H. Kerem Cigizoglu

Comparison of different ANN techniques in river flow prediction

Forecasts of future events are required in many of the activities associated with the planning and operation of the components of a water resource system. For the hydrologic component, there is a need for both short- and long-term forecasts of river flow events in order to optimize the system or to plan for future expansion or reduction. This paper presents the comparison of different artificial neural network (ANN) techniques in short- and long-term continuous and intermittent daily streamflow forecasting. The studies in modelling the intermittent series are quite limited because of the complexity of fitting models in to these series. The available conventional models necessitate the adjustment of numerous parameters for calibration. Three different ANN techniques, namely, feed-forward back propagation (FFBP), generalized regression neural networks, and radial basis function-based neural networks (RBF) are applied to continuous and intermittent river flow data of two Turkish rivers for short-range and long-range forecasting studies. The k-fold partitioning method is employed for preparing the ANN training data successfully. In general, the forecasting performance of RBF is found to be superior to the other two ANN techniques and a time series model in terms of the selected performance criteria. It was observed that the FFBP method had some drawbacks such as a local minima problem and negative flow generation.

Prediction of daily precipitation using wavelet-neural networks.

Abstract This study aims to predict the daily precipitation from meteorological data from Turkey using the wavelet—neural network method, which combines two methods: discrete wavelet transform (DWT) and artificial neural networks (ANN). The wavelet—ANN model provides a good fit with the observed data, in particular for zero precipitation in the summer months, and for the peaks in the testing period. The results indicate that wavelet—ANN model estimations are significantly superior to those obtained by either a conventional ANN model or a multi linear regression model. In particular, the improvement provided by the new approach in estimating the peak values had a noticeably high positive effect on the performance evaluation criteria. Inclusion of the summed sub-series in the ANN input layer brings a new perspective to the discussions related to the physics involved in the ANN structure.

Rainfall-Runoff Modelling Using Three Neural Network Methods

Three neural network methods, feed forward back propagation (FFBP), radial basis function (RBF) and generalized regression neural network (GRNN) were employed for rainfall-runoff modelling of Turkish hydrometeorologic data. It was seen that all three different ANN algorithms compared well with conventional multi linear regression (MLR) technique. It was seen that only GRNN technique did not provide negative flow estimations for some observations. The rainfall-runoff correlogram was successfully used in determination of the input layer node number.

Artificial neural network methods for the estimation of zeolite molar compositions that form from different reaction mixtures

The possibility of using artificial neural network (ANNs) methods for the estimation of the zeolite molar composition and hence the zeolite phase that may be obtained from a certain initial reaction mixture composition was investigated. Three different artificial neural network methods, namely feed forward back propagation (FFBP), radial basis function-based neural networks (RBF) and generalized regression neural networks (GRNN), were tested for this purpose. A data set obtained from the literature was used in the training of the neural networks. The results obtained for a second data set were compared to experimental findings as well as to estimations made by using multilinear and non-linear regression. It was determined that the neural networks learn quite efficiently from experimental zeolite synthesis data. The predictions made by using artificial neural network methods were, in general, more reliable than those performed by regression. The best prediction of the Si contents of the zeolites investigated were made by the GRNN and FFBP methods while the H2O content was predicted better by the RBF method. The results indicate that using artificial neural network methods may decrease significantly the number of experiments that have to be performed to discover new synthesis compositions.

Daily precipitation predictions using three different wavelet neural network algorithms by meteorological data

In this study, three different neural network algorithms (feed forward back propagation, FFBP; radial basis function; generalized regression neural network) and wavelet transformation were used for daily precipitation predictions. Different input combinations were tested for the precipitation estimation. As a result, the most appropriate neural network model was determined for each station. Also linear regression model performance is compared with the wavelet neural networks models. It was seen that the wavelet FFBP method provided the best performance evaluation criteria. The results indicate that coupling wavelet transforms with neural network can provide significant advantages for estimation process. In addition, global wavelet spectrum provides considerable information about the structure of the physical process to be modeled.

The analysis of 2004 flood on Kozdere Stream in Istanbul

The climatic and physiographical factors of the Kozdere Creek Watershed are examined to find out the causes of the flood of August 11, 2004. Actual land use has been obtained from forest management stand plans and classified satellite images. The multispectral digital satellite data set belonging to years 1992, 1993, and 2005 was used to determine the status of land use. Physiographic factors, including the slope and the aspect, have been identified from digitized ortho-photo maps in the GIS environment. Since flow records of Kozdere are not available, flow values corresponding to different return periods were obtained using regression analysis of neighboring streams. No noticeable alteration of the land use occurred between 1992 and 2005. Since the physical factors are the same as they were during the 1985 rain, the flood after the 2004 rainfall cannot be due to the physiographical factors of the upper watershed. The existing channel in the neighborhood is not enough to convey even the 10-year return period flow. Thus, the 2004 flow did not fit into this cross-section and flooded houses on the alluvial fan. The cross-section of the Kozdere Creek passing through the PTT neighborhood should be increased, and the surface roughness should be decreased by covering the channel with concrete in order to prevent floods.

The effect of the relative humidity and the specific humidity on the determination of the climate regions in Turkey

All elements of climate that affect climatic events must be taken into account such that the climate regions are determined with exactitude. To this end, data on maximum temperature (Tx), minimum temperature (Tn), mean temperature (Tm), and precipitation (Pt) as well as local pressure (Ps), mean wind (WN), relative humidity (RH), and specific humidity (SH) have been investigated statistically and graphically. The specific humidity data calculated using Tm, Ps, and RH data and statistical comparisons have shown that there are no drawbacks to using SH in climatologic studies. According to principal component analysis, it was concluded that RH and SH should be used together with Tx, Tm, Tn, and Pt for the determination of the climate regions. Two cluster analysis methods, Wards method and Kohonen neural network technique, were used to show the effect of RH and SH. A comparison of the clusters stability between the limited and high number of stations shows that Wards method and Kohonen neural network are very stable in both cases. It was also determined that RH does not change the outline of climate regions but that it affects the zones of climate transition. It was observed that clusters determined by using Tm, Pt, and RH provide relatively more distinctive clusters in the data space than clusters determined by using Tm, Pt, and SH.

Suspended sediment forecasting by artificial neural networks using hydro meteorological data

Estimates of sediment yield are required in a wide spectrum of water resources engineering problems. The non-linear nature of suspended sediment time series necessitates the utilization of non-linear methods for the forecasting study. In this study artificial neural networks, a well known non-linear method, are employed to forecast the daily total suspended sediment amount on rivers. The neural networks are trained using the rainfall data, recorded on the river catchment, the river flow and suspended sediment data belonging to Juniata Catchment in USA. The simulations provided satisfactory forecasts in terms of the selected performance criteria comparing well with conventional methods.

Forecasting of Meteorologic Data by Artificial Neural Networks

The forecasting of the meteorologic data carries significance for meteorology and water resources engineering. The forecasting of rainfall particularly is complicated because the data is intermittent, i.e. the observed time series data contains zero values as well as the positive non-zero observations. The available meteorologic models having physical basis are difficult to establish since they require tremendous amount of data for calibration. In this study artificial neural networks are employed to forecast the rainfall data using with and without periodic components. The forecasting results are evaluated in terms of mean square error (MSE) and the total rainfall for the testing period. It is shown that the forecasted series capture the general behaviour of the observed one and compare well in terms of the rainfall total.

Investigation of sea level anomalies related with NAO along the west coasts of Turkey and their consistency with sea surface temperature trends

It is well-known that North Atlantic Oscillation (NAO), which is one of the large-scaled climate modes effective in the Northern Hemisphere, has a considerable affect on the water resources and climatic indicators especially in the Mediterranean basin. In recent years, also crucial studies about the sea level rise in relation to climate change have been accelerated. Turkey has about 20 modernized tide gauge stations equipped with permanent GPS receivers and targets to contribute to global sea level rise studies in the future. The aim of this study is to find out the effects of North Atlantic Oscillation on the national shores using the data of four tide-gauge stations located on the Aegean and Mediterranean coasts of Turkey. Implications from these four tide gauges would motivate researches to take into account the effect of NAO in calculating the true sea level rise at the national coasts. While studying the sea level changes, vertical crustal movement has been observed using the data of tide gauge GPS stations, and this situation has been taken into consideration in the evaluation of sea levels. Besides, in order to investigate the influences of thermal expansion on sea levels, sea surface temperature data of the meteorology stations near the tide gauges have been evaluated. The homogeneity of the data sets was analyzed using four statistical tests. As a result, all of the meteorology stations’ temperature series and tide gauges’ data are subjected to trend detection after the homogeneity analysis. Eventually, the effects of North Atlantic Oscillation on both sea levels and sea surface temperatures have been introduced. The study results indicate high correlation between North Atlantic Oscillation and the sea level and sea surface temperature events. It is seen that the linear correlation between the sea level trends of the considered stations and the sea surface temperature data of the related meteorology stations is considerably significant.