Summary of approach and findings
In an attempt to reproduce Butry and Thomas’ results at the national level, this analysis used NFIRS and MODIS6 data from 2017 covering the entire U.S.
The 2017 NFIRS data was geocoded, so it was possible to find matches at a much more granular level than by county. All fire incident types were included (rather than attempting to define "wildfires" with NFIRS criteria), as well as authorized and unauthorized fires (incident types 631 and 561) and prescribed fires (incident type 632).
To eliminate issues related to time zones and to be able to compare incidents within a 24-hour window, date and time of all NFIRS incidents was standardized by converting the alarm time to Universal Coordinated Time. Nationwide, we found that about 1.1% of the MODIS observations and 1.6% of the MODIS incidents occurred within 1 kilometer and 24 hours of an NFIRS incident.
The 2017 NFIRS data was geocoded using location information provided by reporting fire departments and the GEOCODE Procedure in SAS 9.4. The precision of geographical coordinates returned by the geocoder is limited by the quality of the input information and the sophistication of the geocoding algorithm.
Overall, 72% of the NFIRS incidents examined (where Incident Type was in the 100 series or was equal to 561, 631 or 632) were geomatched to an address, 11% were geomatched to the center point of a named street and 17% were geomatched to the center point of a ZIP code.
However, the geocoding precision varied by type of incident. Structure fires (those where Incident Type was between 110 and 123) were geocoded to an address 83% of the time, while fires in forests and grasses (Incident Types 141 and 143) were geocoded to an address only 51% of the time. Some incidents (1,454) lacked sufficient location information to be geocoded and most (953) of these were natural vegetation fires (Incident Types 140-143). Even where the geocoding algorithm provided a geomatched address, it is possible that the location information was incorrectly reported or interpreted.
The subsequent analysis is based on identifying NFIRS fires that occurred within 1 kilometer and 24 hours of a MODIS observation; bear in mind that this distance calculation may be based on imprecise NFIRS location information. A summary of the geocoding precision by NFIRS incident type is shown in Table 5.
Table 5: Geocoding Precision of 2017 NFIRS Data Fire Locations by Fire Incident Type
| Address | % | Street | % | ZIP | % | City | % | None | % | |
|---|---|---|---|---|---|---|---|---|---|---|
| Structure | 405,952 | 83 | 14,264 | 3 | 66,125 | 14 | 13 | 0 | 247 | 0 |
| Vehicle | 87,615 | 55 | 31,747 | 20 | 40,733 | 25 | 9 | 0 | 78 | 0 |
| Natural | 18,881 | 66 | 4,673 | 16 | 5,204 | 18 | 1 | 0 | 48 | 0 |
| Forest | 13,694 | 51 | 5,185 | 19 | 7,569 | 28 | 6 | 0 | 297 | 1 |
| Brush | 58,860 | 60 | 17,133 | 17 | 21,900 | 22 | 5 | 0 | 295 | 0 |
| Grass | 45,726 | 51 | 17,863 | 20 | 25,803 | 29 | 3 | 0 | 313 | 0 |
| Authorized, unauthorized, prescribed | 127,998 | 77 | 18,940 | 11 | 19,730 | 12 | 3 | 0 | 45 | 0 |
| Other | 172,647 | 71 | 36,128 | 15 | 34,540 | 14 | 8 | 0 | 131 | 0 |
| Total | 931,373 | 72 | 145,933 | 11 | 221,604 | 17 | 48 | 0 | 1,454 | 0 |
We were unable to identify shapefiles designating local direct protection areas for all states, but shapefiles for national parks and national forests, grasslands, and wilderness areas to examine MODIS incidents in areas of federal control were used. This omits land under the management of the Bureau of Land Management, Bureau of Reclamation, Bureau of Indian Affairs, and Fish and Wildlife Service as well as state parks and state forests. However, we found that 6.4% of MODIS incidents occurred within national parks and national forests, grasslands, and wilderness areas.
6 https://fsapps.nwcg.gov/afm/gisdata.php