Forecasting Development of Mine Pit Lake Water Surface Levels Based on Time Series Analysis and Neural Networks

Sustainable mine closure is one of the main priorities of the mining industry. This aim of this research was to predict the spatiotemporal development of water levels in a mined-out pit by generating forecasts of the dependent variables (rainfall and temperature) via linear (autoregressive integrated moving average) and non-linear (artificial neural network) models. We investigated natural water level development in one mined-out pit of the closed lignite mines in Amynteon, north Greece, with no artificial recharge. The forecasted rate of water level increase was estimated to be ≈ 10 m per year in the ‘early’ stage of pit lake spatiotemporal evolution (first 10 years), and 0.1 m per year in the ‘last’ stage of potential lake development (after year 2060). Also, the optimum lake surface (i.e. the level where no significant increase in water level rate appears) was estimated at + 520 m, which was predicted to occur in ≈ 40 years. The proposed methodology was validated via water level measurements performed during the first year of lake development, where field measurements of water elevations closely followed predictions. Forecasting pit lake water levels is essential for strategic planning, examining pit lake repurposing options, and informing decisions about post-mining futures and economic transitions.