy targets. Inflation is a major concern for a country’s central bank, but forecasting inflation is not a trivial task for any economy espe- cially in a world influenced by globalisation (Hu et al., 2007). Building a forecasting model that performs adequately well has always been a challenge for many applied econometricians and statisticians. Theory-based models (such as the expectations-augmented Phillips Curve model and the traditional Monetarist model) are often used to describe the inflation process. Econometric forecasting models differ not only in how they are specified but in the quality of inflation forecast data that they produce (Pretorius and Janse van Rensburg, 1996). Although traditional statistical time series methods perform well in forecasting, many have inherent limitations due to the manner in which they are constructed. Time series models are often linear and thus may not capture nonlinear behaviour. (Hill et al., 1996).
In this work, combining three ANNs to construct real time Flood Forecasting System. The Architecture of the system is shown in following figure 1. In this forecasting system, data are gather from stations and collected to server. Good quality data Gathered from the Gamma test are used for training. In the training, ,the training data is fed into to the input layer. The data is propagated to the hidden layer and then to the output layer. Read the input features, label output and define layers of network. After that, we have apply Back propagation neural network and Elman neural network applying for the calculating delta in each layer of the network.
Reservoir operation concerns taking decisions on water re- lease from a reservoir based on the amount of water available vis-à-vis the demand placed on the system. The available water is the sum of starting period storage and the inflow expected during the period. Consequently, effective reser- voir operation relies on reliable forecast of the inflow into the reservoir. Traditional forecasting methods using hydro- logic, hydraulic and time-series models require specification of the functional relationship of the model which can be problematic (Zhang et al., 1998), which is why focus has re- cently shifted to the use of data-driven techniques that do not require knowledge of this functional relationship. In par- ticular, artificial neuralnetworks (ANN) have been widely used to forecast reservoir inflows (see e.g. Edossa and Ba-