I developed a Graph Neural Network model that uses historical rainfall data from a few locations in India to predict rainfall variability. I used ISRO’s GSMaP satellite dataset for the task at hand. The model allows us to:

Predict rainfall stability for a given year based on historical data
Plan water management for floods and droughts well in advance
Identify optimal locations to build raingauge stations with minimal cost

In the above picture, we can see the variation of rainfall across mainland India. By identifying the various zones of variances, models can be trained better to yield better results across the country for minimal training.