DISCOVER PROJECT
Time Series Analysis
From Monitoring to Forecasting – River Flow Time Series Analytics
Project centered on modeling and forecasting river water levels and flow speeds using univariate time series techniques on hydrological data from the Vigicrues network. The codebase is designed as a modular, notebook-style pipeline that reads raw CSV exports, preprocesses them into clean time series, performs stationarity analysis, and generates short-term forecasts with ARIMA models.
PYTHON - PANDAS - NUMPY - MATPLOTLIB - STATSMODELS (ADF, ARIMA)
Project Overview
The objective of this project is to build a reusable and parameterizable framework to analyze and forecast river water levels and discharges for French rivers using Vigicrues data, focusing on the Nartuby and La Pique rivers. The goal is to go from raw CSV downloads to validated ARIMA-based forecasts, supported by diagnostic plots, stationarity tests, and quantitative error metrics such as RMSE.
The project ingests Vigicrues CSV files, parses timestamps and measurement columns, and restructures them into time-indexed series for water level and flow at multiple stations. Stationarity is assessed through rolling statistics and the Augmented Dickey–Fuller test, followed by log transformation and differencing to satisfy ARIMA assumptions. Separate ARIMA models are trained for each variable (level and speed for both Nartuby and La Pique), with fitted values transformed back to the original scale and compared visually and via RMSE to assess forecasting performance.
