LSTM and ARIMA models to forecast population by age and sex within the Mexico City Metropolitan Area

Population projections and forecast are usually produced for national and regional levels. At more granular geographic scales, such as the intraurban level, population estimates are seldom produced due to insufficient time-series data and a lack of suitable methods. Yet, projecting population size and demographic structure within urban areas is essential for spatial planning. We input WorldPop data from 2000 to 2020 into a series of Long-Short Term Memory (LSTM) and Autoregressive Integrated Moving Average (ARIMA) models to forecast population by age and sex to the year 2030 in the municipalities of the Mexico City Metropolitan Area. We aim to analyse potential changes in population size and age-sex composition by 2030, and evaluate the accuracy of respective outputs from LSTM and ARIMA models based on median errors in a 2020 forecast by comparing estimated and actual values. Our results show that the total population of the Mexico City Metropolitan Area could reach between 23.8 and 26.2 million by 2030. The municipalities of Tacamác, Chalco, Nicolás A. Romero, Chimalhuacán, and Ixtapalupa will likely register the largest increases, while a population decline will likely be observed in Iztapalapa, and Ecatepec de Morelos. Increasing levels of ageing are expected in all municipalities, with significant increases of population groups aged 65+ and 40-64, particularly in Tecámac, Chalco, Nicolás Romero, Chimalhuacán, Ixtapaluca, and Benito Juárez. The number of children aged 0-14 will likely remain constant in the majority of the metropolitan area, but will decline in a number of municipalities, including Iztapalapa, and Ecatepec de Morelos. Forecasts of young adults aged 15-39 suggest population stability, small growths or declines across municipalities. Both LSTM and ARIMA models return low median percentage error forecasting 2020, 1.4% and 0.2% respectively.