Model Paths

My new illuminem article proposes a "community first" paradigm for defending against wildfires spreading into urban areas. While fire-hardening existing homes and buildings should be done to the extent possible, it is a bottom-up approach that will always be limited by cost and scalability. A more cost-effective and scalable solution might be a top-down approach, where the community itself is the unit that must survive wildfires. Buffer zones that are located strategically at the peripheries of communities could move us closer to preserving whole communities as human ecosystems.

Over the next 2 weeks, our forecast indicates that wildfire susceptibility (potential) intensifies in west Texas, Arizona and New Mexico, and extends north to Colorado, Utah, Nevada and inland parts of central California. Looks like a typical March so far with fire potential increasing and expanding northward as we get into April. FireCasterAI: https://lnkd.in/exseazqB

This WWA study confirms our general understanding of the climate fingerprint in the LA wildfires: an extended wet period (which produced excessive vegetation) followed by an extended dry period + stronger Santa Ana winds during the actual fires. If the ignition of at least one of the fires was likely caused by sparking power lines in heavy winds getting up to a 100 mph (https://lnkd.in/gjPFh5Mc), that too might be partially climate driven. Thinking about future wildfire risk, just the most likely projected climate in any future year might not capture the entire wildfire risk for that year because it depends on multi-year climate sequences as demonstrated in the LA fires.

Screenshots from FireCasterAI show how wildfire potential is likely to change over the next two weeks for the western US. Wildfire potential decreases for the LA area given some precipitation over the next couple of days and fairly high relative humidity for the next 10 days. In comparison, the ML model predicts increasing potential for larger wildfires in West Texas 1-2 weeks out due to dry conditions and high winds. FireCasterAI: https://lnkd.in/g5uRfgUr Map legend: darker color = potential for larger fire if ignited

Making individual homes resilient to fire might not be enough. If entire neighborhoods have been wiped out, it is hard to see the value of a few scattered homes that are still intact in a toxic wasteland. To all the smart people working on solutions to the wildfire crisis: We should be thinking about saving entire neighborhoods, not just individual homes. We need to focus on the wildland-urban interfaces and build sufficient buffer zones to dramatically improve our chances of keeping wildfires out of neighborhoods, towns and cities -- at a fraction of the cost of upgrading every building in an area.

What started the LA fires? I've seen some news reports suggesting that lightning has been mostly ruled out. The map below from FireCasterAI supports that hypothesis: lightning-induced fire risk is forecasted to be mostly low in that area for this time of the year. If this proves to be true and the fires were basically human caused (which could be anything from power lines and machinery to recreational activities and arson), then those ignitions were fundamentally preventable. While we need to work on fuel reduction and more fire-resilient structures, we should be doing a lot more to simply stop the human-caused fires from starting. FireCasterAI: https://lnkd.in/gPAGQXgD Map legend: red = high risk; orange = medium; yellow = low

A simple explanation of how climate change played a critical role in the LA wildfires: Unusually large amounts of winter rainfall in the last couple of years led to heavy vegetation growth which (without thinning) then turned into dry fuel in the subsequent drought-like conditions. Add powerful offshore winds -- the Santa Ana winds are now overlapping with the extended fire season -- and all it needed was some ignition events. The ignition cause is still under investigation as of today. Part of the response (call it climate resilience) *must* be strategic thinning and creating buffer areas in the wildland urban interface.

Out of more than 22000 wildfires in the first 8 months of 2024 in the western US, only 364 (1.7%) grew larger than 1000 acres. Imagine being able to identify the few fires out of the dozens reported every day that have the potential to grow big enough to be a threat. Imagine being able to allocate additional resources to prevent them from becoming large fires. Imagine preemptively limiting fire risk through fuel reduction and other measures in specific areas that are likely to support large fires. Imagine getting local residents in and around these areas engaged to prevent human-caused ignitions. This is what a modern wildfire mitigation strategy might look like. This is what machine learning models can help us do by providing actionable wildfire risk predictions on both short and long timescales and small spatial scales. And this is what FireCasterAI (https://lnkd.in/gmjEhajE) is designed to do. Check out the article below for details on how FireCasterAI's ML prediction models performed on the 2024 wildfires (for a quick summary, just go to the end and read the takeaways). #WildFire #WildfireRisk #WildfirePrediction #FireWeather #FireCasting #FireCast #DeepLearning

Kumar Venkat?explores the effectiveness of ML models in predicting wildfire sizes and causes, contributing to better resource allocation and risk mitigation. He highlights the innovative improvements in the FireCasterAI platform, including enhanced deep learning pipelines, synthetic data generation, and rigorous testing on real-world data, exemplifying the transformative potential of AI in addressing critical environmental challenges. https://lnkd.in/eJhHBH2h

I've posted frequently on our landscape-scale wildfire risk predictions using various maps generated by FireVision. We also have a recently updated dashboard for running the machine learning model in FireVision to explore wildfire characteristics at a single location. Results include both long-term projections through 2040 (example in the screenshot below) and short-term forecasts for the next two weeks. Check it out: https://lnkd.in/g8sQfp7r #WildFire #WildfireRisk #WildfirePrediction #FireWeather #FireCasting #FireCast #DeepLearning