January brought a challenging start for California, as a series of devastating wildfires swept through the state, endangering lives, destroying homes, and reshaping entire communities. These fires were not only among the most destructive in California’s history but also ranked among the costliest ever recorded in the United States. The scale of destruction was unprecedented, leaving a lasting impact on both the environment and the people affected.
At the time of our investigation, estimates indicated that approximately 57,000 acres (equivalent to around 23,000 hectares) had been reduced to ashes. To put this staggering number into perspective, the wildfires consumed an area twice the size of Barcelona. This comparison highlights the severity of the catastrophe, as the flames not only ravaged vast landscapes but also wiped out properties, businesses, and the livelihoods of thousands of people. The destruction extended beyond physical structures; it represented the loss of years of effort, dedication, and personal histories embedded in those communities.
For our research, we chose to focus on the Palisades Fire, one of the most significant wildfires during this period. It was the fourth most destructive fire in California’s history, making it a crucial case to analyze. Additionally, this fire was particularly notable because it originated in an area that played a key role in the spread of the overall wildfire crisis. Given the intensity and scale of the fire, we initially assumed that the entire affected region would be completely burned, leaving little to no untouched areas. However, upon closer examination, we found that this was not the case.
We obtained a dataset containing information on both burned and unburned buildings within the fire-affected zones. As we analyzed the data by creating buffers around the buildins, we observed an unexpected pattern, certain areas within the burned region remained intact, creating gaps in the otherwise devastated landscape. This phenomenon immediately caught our attention, raising questions about what factors contributed to these unburned sections. Were they protected by specific environmental conditions, structural materials, or firefighting efforts?
Through our research, we discovered that many California residents took matters into their own hands to protect their homes from the advancing wildfires. They stayed behind, using pool water to drench their trees, lawns, and roofs in an effort to create a barrier against the flames. This widespread practice led us to think more deeply about water consumption across the city and how it might be linked to fire patterns. Could the availability and strategic use of water in certain areas influence the way wildfires spread and the extent of the damage they cause?
To explore this question, we established three key data sets to analyze: swimming pools, trees, and healthy vegetation (such as well-maintained lawns) within the Palisades area. These elements were crucial in understanding the relationship between water sources, greenery, and fire resistance. To collect and process this information efficiently, we developed two data pipelines.
The first pipeline focused on data collection through machine learning. We used Roboflow to train two specialized models capable of detecting pools and trees from satellite imagery. Once the models were trained, we converted them into ONNX format, making them compatible with QGIS, a powerful geographic information system. By integrating these models into the QGIS plug-in Deepness, we were able to analyze large-scale satellite data and accurately pinpoint the locations of pools and trees throughout the region.
The second phase of our data collection involved leveraging Google Earth Engine to extract key environmental indicators. Specifically, we focused on NDVI (Normalized Difference Vegetation Index) to assess healthy vegetation and NDWI (Normalized Difference Water Index) to detect the presence of water. These indices provided crucial insights into the distribution of greenery and water sources across the study area. After processing the data, we downloaded a GeoTIFF file, which could be seamlessly integrated into QGIS for further analysis.
Once all the necessary data layers were collected, we moved on to the analysis phase. To visualize the relationships between water, vegetation, and fire patterns, we overlaid all the data layers and converted them into heat maps. We then divided the site map into a grid system, placing data points at the intersections of the grid. By assigning values to these points based on the corresponding heat map color gradients, we were able to create a comprehensive dataset representing the spatial distribution of pools, trees, vegetation health, and water presence.
From this process, we generated a CSV file containing numerical values for all the analyzed layers. This structured dataset allowed us to move forward with advanced data visualization. Using dimensionality reduction techniques, we plotted the information into a specialized graph to uncover hidden patterns and correlations between wildfire behavior and water consumption. Through this approach, we aimed to gain a deeper understanding of how access to water and vegetation management influenced fire resistance and overall damage mitigation.
Relationship of correlation graph for data has been plotted to investigate different root causes and catalyst for this event.
For next term, we expect that by adding more layers into the analysis [non-human] (moisture, wind, topography) and [human] (materials, nearest firefighters station, water supply system) we can create a planning tool which can be used to predicts patterns with AI models in order to prevent society and governments for future fires in places around the globe that struggle with this issue.
