Evaluation of river bio-resilience with artificial intelligence models(case study: Aliabad river)

Resilience is a criterion for measuring the strength and stability of a system and its ability to absorb changes and distribute them, while at the same time maintain relationships between system variables. The ability of societies to live and develop with dynamic environments is known as ecological resilience. When the accessibility of a vital resource (such as water) varies between overabundance and extreme scarcity due to natural or man-made phenomena, management should be flexible and the authorities adapted with maintaining legitimacy. So, the prediction of ecological resilience in water resources such as rivers is an important issue that needs to be considered for better management of land-use systems and water supplies. For this purpose physical and chemical parameters in rivers should be monitored to predict and understand the behavior, flexibility and interaction between living-beings of rivers. The ecological importance and reorganization features of algae, as well as being directly influenced by chemical and physical parameters make them eligible to be considered as indicators of nutrient pollution and to be the endpoints for numeric nutrient criteria developed for water quality management aims. But, most environmental models do not address water quality in relation to river biology over time and offer little prediction for future periods. Time series modeling and forecasting have importance in various applied studies, such as resilience in which resistance to chronic stress and time series are to be significant. In spite of various studies on intelligent modeling in the field of water management, no study has yet investigated the environmental resilience of the river in Iran, using the time series and artificial intelligence models. The resilience indicators examined for rivers have included four criteria: the biology, impact of pollutants, climate change, and time. The mentioned criteria were investigated in accordance with the following factors: Diatom algae, chemical parameters, discharge variations, and 10-year time series. The input data for modeling relations in the river ecosystem for Diatom were based on the factors influencing the physical and chemical parameters of these algae (EPA2017), and also on the basis of statistical methods of their correlation coefficients. Resilience index in the current study was determined based on Diatoms population with regard to Diatomic-Trophic index. In this respect, this study proposes a gene expression programming (GEP), hybrid wavelet-gene expression programming (WGEP), support vector machine (SVM) and wavelet support vector machine (WSVM) for prediction of monthly variations in Lavarak Station’s water quality that affect bio indicators. The 10-years (2002-2012) monthly data used in this study were measured from Aliabad River located in Tehran, Iran. First, the measured discharge (Q) and other quality parameters that affect the bio indicators data sets were initially decomposed into several sub-series, using discrete wavelet transform (DWT). Then, this new sub-series was imposed on the (GEP) and (SVM) models as input patterns to predict monthly bio indicator one month ahead. The results of the new proposed WGEP and WSVM models were compared with SVM and GEP models. The performance of this model was evaluated using Nash-Sutcliffe efficiency (NS), root mean square error (RMSE), and mean absolute error (MAE). A comparison between the four models showed the superiority of the hybrid models over the classic models. The achieved results even pointed to the superiority of a single SVM model over the GEP model. With regard to the studies conducted to determine the bio - resiliency index, the abundance of Diatom algae in the river within the standard of resilience the WSVM hybrid model was better while the WGEP was the second best. But due to the modeling process and the results, the WGEP model was used to determine the formula and the effect of each parameter was defined in the scenario. This model can also be effective in expressing changes in one or more independent parameter. The results of this study indicated that considering the capacity and the ability of AI models to deal with the nonlinear nature and dynamics of hydrological processes, the ability of wavelet analysis to extract certain periods of a time series is potentially more to gain reasonable prediction in different environmental processes and planning for them.