Strategic Environmental Assessment for Municipal Water Demand Based on Climate Change

Accurate urban water demand forecasting plays a key role in the planning and design of municipal water supply infrastructure. The reliable prediction of water demand is challenging for water companies, specifically when considering the implications of climate change (Zubaidi et al., 2018). Several studies have documented that weather variables drive water consumption in the short-term, and it enhances the accuracy of the prediction model when it is combined with socio-economic factors. However, the impact of climate change on the municipal water demand has yet to be challenged. To surmount this challenge, more research work is needed to accurately estimate the required quantity of water with increasing water demands. Recently, Artificial Neural Networks (ANNs) have been found to be an innovative approach to predict water demand. This PhD study aims to develop a novel methodology to forecast the impact of climate change on municipal water demands for a long-term time series based on the baseline period 1980-2010. It should be highlighted that, based on our knowledge, this is the first study of substantial duration, based on data collected from 1980-2010, which focuses on the associations between monthly climate change and municipal water consumption. A new approach is therefore proposed to quantifying municipal water demands through the assessment of climatic factors, using a combination of a Singular Spectrum Analysis (SSA) technique, three hybrid computational intelligence algorithms and an ANN model. These hybrid algorithms include a Lightning Search Algorithm (LSA-ANN), a Gravitational Search Algorithm (GSA-ANN) and Particle Swarm Optimisation (PSO-ANN). The SSA technique is adopted to decompose the time series of water consumption and climate variables to detect the stochastic signal for each time series. In the same context, the hybrid algorithms are used to find the best value of learning rate coefficient and the number of neurons in both hidden layers of the ANN model. Based on the performance of each hybrid algorithm, the most accurate and reliable water demand forecast model will be selected and used for estimating future water consumption. The considered environments of this study are applied in Australia and the United States from America for mitigating the uncertainty associated with the geographic location (the data of the United States of America was used to support the reliability of developing the municipal water demands prediction model). Furthermore, the Long Ashton Research Station Weather Generator (LARS-WG) model is utilised to simulate future climate factors over three periods (2011-2030, 2046-2065 and 2080-2099) based on the B1, A1B and A2 emission scenarios and seven General Circulation Models (GCMs). The future projection of these climate factors is applied directly to the impact model of water consumption to obtain the projected municipal water demand for different future periods and different greenhouse emission scenarios. The principal findings of this research are the following: from the model perspective, 1) the SSA is a powerful technique when used to remove the effect of socio-economic factors and noise, and detect the stochastic signal time series for water consumption. 2) The ANN model has better performance in term of optimising the correlation between observed and predicted water consumption when using the (LSA-ANN) algorithm. 3) The evaluation of the ANN model (using a validation data set) for Melbourne and Columbia Cities gives a correlation coefficient of 0.96 and 0.95, and the root mean square errors are 0.025 and 0.016 respectively. These findings indicate the capability of the proposed model to predict water demands with high accuracy in different continents. 4) The high performance of LARS-WG model results are found to be appropriate for the simulation of future climate variables. 5) The harmonisation between future monthly water demand (for the periods 2011-2030, 2046-2065 and 2080-2099) and stochastic signals of climate variables, relative to baseline period 1980-2010, emphasises the reliability of the present methodology. However, from the water demand perspective, the water percentage demand (WPD) are likely to rise in winter, drop in summer and fluctuate in both spring and autumn seasons for all periods and under all greenhouse emission scenarios. The results of WPD distribute between -3.5% and 3% for all periods and emission scenarios. The A2 scenario shows the highest and lowest values of WPDs compared to the A1B and B1 scenarios, in particular, in the 3rd period. The mean of seasonal WPD values shows that there is no dominant scenario as the best or the worst case of water demand over all future periods. The highest amount of seasonal demand happens in winter (A2 scenario, 3rd period), and the lowest amount of seasonal demand occurs in autumn (A1B scenario, 3rd period). In conclusion, this study facilitates the conception of the impact of climate change on municipal water demand from the baseline period 1980-2010