Analysing and Forecasting Precipitation Trends in Vancouver

Abstract

This study analyzed trends and forecasted monthly temperature and precipitation in Vancouver using time-series modeling. A dataset spanning several decades (1941 to 2024) was preprocessed to handle missing values and included derived features, such as lagged variables for precipitation. Prophet, a robust time-series forecasting model, was used to identify seasonality and long-term trends. Additional regressors, including rain intensity and relative humidity, were incorporated to improve precipitation forecasts.

The analysis revealed a steady upward trend in temperature, with an estimated rate of increase of 0.1°C per year, consistent with global warming patterns. Model diagnostics showed that the temperature model performed well, achieving a Root Mean Squared Error (RMSE) of 0.30 and a Mean Absolute Percentage Error (MAPE) of 4.03% when regressors were included. In contrast, the precipitation model exhibited significantly higher residual variability and error metrics, underscoring the challenges of forecasting this variable. A moderate correlation (0.50) was observed between temperature and precipitation, reflecting some seasonal alignment but highlighting the complexity of their relationship.

These findings contribute to understanding local climate dynamics in Vancouver and emphasize the need for more sophisticated modeling approaches, particularly for highly variable parameters like precipitation. This study provides a foundation for refining predictive models to support climate adaptation strategies.