Mohammad Reza Yazdani; Ali Asghar zolfaghari
Abstract
Watershed outflow has influenced by different factors such as climatic, human and physical aspects and this Variability of effective factors can cause complex conditions, difficulty of flow forecasting and it mainly originates by different local and temporal scales of these factors. Also, some remote ...
Read More
Watershed outflow has influenced by different factors such as climatic, human and physical aspects and this Variability of effective factors can cause complex conditions, difficulty of flow forecasting and it mainly originates by different local and temporal scales of these factors. Also, some remote meteorological signals can cause changes in meteorological conditions in different regions. Hablehrud river flow has a vital role in regional development, especially for agricultural section. Thus research of river flow forecasting should be done for water resources management especially when there are drought and climate change conditions in order to facilitate sustainable development. In this study four nonlinear models of artificial neural networks including Generalized Feed Foreward Networks (JFNNs), Jordan/Elman Networks(JENs), Time Lag Recurrent Networks(TLRNs) and Radius Basis Function Networks(RBF) was used to modeling Hablehrud river flow(Bonkuh station) during 1982 to 2011. Input variables after sensitivity analysis were used in 4 models and 4 scenarios. Ten teleconnection indexes were used as input of the model to evaluate their roles in model capability. Results indicated that in the test stage Jordan/Elman Networks represented lower error compared with selected models (RMSE for 4 scenarios are5.57, 4.9, 5.35 and 4.62 respectively). In general error showed decreasing trend from first scenario to the last. Error was decreased of 15 to 31 percent by using teleconnection patterns as inputs (GFFN=%26, JEN=%15.8, TLRN=%25.5 and RBF=%31.7). Totally using teleconnection indexes as inputs in the modeling stage can diminish error of flow forecasting, although selected models indicated different results due to its variable topologies.
Ommolbanin Bazrafshan; Ali Salajegheh; Ahmad Fatehi; Abolghasem Mahdavi; Javad Bazrafshan; Somayeh Hejabi
Abstract
Drought is random and nonlinear phenomenon and using linear stochastic models, nonlinear artificial neural network and hybrid models is advantaged for drought forecasting. This paper presents the performances of autoregressive integrated moving average (ARIMA), Direct multi-step neural network (DMSNN), ...
Read More
Drought is random and nonlinear phenomenon and using linear stochastic models, nonlinear artificial neural network and hybrid models is advantaged for drought forecasting. This paper presents the performances of autoregressive integrated moving average (ARIMA), Direct multi-step neural network (DMSNN), Recursive multi-step neural network (RMSNN), Hybrid stochastic neural network of directive approach (HSNNDM) and Hybrid stochastic neural network of recursive approach(HSNNRM) with time scale monthly and seasonally for hydrology drought forecasting and SDI selected as predictor in the Karkheh river basin. The results shown performances of HNNDA was found to forecast hydrological drought with greater accuracy for SDI forecasting, so performances model in monthly scale was greater accuracy to seasonality scale.