<p>Water demand, as a time-related phenomenon, exhibits dependency on external parameters, which implies that demand forecasting requires modelling demand time series and the impact of external features. The focus of the current study is on enhancing short-term water demand forecasting by utilizing a comprehensive combination of external environmental and sociocultural parameters, along with demand time series properties, through Dynamic Neural Networks models. The environmental and sociocultural inputs include weather, calendars, and cultural events. In this regard, Nonlinear Autoregressive with Exogenous Input, Nonlinear Autoregressive, and Nonlinear Input–Output Network Dynamic Neural Networks models are analyzed. The results show that the prediction power increases with the neural network layer size and the number of feedback and input delays, at the cost of an exponential increase in computational time. For one time step, one hour, one day, and one week ahead, the Nonlinear Autoregressive with Exogenous Inputs model demonstrates greater predictive power. The Nonlinear Autoregressive model RMSE values remain less than 0.05 for around 1000 time steps and then increase rapidly. This threshold for the Nonlinear Autoregressive with Exogenous Input and Nonlinear Input–Output Network models is around 2500 time steps. Overall, the findings indicate that the developed models enhance the accuracy of demand prediction.</p>