Extreme weather events are sweeping the globe and range from heat waves, wildfires and drought to hurricanes, extreme rainfall, and flooding. These weather events have multiple impacts on agriculture, energy, and transportation, as well as low-resource communities and disaster planning in countries across the globe.
Accurate long-term forecasts of temperature and precipitation are crucial to help people prepare and adapt to these extreme weather events. Currently, purely physics-based models dominate short-term weather forecasting. But these models have a limited forecast horizon. The availability of meteorological data offers an opportunity for data scientists to improve sub-seasonal forecasts by blending physics-based forecasts with machine learning. Sub-seasonal forecasts for weather and climate conditions (lead times ranging from 15 to more than 45 days) would help communities and industries adapt to the challenges brought on by climate change. We will focus on longer-term weather forecasting to help communities adapt to extreme weather events caused by climate change.
The dataset was created in collaboration with Climate Change AI (CCAI). Participants will submit forecasts of temperature and precipitation for one year, competing against the other teams and official forecasts from NOAA.
In this 4-week project, the team will model data to predict the arithmetic mean of the maximum and minimum temperature over the next 14 days for each location and start date for longer-term weather forecasting to help communities adapt to extreme weather events caused by climate change.
Various machine learning methods can be used to make these predictions, such as Random Forests, XGBoost, and Convolutional Neural Networks (CNNs). We will explore several choices.
Data can be augmented with meteorological data such as temperature, wind speed, and vapor pressure from National Oceanic and Atmospheric Administration (NOAA).
One of the key challenges will be to choose a subset of appropriate features that impact a weather forecast’s predictions to be used in model training.
Data Pre-processing & Data Insights/Exploration
Feature Extraction and Baseline Models
Model Development and Fine tuning
Model Finalization and Deployment to Dashboard (Optional based on interest)
Final report and Presentation
Take your skills and show them off.
1. Data Pre-processing & Data Insights 2. EDA And Feature Engineering 3. Developing Models 4. Fine tuning Models 5. Deploying to a Dashboard or App