Inspiration

Climate change is increasingly causing extreme weather events like floods and droughts, severely impacting agriculture and livelihoods in Rwanda. Witnessing these challenges, I was inspired to create a predictive framework that leverages historical weather data to forecast local climate risks. The goal is to support early warning systems and sustainable adaptation strategies to help vulnerable communities become more resilient.

What it does

This project analyzes temperature data collected from multiple meteorological stations over several decades to identify trends, seasonal patterns, and anomalies. Using advanced forecasting models such as ARIMA, Prophet, and machine learning techniques, it predicts upcoming climate risks, including temperature extremes related to floods and droughts. These predictions can inform farmers, policymakers, and disaster preparedness teams to take proactive action.

How we built it

  1. Data Cleaning

    • Replaced placeholder negatives (e.g., -1.96) with NaN.
    • Imputed missing values using rolling mean methods.
    • Standardized station data and converted DATE columns to time-series formats.
  2. Exploratory Data Analysis (EDA)

    • Identified progressive warming trends over decades.
    • Highlighted wet/dry cycles affecting agriculture.
    • Detected anomalies relative to 30-year baselines.
  3. Feature Engineering

    • Created rolling 7-day and 30-day temperature averages.
    • Computed anomalies relative to long-term mean.
    • Encoded seasonality and station identifiers.
  4. Predictive Modeling

    • Time-series models: ARIMA, SARIMA, Prophet.
    • Machine learning: Random Forest, LSTM for extreme event classification.
    • Evaluated performance with RMSE and MAE metrics; Prophet achieved best seasonal forecasts.

Challenges we ran into

1.Handling incomplete and inconsistent data with many missing or placeholder values.

  1. Capturing localized climate variations that global models fail to represent accurately. 3 Balancing model complexity and interpretability to ensure actionable outputs. ## Accomplishments that we're proud of
  2. Successfully developed a replicable forecasting pipeline tailored to Rwanda’s microclimates.
  3. Detected clear warming trends and seasonal patterns validated by historical extreme events.
  4. Achieved accurate forecasts of temperature extremes to support early warning systems.
  5. Created a scalable framework that can extend to other regions and integrate more environmental data.
  6. Delivered actionable insights aligned with global climate adaptation goals.

What we learned

  1. Data cleaning and preparation are critical for robust time-series modeling.
  2. Combining statistical models with machine learning provides complementary strengths.
  3. Localized forecasting requires careful feature engineering and domain knowledge.
  4. Clear visualizations enhance stakeholder understanding and trust in the model.
  5. Climate data science can make a tangible impact on food security and disaster resilience.

What's next for Climate Change and Environmental Sustainability

  1. Incorporate additional climate variables like rainfall, humidity, and soil moisture for multi-factor risk modeling.
  2. Expand the model coverage to neighboring districts and countries with similar climates.
  3. Develop mobile and SMS-based alert systems to deliver forecasts directly to farmers.
  4. Collaborate with policymakers to embed these insights into climate adaptation plans.
  5. Continuously update models with new data to improve accuracy and responsiveness.

Built With

Share this project:

Updates