Complicated Complexes! Why do Wildfires get “Complexed”?¶
Background:¶
Across the United States, recent years have seen an increase in the number of wildfires being managed as “wildfire complexes”. Wildfire complexes are groups of individual fires that are all managed together by a single Incident Management Team. This can have a number of benefits, however it also creates unique challenges for data reporting on wildfires. Creating a complex can make it more difficult to track individual fires that are members of complexes, and it can create data errors that lead to inaccurate estimates of size or cost. These errors make it more difficult to correctly allocate expenses after the fire is over, and to conduct research on wildfires.
Our project focused on investigating the factors that influence the creation of complexes. We used data from our partner, Lise Ann St. Denis of Earth Lab. The data we used were refined versions of the ICS 209 situation reports filled out during wildfire incidents - Lise’s work has been to streamline these by standardizing data formats, removing errors, and other tasks needed to transform the sitreps into science-grade datasets (St. Denis et al., 2023).
Findings and Methods (project phases):¶
Our project had three phases: 1, 2, and 3 (a & b).
Phase one:¶
The first phase involved exploratory data analysis: we wanted to understand what was contained in the datasets Lise gave us, since these datasets have not been published yet and have not been thoroughly analysed. This phase consisted of creating numerous plots from the data, and using python code to create metadata. For example we used box plots and histograms to get a sense of the data’s distribution, and to see outliers. This process also helped to familiarize us with the data, and give us a sense of what the different variables meant. Following are some plots we created during the process of exploratory data analysis.
Figure 1: Qualitatively, this plot appears to show an increase over time in the number of structures damaged over time. A task for future analysis would be to look at this trend quantitatively, to determine its statistical significance.
Phase two:¶
The second phase involved identifying data errors, which we did by counting the number of duplicated ID numbers for particular fires. Since each fire is supposed to have only one ID number, we looked through the datasets to find ID numbers that were repeated. This gave us a sense of which fires might have been counted twice, due to merging with other fires, splitting into two, or being exchanged between complexes.
Note: this image is not working at the moement. It will be fixed soon!
Figure 3: Extracting duplicated identifiers to flag data errors.
Phase three (a & b):¶
The third phase had two sub-phases. In the first sub-phase we counted the number of member fires per complex, and looked at how this number changed over time. In the second phase, we looked at both the number of member fires per complex and the discovery date of all the complexes, and we related this to the change over time in National Preparedness Level.
National Preparedness Level (NPL or PL) is a number between 1 and 5 which compares the availability of firefighting resources to the number of active wildfires. A PL of 1 means that many resources are available during a time with minimal fire activity, whereas PL 5 indicates severe resource strain, as firefighting resources are stretched thin by many or large wildfires. Comparing PL to discovery date and members per complex allowed us to see if times of greater resource strain correlated with increased numbers of complexes being created, or more member fires per complex (spoiler: they did).
Figure 4: Increase in average number of member fires over time.
Figure 5: Fires are more likely to be complexed, and to contain more member fires, during times of resource strain (NPL=5). The size of the red dot in the figure indicates the number of member fires per complex.
Final Discussion and Analysis:¶
There are several main findings of our work. The first is that the ICS 209+ datasets are a treasure trove of information, and that much could be learned from continuing to study them. Our brief exploration of the data hinted at trends including increases in wildfire destructiveness and cost, increased resource strain in recent summers, and many other things. These could all be explored with continued analysis of this data.
Other main findings were that errors do still exist in the data, and that tracking individual fires within complexes continues to be a challenge. We also found that the number of member fires per complex is increasing over time. Finally, we found that complexes are generally more likely to be created during times of high resource strain, and that during times of high resource strain, they are likely to have more member fires.
These last observations were qualitative, and need to be backed up by numerical analysis in the future. Given more time, we would also like to have looked at the average PL levels over the course of a given year, and other factors driving the creation of wildfire complexes. Wildfire complexes have wide-reaching implications for wildfire research and cost allocation. Continuing to understand what factors drive complexing presents a critical question, which the ICS 209 data is ideally suited for answering.
Additional Resources & Citations:¶
- St. Denis et al., 2023, “All-Hazards Dataset Mined from the US National Incident Management System, 1999-2020”
- NWCG Memo: Data Management Committee Task Team, May 17, 2016, “Data management for Incidents Managed as a Complex and Wildfires that Merge”.
- Nguyen, D., Belval, E.J., Wei, Y. et al. Dataset of United Stated Incident Management Situation Reports from 2007 to 2021. Sci Data 11, 23 (2024). https://doi-org.colorado.idm.oclc.org/10.1038/s41597-023-02876-8
- Moench, Michael, 2025, “Wildfire Data Reporting: Errors Caused by Wildfire Complexing”. https://github.com/allenmoench/wildfire_complexing2