<p>Marine waters’ physicochemical and biological parameters, such as temperature, salinity, oxygen content, and nutrient levels, play a critical role in the balance and functioning of marine ecosystems. Accurate prediction of these parameters provides valuable insights for sustainable marine resource management and assessing climate change’s impacts on ocean productivity. In this study, we evaluated the performance of a Long Short-Term Memory (LSTM) neural network model using original input variables and a second model using input variables preprocessed through Discrete Wavelet Transform (DWT). The latter approach—the DWT-LSTM model—aims to assess the influence of decomposed input variables on prediction performance. Ten key oceanic parameters were predicted in Al Hoceima Bay (Morocco, Central Rif), using daily-resolution data (2000–2021) from the Copernicus Marine Platform. The models were trained on data from 2000 to 2020 and tested in 2021. Results show that the DWT-LSTM model achieves higher predictive accuracy, capturing both short-term fluctuations and long-term trends more effectively. These results enhance our understanding of the marine dynamics in Al Hoceima Bay and support improved forecasting of environmental responses to climate variability.</p>