Which Health Facilities Have Been Impacted by L.A.-Area Fires? AI May Paint a Clearer Picture

On the morning of January 7, a fire broke out in the Topanga State Park area of Los Angeles County. Spurred by drought and heavy wind conditions, the Palisades Fire quickly spread to hundreds and then thousands of acres, consuming thousands of structures along the way and displacing whole communities. Across the region, three other significant fires also broke out in quick succession, the largest of which, the Eaton Fire just north of Pasadena, by January 13, had burned over 14,000 acres and destroyed thousands of structures.

Meanwhile, even for those not directly in the line of fire, smoke blanketed most of the city, with over 900,000 individuals exposed to between three and five days of heavy smoke between the 7th and the 13th according to data from NOAA’s smoke forecasting system. CNN reported on January 13 that “at least one medical clinic” burned down, while medical evacuations were undertaken for hundreds of patients.

One of the most important factors for humanitarian responders in these types of large-scale disaster situations is to understand the effects on the formal health system, upon which most people — and vulnerable communities in particular — rely upon in their neighborhoods. Evaluation of the impact of disasters on individual structures, including critical infrastructure such as health facilities, is traditionally a relatively slow and manually arduous process, involving extensive ground truth visitation by teams of assessment professionals.

Speeding up this process without losing accuracy, while potentially improving the safety and efficiency of assessment teams, is among the more important analytical efforts Direct Relief can undertake for response and recovery efforts. Manual assessments can now be effectively paired with AI-based analysis of satellite imagery to do just that.

Understanding the Impact

A widespread manual assessment process is ongoing in Los Angeles under the auspices of county authorities. It appears to be moving somewhat faster than some might expect given the enormous scale and intensity of the impacts. In the Eaton Fire impact zone around Altadena, for instance, ground crews had by Monday, January 13, in just a few days of work, already evaluated 4,547 structures within and near the fire perimeter, and determined that 1,897 (41.7%) of those had been “Destroyed”, meaning that damage to the structure was 50% or greater.

Another 51 structures were determined to have “Minor” or “Major” damage, of between 10% to 50% of the structure. Ground assessment areas were also clustered around the fire perimeter, predominantly at the edge where the probability of damage and likelihood of detection was greatest. While this makes sense in terms of prioritization for the work of ground crews, the clustering also left large areas within the perimeter but further away from the edge largely unassessed. Likewise, the principal health infrastructure serving areas within the Eaton Fire perimeter was entirely unassessed by this process as of January 13, most likely due to the prioritization landscape.

With the advent of geospatial AI models trained on disaster damage impacts, ground assessment is not the only tool available to response agencies and others seeking to understand how much damage has occurred and the degree to which that damage may affect essential services for communities. The work of the Oregon State University team of experts in remote sensing-based post-disaster damage detection, led by Jamon Van Den Hoek and Corey Scher, was featured in the Financial Times on January 9.

Their modeling, based on Sentinel-1 satellite imagery, identified 21,757 structures overall, of which 11,124 were determined to have some level of damage. The Oregon State model does not distinguish between different levels of damage, and therefore cannot respond to certain types of questions that the manual inspections can respond to, but nevertheless the coverage area and the speed of detection have been much greater.

While the findings of the ground assessment and the Oregon State analysis tend to correlate at a relatively high level in terms of whether or not damage occurred, there were also 874 points in the ground assessment dataset for the Eaton Fire for which no corresponding polygon existed in the Oregon State remote sensing analysis, making its findings still somewhat more comprehensive in certain areas of the fire perimeter, and most effectively used in conjunction and comparison.

By far the largest effort undertaken to analyze the extent of building damage post-fire, particularly in the Eaton Fire area, has been conducted by Microsoft’s AI for Good team. In the update to their dataset produced on January 11th (Saturday) the Microsoft team, using satellite imagery from Maxar, had analyzed a total of 156,102 buildings across the entire area surrounding the Eaton Fire perimeter. From that group, they detected 8,682 damaged buildings, 7,026 of which were estimated to be damaged above 50% of the structure.

Which Health Facilities May Be Damaged?

Information on the location of health facilities comes from three primary sources. The first is the state of California’s OSHPD dataset on health infrastructure locations, which tracks a wide array of types of facilities ranging from acute care hospitals to skilled nursing facilities and hospices. The federal Bureau of Primary Health Care’s Uniform Data System tracks locations for all federally qualified health center sites, which treat a disproportionate number of low-income patients. HealthcareReady.org also maintains an extensive database of private pharmacy locations for their RxOpen online tool which reports on the operational status of pharmacies during disasters. While each dataset contains coordinate locations for facilities, geocoding quality relative to precise locations varies somewhat, with OSHPD being the most likely to reflect accurate structure locations.

Correlating all three of these health facility databases with three of the primary available assessments (County of LA, Oregon State, and Microsoft AI for Good) for the Eaton Fire area reveals potentially a more extensive picture of damage than has so far been reported by CNN and others. According to model-based estimates which have been partially confirmed by ground truth assessment, at least four facilities in the Eaton Fire area (Pinnacle Health, Alta Loma Hospice, Two Palms Care Center, and AltaMed Medical Group: Pasadena) have likely been or confirmed destroyed. One other facility (Pasadena Park Healthcare and Wellness Center) is damaged but likely not destroyed.

One dialysis facility (Eaton Canyon Dialysis Center) does not appear to be damaged but sits directly proximate to areas of high destruction so as to make its operational status highly questionable. At least two other private pharmacy locations (Rite Aid on East Altadena Dr. and Webster’s Pharmacy on N. Lake Ave.) do not appear to be damaged but are situated in areas where virtually every other building has been highly damaged or destroyed.

In addition to other complicating factors for service availability, such as road closures affecting access, personal impacts on staff, electrical outages affecting the ability to remain open, and the ability to re-supply facilities with essential health commodities, the combined use of AI and manual assessment teams is providing a remarkable combination of speed, scope, and accuracy of damage predictions alongside ground-truth evidence to corroborate and verify model findings. This type of emerging socio-technical system for health assessment is one of the signs of progress and hope amidst an otherwise catastrophic situation unfolding in Los Angeles.