Lake evaporation rate and evaporation volume prediction is getting attention from the scientific community and government policy makers because of the dramatic global climate change and the key role evaporation plays in hydrological cycle management and water budgeting. Different researchers tried to predict lake evaporation through direct measurement methods like energy transfer, mass transfer and energy balance or indirect measurement methods like pan evaporation and machine learning prediction models. ANN, CNN, RNN and LSTM are among the commonly used machine learning models. However, one evaporation prediction method does not work for all lakes because evaporation is dependent on climate variables and regional topography of the lakes. Additionally, direct evaporation measurements, which are dependent on manual calibrations, and pan evaporation measurement, which is pan and pan coefficient dependent, are inefficient for such long-term evaporation predictions because lake evaporation prediction by itself is dependent on the open nature of the lake, and the time-series, dynamic and non-linear nature of the evaporation data on which these prediction methods are unfit for. Therefore, an LSTM prediction model that can manage time-series, dynamic, non-linear data for long-term prediction is necessary. Hence, we developed an evaporation rate and evaporation volume prediction model for Lake Tana using LSTM which can predict the lake’s evaporation rate and evaporation volume for 20 years. We used historic climatic and evaporation dataset collected from NASA POWER and GLEV for 34 years to train the model and we applied MSE, RMSE and MAE to evaluate model performance. We developed a performant LSTM prediction model that can effectively predict Lake Tana’s evaporation rate and evaporation volume for the next 20 years with a possible minimal loss and error values of 0.0044 MSE, 6.67% RMSE and 4.69% MAE.