It seems like everything these days is all about AI this, AI that. When it comes to weather and climate, AI and machine learning is being integrated in numerous ways with the hopes of improving forecasting to better help communities prepare.
One of the latest AI weather integrations is taking place in India. The Indian monsoon is responsible for providing three quarters of the country’s annual rainfall each year, and is essential for agriculture in the region. However, the monsoon is notoriously hard to predict, and climate change is making it even more difficult.
The Indian monsoon is driven by differences in temperature between land and sea, the amount of snow cover in the Himalayas, soil moisture, and other factors. The India Meteorological Department (IMD) has used numerical weather prediction (NWP) models for decades to try to simulate how the weather patterns will develop over time. They offer broad guidance such as if more or less rain than usual is expected a month or so ahead of time, but are not considered reliable more than about five days out.
This year, the IMD provided forecasts to their 38 million farmers that were generated by AI rather than NWP. Instead of trying to simulate the atmosphere using physics and equations, the AI model provided predictions by comparing the patterns they see in weather data with previous, similar patterns in historical weather records.
Initial findings show that the AI model was able to accurately predict when the monsoonal rains would arrive 30 days in advance. It also correctly forecasted that rainfall would stall in the middle of the season, as it did for 20 days, despite not appearing in NWP forecasts.
Not only was the AI model more accurate, but it demanded less computing power and equipment than NWP models. Poorer countries often lack both, often leading to worse predictions. The AI model may be able to help close that gap. It is noted however that there are known issues with AI weather predictions using this method. AI forecasts could become unrealistic due to their lack of constraints by the laws of physics and the fact that they can only make predictions based on data from past events, causing them to struggle with rare or extreme events.
