Project 04

River flow forecasting

The 2013 Calgary floods caused over $6 billion in damage and displaced 100,000+ residents. This project builds a multi-model forecasting tool using 9.5M+ five-minute Bow River observations for early-warning systems that protect lives and infrastructure.

XGBoost ARIMA Time series Forecasting 9.5M readings

0.89 R-squared (XGBoost)

Interactive app

Streamlit dashboard pages

Current flow

Real-time Bow River flow visualization with historical context and rolling averages

Forecast

Multi-day ahead flow predictions with confidence intervals and flood-risk thresholds

Model comparison

Side-by-side evaluation of ARIMA, Random Forest, and XGBoost forecast accuracy

Alerts

Configurable flow thresholds with automated alerts when readings exceed safe levels

Performance

Key results

0.89

R-squared

XGBoost regressor

~10

MAE (m3/s)

Best model performance

9.5M+

Observations

Five-minute intervals

Approach

Methodology

Fetched 9.5M+ five-minute river level and flow observations via the Socrata API. Resampled to daily means and engineered rolling averages (7-day, 30-day) and lag features. Trained ARIMA/SARIMA for classical time-series modeling with confidence intervals. Trained Random Forest and XGBoost regressors on engineered lag and calendar features for comparison.

01 Fetch 9.5M river observations

02 Resample to daily means

03 Engineer rolling and lag features

04 Train ARIMA/SARIMA baseline

05 Train XGBoost and Random Forest

06 Deploy Streamlit dashboard

Resources