Input selection and performance optimization of ANN-based streamflow forecasts in the drought-prone Murray Darling Basin region using IIS and MODWT algorithm

By R. Prasad, Ravinesh C. Deo, Yan Li, Tek Maraseni. 2017

157 citations recorded by the source corpus.

Forecasting streamflow is vital for strategically planning, utilizing and redistributing water resources. In this paper, a wavelet-hybrid artificial neural network (ANN) model integrated with iterative input selection (IIS) algorithm (IIS-W-ANN) is evaluated for its statistical preciseness in forecasting monthly streamflow, and it is then benchmarked against M5 Tree model. To develop hybrid IIS-W-ANN model, a global predictor matrix is constructed for three local hydrological sites (Richmond, Gwydir, and Darling River) in Australia's agricultural (Murray-Darling) Basin. Model inputs comprised of statistically significant lagged combination of streamflow water level, are supplemented by meteorological data (i.e., precipitation, maximum and minimum temperature, mean solar radiation, vapor pressure and evaporation) as the potential model inputs. To establish robust forecasting models, iterative input selection (IIS) algorithm is applied to screen the best data from the predictor matrix and is integrated with the non-decimated maximum overlap discrete wavelet transform (MODWT) applied on the IIS-selected variables. This resolved the frequencies contained in predictor data while constructing a wavelet-hybrid (i.e., IIS-W-ANN and IIS-W-M5 Tree) model. Forecasting ability of IIS-W-ANN is evaluated via correlation coefficient (r), Willmott's Index (WI), Nash–Sutcliffe Efficiency (ENS), root-mean-square-error (RMSE), and mean absolute error (MAE), including the percentage RMSE and MAE. While ANN models are seen to outperform M5 Tree executed for all hydrological sites, the IIS variable selector was efficient in determining the appropriate predictors, as stipulated by the better performance of the IIS coupled (ANN and M5 Tree) models relative to the models without IIS. When IIS-coupled models are integrated with MODWT, the wavelet-hybrid IIS-W-ANN and IIS-W-M5 Tree are seen to attain significantly accurate performance relative to their standalone counterparts. Importantly, IIS-W-ANN model accuracy outweighs IIS-ANN, as evidenced by a larger r and WI (by 7.5% and 3.8%, respectively) and a lower RMSE (by 21.3%). In comparison to the IIS-W-M5 Tree model, IIS-W-ANN model yielded larger values of WI = 0.936–0.979 and ENS = 0.770–0.920. Correspondingly, the errors (RMSE and MAE) ranged from 0.162–0.487 m and 0.139–0.390 m, respectively, with relative errors, RRMSE = (15.65–21.00) % and MAPE = (14.79–20.78) %. Distinct geographic signature is evident where the most and least accurately forecasted streamflow data is attained for the Gwydir and Darling River, respectively. Conclusively, this study advocates the efficacy of iterative input selection, allowing the proper screening of model predictors, and subsequently, its integration with MODWT resulting in enhanced performance of the models applied in streamflow forecasting.

Subfields: Electrical and Electronic Engineering, Environmental Engineering, Water Science and Technology

View Paper: https://www.semanticscholar.org/paper/cb64dae6223c2b350d839906bdf2a2217c5ba589

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Electrical and Electronic Engineering

In Environmental Engineering

In Water Science and Technology

Citations in the corpus

Implementation of a hybrid MLP-FFA model for water level prediction of Lake Egirdir, Turkey (2018, methodology): “The validation of non-decimated, maximum overlap discrete wavelet transform algorithm (i.e., moDWT) integrated with artificial neural networks (Prasad et al. 2017) and extreme learning machines compared with an artificial neural network model (Deo and Şahin 2016) have also been demonstrated for…”
Design and evaluation of SVR, MARS and M5Tree models for 1, 2 and 3-day lead time forecasting of river flow data in a semiarid mountainous catchment (2018, methodology): “…methods based on multi-resolution analysis of input dataset (e.g., wavelet transformation and, ensemble empirical mode decomposition) (Prasad et al. 2017), the prediction accuracy of river flow discharge over daily and monthly time scales were found to be better than the standalone…”