The goal of this research is to develop accurate and reliable forecasting models for chlorophyll ɑ concentrations in seawater at multiple depth levels in El Mar Menor (Spain). Chlorophyll ɑ can be used as a eutrophication indicator, which is especially essential in a rich yet vulnerable ecosystem like the study area. Bayesian regularized artificial neural networks and Long Short-term Memory Neural Networks (LSTMs) employing a rolling window approach were used as forecasting algorithms with a one-week prediction horizon. Two input strategies were tested: using data from the own time series or including exogenous variables among the inputs. In this second case, mutual information and the Minimum-Redundancy-Maximum-Relevance approach were utilized to select the most relevant variables. The models obtained reasonable results for the univariate input scheme with σ¯¯¯ values over 0.75 in levels between 0.5 and 2 m. The inclusion of exogenous variables increased these values to above 0.85 for the same depth levels. The models and methodologies presented in this paper can constitute a very useful tool to help predict eutrophication episodes and act as decision-making tools that allow the governmental and environmental agencies to prevent the degradation of El Mar Menor.