Climate change impacts on the flow regime and water quality indicators using an artificial neural network (ANN): a case study in Saskatchewan, Canada

Abstract

In this study, the artificial neural network (ANN) method was applied to investigate the impacts of climate change on the water quantity and quality of the Qu'Appelle River in Saskatchewan, Canada. First, the second-generation Canadian earth system model (CanESM2) was adopted to predict future climate conditions. The Statistical DownScaling Model (SDSM) was then applied to downscale the generated data. To analyze the water quality of the river, concentrations of dissolved oxygen (DO) and total dissolved solids (TDSs) from the river were collected. Using the collected climate and hydrometric data, the ANNs were trained to simulate (i) the ratio of snowfall-to-total precipitation based on the temperature, (ii) the river flow rate based on the temperature and precipitation; and (iii) DO and TDS concentrations based on the river flow and temperature. Finally, the generated climate change data were used as inputs to the ANN model to investigate the climate change impacts on the river flow as well as DO and TDS concentrations within the selected region. Hydrologic alteration of the river was evaluated via the Range of Variability Approach (RVA) under historical and climate change scenarios. The results under climate change scenarios were compared with those under historical scenarios and indicated that climate change would lead to a heterogeneous change in precipitation and temperature patterns. These changes would have serious degrading impacts on the river discharge as well as DO and TDS concentration levels, causing deterioration in the sustainability of the river system and ecological health of the region.

Forecasting of stage-discharge in a non-perennial river using machine learning with gamma test

“…Machine learning and artificial intelligence models have become very popular in recent decade [ [34] , [35] , [36] , [37] , [38] ]. Forecasting of the stream discharge various models such as multiple-linear regression (MLR) [ 2 , [39] , [40] , [41] , [42] ], rating curve [ [43] , [44] , [45] , [46] , [47] ], wavelet-based MLR (WMLR) [ 48 , 49 ], support vector machine (SVM) [ 39 , 44 , [50] , [51] , [52] , [53] ], artificial neural network (ANN) [ 45 , [53] , [54] , [55] , [56] , [57] ], wavelet-based artificial neural network (WANN) [ 2 , 39 , 58 ], adaptive neuro-fuzzy inference system (ANFIS) [ [59] , [60] , [61] ], wavelet-based support vector machine (WSVM) [ 39 , 62 ], wavelet–bootstrap–ANN (WBANN) [ 48 , 63 ], M5-model trees [ 46 , 64 ], random forest (RF) [ 65 ], ARIMA [ 65 , 66 ], gene expression programming (GEP) [ 32 , 67 , 68 ], genetic algorithm (GA) [ 3 , 33 , 69 ], genetic programming (GP) [ 32 ], Bagged M5P [ 65 ], integrating long-short-term memory (LSTM) [ 69 , 70 ], wavelet–bootstrap–multiple linear regression (WBMLR) [ 48 ], Fuzzy logic and fuzzy neuro systems [ 59 , 71 ] multi-objective evolutionary neural network (MOENN) [ 59 ], and Gaussian process regre...…”