With a global of 50 AI change makers, the team will develop a model for 12-hour rainfall forecasting to help vulnerable communities and children plan and mitigate adverse weather conditions such as drought, floods, and storms in both the short and long term, in this high-impact 8-weeks challenge.

The problem

Smallholder farmers in West Africa are very sensitive to rainfall and flooding events. Yet they do not have access to weather information, and furthermore, thunderstorm forecasts are poor, because satellites do not have the resolution to accurately predict these relatively small-scale weather features. 

The Kanda Weather Group has a working IoT product that collects upper-air data using a weather balloon at a very low cost and sends the data back to the ground receiver. We are also completing a weather app that can show a forecast in a dashboard format. We have been testing these weather balloon (also called radiosonde) launches at two universities in West Africa. 

Our most recent work has been to create an initial machine learning model using precipitation data to make a 12-hour rain forecast for the 500 square kilometer region that initiated the launch. Now, we would like to utilize a more complex ML or AI method to achieve the same outcome. 

The project goals

Develop a model that forecasts 12-hour rainfall at a skill level better than the previous model and better than climatology. The training datasets to be used are (6000+) radiosonde weather balloon launches from National Weather Stations located in the United States that are cross-referenced to corresponding rainfall data 12 hours in the future for the same location. The creation of a model of this type implies that a good 12-hour rainfall forecast can be made after sending up a single weather balloon in the morning. 

Future work includes further atmospheric modeling, including fire weather over California or air quality forecasts in polluted regions.