Abstract

Prescribed burn officials rely on day-of NWS Spot Forecasts to understand the day's meteorological conditions and whether they can burn within the prescription, leaving little time for replanning or day-of operational shifts as weather conditions change. Although forecasts from mesoscale products like NOAA’s HRRR model can provide forecasts up to 48 hours in advance, hyper-local variability in coastal mountain-valley systems can introduce biases or inadequately capture near-surface conditions. Leveraging distributed, modular sensor suites developed by Stanford University's SMesh Network, in part adapted from UCAR COMET 3D-PAWS, we demonstrate a bias-correction approach to enable short-horizon hyperlocal sensor forecasts that can complement NWS Spot and support fire personnel through convergence science. We further explore machine-learning emulators trained using the NCAR CREDIT Generation 2 development framework that could enable emulated forecasts on consumer-grade hardware without connecting to external servers, even deep in the wilderness.

Here is the public livestream link. Staff members can look for a Google Calendar invitation for the talk. 

Please reach out to Sam Scalice (sscalice@ucar.edu) with any questions you may have.