With peak wildfire season underway in California, Scott Stenfel, PG&E’s senior director of meteorology and wildfire science, held a virtual wildfire season outlook briefing to provide an advanced seasonal outlook for PG&E’s service area.
A virtual event podcast from the Oakland PG&E office on June 26 highlighted weather patterns, fuel conditions and other factors that will impact wildfire risk in the coming months.
Strenfel said this is the driest time of year for fire fuel moisture until mid-October, when rain typically begins. “While Northern California will experience Santa Ana wind events until the rain arrives, machine learning and modeling can help determine when a Public Safety Power Shutoff (PSPS) should be implemented,” Strenfel explained. “To reduce the risk of our equipment starting to ignite, PG&E is taking a snapshot of our outlook so far this season.”
Year-to-date, residents have experienced 80,000 acres of fire in the PG&E service area, according to Cal Fire. This is slightly below 2025, but 20,000 acres more than the five-year average. Wildfire risk continues until typical winter rains begin. “We are entering a summer of heightened wildfire risk, and we need to get serious about wildfire mitigation and what we can all do together as community members to reduce wildfire risk,” Strenfel said. “Our layers of protection start with equipment replenishment, undergrounding, or removing lines with microgrids, which is why there is an ongoing effort to update our networks and make them more resilient to all weather events.”
Meanwhile, PG&E is taking operational mitigation measures, with its meteorology and data science teams working with the company to run a machine learning Environmental Potential Index model. It implements the use and development of computer systems that can learn and adapt without following explicit instructions by using algorithms and statistical models to analyze patterns in data and make inferences. Strenfel also noted that the application of machine learning to biological databases is increasing, and there is significant room for improvement in machine learning applications.
PG&E uses the model to review its own operations to prevent fires. In some cases, Strenfel noted, the PG&E grid becomes more sensitive to instances of damage. “This is where our power line safety settings serve as an important tool to reduce the chance of fire, and this is the area that the San Ramon Institute is monitoring. They are testing enhanced power lighting safety settings.”
“Eventually, if the fire likelihood index reaches a very high point where it’s very dangerous and we have a Diablo-style event, that’s where we’ll be running PSPS,” Stenfel said. However, there have been cases where trees have gotten stuck on power lines, or foreign objects (balloons) have flown onto power lines. If that happens, PG&E will initiate other efforts to prevent the fire, such as putting lines underground and managing vegetation.
Events that could lead to PSPS. Strong winds, dry conditions, and low humidity can lead to a high fire index. When there is a high probability of an outage leading to a fire, PG&E has AI mechanical models in both the fire probability space and the outage space. Companies can also overlay outage spaces to determine where in their PG and E footprint conditions require PSPS. “We’re doing this in predictive mode because we want to communicate to our customers through PSPS that there is a possibility of an outage,” Strenfel said.
In addition to computer forecasting models, PG&E also relies on observations from weather stations already in place around the state, such as the National Weather Service, and remote weather stations, supplemented by 1,600 PG&E weather stations. These companies’ stations send a report every 10 minutes through an app when switched on. However, these stations can make observations every 30 seconds, allowing them to monitor the situation in real time and make potentially important decisions quickly.
Here, live weather maps are deployed internally at PG&E and developed for situational awareness and decision support. “As one of the lead meteorologists making difficult decisions regarding the PSPS, I need to know where all of our assets are, along with their context. Combined with high-resolution modeling, we can make the best decisions about whether to move forward with operational weather mitigation,” Strenfel said.
Sometimes, when the company is on the brink of doing a PSPS, Strenfel will tune in to a particular weather station doing a 10-minute microcast observation, then switch to a station that launches a 30-second microcast to see really detailed observations before making a PSPS decision.
As an example, Strenfel monitored weather conditions in Altamont Pass at 11 a.m. on June 26 and found that the likelihood of a catastrophic wildfire was low. That morning’s particular weather system brought cold air and rain showers to San Ramon that morning, limiting the potential for fires.
Another valuable tool is the PSPS Dashboard. This is a learning model that can monitor all deployed PG&E weather stations and make predictions for the next 129 hours. “The dashboard has a geographic view as well as a tabular view that shows what regions and what the risk matrix is,” Strenfel said. “Thus, this tool will allow PG&E to decide whether to undertake PSPS, which will be in the best interest of PG&E’s customers,” he added.
So the big question is: What will the 2026 fire season look like? According to the National Interagency Fire Center, an above-normal severe wildfire risk is predicted for most of PG&E’s service area except Southern California. The reason is that rising ocean temperatures off the coast of Baja California can lead to monsoon rains.
There’s an El Niño weather system just off its coast, but that doesn’t necessarily mean extra precipitation, Strenfel noted. According to machine learning models, dry Diablo winds will prevail throughout the summer and early fall. That means catastrophic wildfires are unlikely to occur this season.
