Issue link: https://read.dmtmag.com/i/35322
Windows Extend Information into ‘No-Data’ Areas BEYONDMAPPING W BY JOSEPH BERRY ildfire modeling increasingly intrigues me. For a spatial-analysis enthusiast, it has it all: headline-grabbing impact, real-world threats to life and property, and “action hero” allure. It also features a complex mix of geo- graphically dependent “driving variables” (fuels, weather and topography) and extremely challenging spatial analytics. Too Difficult for Data However, with all their sophistication, most wildfire models tend to overlook some very practical considerations. For example, Figure 1 identifies an extension that “smoothes” the salt-and-pepper pattern of the individual estimates of flame length for individual 30- meter cells (left side) into a more continuous surface (right side). This is done for more than cartographic aesthetics, as surrounding fire-behavior conditions are believed to be important. It makes sense that an isolated location with pre- dicted high flame-length conditions adjacent to much lower values is presumed to be less likely to attain the high value than one surrounded by similarly high flame-length values. Also, the mixed-pixel and uncertainty effects at the 30-meter spatial resolution suggest using a less myopic perspective. The figure’s top-right portion Joseph Berry is a principal in Berry & Associates, consultants in GIS technology. He can be reached via e-mail at jkberry@du.edu. 10 shows the result of a simple-average five-cell smoothing window (150- meter radius), while the lower inset shows results of a 10-cell reach (300 meters). Wildfire professionals seem to vary in their expert opinion (often in heated debate—yes, pun intended) of the amount and type of smoothing required, but they seem to agree that none (raw data) is too little, and a 10-cell reach is too much. The most appropriate reach and type of smoothing to use will likely keep fire scientists busy for a decade or more. In the interim, expert opinion prevails. GEO W ORLD / JUL Y 2O11 The ability to “iteratively ooze” information into an area step-by-step keeps the data bites small and localized, similar to the brush strokes of an artist. An even more troubling limitation of traditional wildfire models is depicted as the “white region” in Figure 1, which represents urban areas as “no data,” meaning they’re areas of “no wildland fuel data” and can’t be simulated with a wildfire model. The fuel types and conditions within an urban setting form extremely complex and variable arrangements of non-burnable to highly flammable conditions. Hence, wildfire models must ignore urban areas by assigning no data to these extremely difficult conditions. But all too often, wildfires ignore this artificial boundary and move into the urban fringe. Modeling the relative venerability and potential impacts within the no-data area is a critical and practical reality. Figure 1. Raw flame-length values are smoothed to identify the average calculated lengths within a specified distance of each map location—from point-specific conditions to localized conditions that incorporate surrounding information (smoothing).