Illinois Mobile App Master Calendar

Heating up forecast skill: AI and microwave soundings from space

Traditional weather forecasting has continually improved at about one day per decade. These skill increases have largely come from the slow and steady increase of observations globally, as well as improvements to the underlying physical models. More recently and much more rapidly has been the emergence of artificial intelligence (AI) weather forecast models. In about 5 years, AI weather forecast models have shown the ability to achieve comparable forecast skill to traditional numerical models by learning the underpinning physics of nature but running at a fraction of the compute cost. Many of the AI weather forecasting models are still reliant on traditional data assimilation methods, and these traditional DA methods are often compute constrained leaving many observations unused. Tomorrow.io is focused on improving weather surveillance and forecast skill through an increase in density of satellite observations from low-earth orbit and AI. The current global microwave sounder fleet sponsored by government agencies often has large temporal gaps, with some places only receiving one observation daily because of the sheer cost of this flagship observational backbone. To fill the gap, Tomorrow.io is in the midst of building out a constellation of Tomorrow.io Microwave Sounders (TMS), that are cubesats based on the NASATROPICS mission design. These small, 12 channel sounders, are low cost which enables many to be built for a fraction of the cost of conventional microwave sounders. When combined with the government fleet of sensors, Tomorrow.io’s sounders will enable hourly revisits of microwave sounder data globally, vastly increasing the available data for weather forecasting. Tomorrow.io currently has seven sounders on orbit, with more scheduled for launch by the end of the year. This presentation will provide an overview of Tomorrow.io, the sounder constellation, as well as some exploration of AI-DA methods that could potentially handle this large volume of novel data.

ZOOM: https://illinois.zoom.us/j/82093903213?pwd=SC7WAE6OPDRxFHHHSI9WGEMadb894t.1

Meeting ID: 820 9390 3213     Passcode: 372504