Forecasting water quality parameters using artificial neural network for irrigation purposes

Abstract

This study was aimed at analyzing the water quality of Ele River Nnewi, Anambra State for irrigation purposes with a view to predicting a one-year water quality index using Artificial Neural Network (ANN). Water pollution has posed a major problem and identifying the points of pollution in the River system is a very difficult task. To overcome this task, the need to determine the pollution level arose by modeling and predicting four water quality parameters at four (4) different locations using the Artificial Neural Network. These parameters include the pH, Total Dissolved Solids (TDS), Electrical Conductivity (EC), and Sodium (Na), respectively. The water quality results showed that the pH values which ranges from 6.01 to 6.87 were within the FAO standard in all the points for both rainy and dry seasons, whereas the TDS (mg/l), EC (dS/m) and Na (mg/l) parametric values range from 2001 to 2506, 3.01 to 5.76, and 40.42 to 73.45 respectively, were above the FAO standard from point 1 to point 3 and falls within the FAO standard at point 4 with values ranging from 1003 to 1994, 2.01 to 2.78 and 31.24 to 39.44, respectively. However, during the dry season, the TDS, EC, and Na values range from 2002 to 2742, 3.04 to 5.82 and 40.14 to 88.45 respectively, were all above the FAO standard. Generally, the artificial neural network modeled the actual water quality data set very well with good prediction. The training model performance evaluation shows that the R2 values ranges from 0.981 to 0.990, 0.981 to 0.988, 0.981 to 0.989 and 0981 to 0.989, for pH, TDS, EC, and Na. The testing model performance shows that the R2 value ranges from 0.952 to 0.967, 0.953 to 0.970, 0.951 to 0.967and 0.953 to 0.968, for pH, TDS, EC and Na while the forecast performance evaluation shows that the R2 values ranges from 0.945 to 0.968, 0.946 to 0.968, 0.944 to 0.967 and 0.949 to 0.965 for pH, TDS, EC and Na respectively. It was also observed that the Root Mean Squared Error (RMSE) ranges from 0.022 to 0.088, 0.012 to 0.087, 0.015 to 0.085 and 0.014 to 0.084 for pH, TDS, EC and Na, respectively. Information from this study will serve as a guide to researchers on the water quality index for irrigation purposes. Also, it will guide the government and agencies on policy, management and decision-making on water resources.

Artificial neural network model with different backpropagation algorithms and meteorological data for solar radiation prediction

“…An artificial neural network, which works similar to the human nervous system 13 , 14 , consists of an input layer of neurons (or nodes, units), one or two or even three hidden layers of neurons, and a final layer of output neurons 15 – 17 . ANNs have self-learning capabilities that enable them to produce better results as more data becomes available.…”

Monitoring and Forecasting of Water Pollution by Heavy Metals

“…The chemical parameters of ground water play an important role in the classification and evaluation of water quality. It is noted that the best results can only be obtained when studying the ion complex in water, and not the concentration of individual ions [9]. Chemical classification also shows the concentration of various dominant cations, anions and their relationship.…”

“…Various analytical models are used to predict water quality, which can be divided into traditional, based on statistical models, and non-traditional, using artificial intelligence (AI) approaches. Methods of KNN, machine learning, artificial neural networks (ANN) can be attributed to non-traditional methods, and regression analysis, time series methods can be attributed to traditional methods [9]. A review of the literature has shown that artificial neural networks (ANN) are at the peak of popularity in modeling the prediction of water pollution [10], [11].…”

The formulation of irrigation and nitrogen application strategies under multi-dimensional soil fertility targets based on preference neural network

“…We select the activation function and learning rate by referring to the neural network structure commonly used in similar fields (1 hidden layer and 64 hidden nodes) 29 , 30 . It is found that ReLU has better performance than other activation functions (sigmoid, tanh).…”