A specific long short-term memory (LSTM) model developed for water quality prediction based on a particular water quality dataset will only apply to that dataset and may fail to make an accurate prediction on another dataset. This paper focuses on improving the tolerance of LSTM prediction models by mitigating the discrepancies in model prediction capability that arise when a model is applied to different datasets. Two predictive LSTM models are developed from two different water quality datasets and are optimised using the metaheuristic genetic algorithm (GA) to create two-hybrid GA-optimised LSTM models subsequently combined using a linear weight-based technique to develop a tolerant predictive ensemble model. The hybrid models contribute equally to the average ensemble model, while one of the hybrid models has a 10% greater weight contribution in the weighted ensemble model. The ensemble models outperform the individual hybrid models, but only marginally at times. The models can successfully predict the quality of river water in terms of dissolved oxygen concentration. When tested on unseen and unrelated datasets, the models make accurate predictions and thus are applicable in domains other than the water sector. The consistent and similar performance of the models on any dataset illustrates the successful mitigation of discrepancies in the predictive capacity of individual LSTM models by the proposed ensemble scheme. Observed model performance outlined the datasets on which the models could potentially make accurate predictions.INDEX TERMS Ensemble model, environment, genetic algorithm, long short term memory, rivers, water, water quality, water conservation, weight based model fusion.