Google is growing its flood prediction AI for India to cover more than 11,000 square kilometers along the Ganga and Brahmaputra rivers, the company announced today. Approximately 20% of global flood fatalities occur in India.
Monsoon rains have been above average in India for three consecutive weeks, Reuters reported Wednesday.
Since Flood Forecast initiative trials in the Patna region first began about a year ago, 800,000 notifications have been sent to smartphone users. Notifications are also sent to a human network of volunteers with the nonprofit SEEDS who spread emergency warnings to people without phones.
The expansion will be supported by new forecast methodologies — like a recently developed approach to creating Digital Elevation Models (DEMs) — that optimize the inundation model to work with tensor processing units (TPU) and supply predictions 85 times faster than with CPUs alone.
“Correlating and aligning the images in large batches, we adjust their camera models (and simultaneously solve for a coarse terrain) to make the images mutually consistent. Then we create a depth map for each image. To make the elevation map, we optimally fuse the depth maps together at each location,” Google senior software engineer Sella Nevo said in a blog post. “For additional efficiency improvements, we’re also looking at using machine learning to replace some of the physics-based algorithms, extending data-driven discretization to two-dimensional hydraulic models, so we can support even larger grids and cover even more people.”
In addition to using TPUs to improve the accuracy of flood predictions, Google began to draw on imagery from the European Space Agency Sentinel-1 satellite constellation.
In January, Google said flood predictions had achieved 75% accuracy. Model prediction levels have remained the same, a company spokesperson told VentureBeat in an email.
Flood prediction systems could become increasingly important as climate change worsens. AI models are being built to protect people and property from all kinds of natural disasters, with researchers this week publishing a paper on machine learning to predict large wildfires.