This paper explores the dynamics of water levels in the Aquifer Petrignano dataset through comprehensive time series analysis. By examining temporal variations and seasonal patterns, we gain insights into the underlying trends affecting water levels. We address the impact of missing values and evaluate various methods for handling them, ensuring the robustness of our analysis. Data smoothing and resampling techniques are investigated to enhance predictive accuracy. Emphasis is placed on achieving stationarity using transformations and differencing methods. We analyze the influence of cyclical features, such as the time of year, on water level predictions. Time series decomposition is utilized to understand the fundamental components of the dataset. The inclusion of lagged variables is explored to improve forecasting models. Autocorrelation analysis guides the selection of appropriate models for water level prediction. We compare the performance of univariate models, such as Prophet, ARIMA, Auto-ARIMA, and LSTM, evaluating their strengths and limitations. The study extends to multivariate models, such as Multivariate Prophet, to assess improved prediction accuracy by incorporating multiple variables. Challenges associated with applying multivariate models are addressed, and solutions are proposed to overcome these obstacles. The findings provide valuable insights into water level dynamics and offer robust methodologies for accurate forecasting.