When Fire Changes the Observer: Negative Controls in Citizen-Science Estimates of Wildfire Effects on Birds

This study looks at how citizen science bird observation data called eBird can support a causal claim about fire. I joined 4,445,934 eBird bird detection checklists from California from 2015-2026, joined with MTBS fire burn severity data and satellite-derived vegetation indices from MODIS. These attempted detections were consolidated into 5-km x 5km cells, each tracked in 16-day windows, giving a total of 622,219 records. Instead of recording whether a square burned, they were recorded by how much they burned and how severely. The detection rates had a sharp first-year drop, a -0.0298 detection rate decrease at a full burn, which is about a 3 percentage-point drop, slowly recovering over a 4-year period. Negative controls were then put in place, using three control/water birds, the Mallard, American Coot, and Black Phoebe, which should theoretically all be unaffected by fires in chaparral, the area where the Wrentit lives. All three species registered significant effects, signaling that the design is measuring something other than fire. This can be explained by a 42% drop in distinct locations birded. Because of this, the tests resulted in per-species detection estimates and migration results being withdrawn, opting for comparisons between whole groups of birds, also known as guild contrasts. The comparisons did not clear their own check either. What stayed is the direction test, that 12 of the 16 non-control species moved the way they were predicted to before the analysis ran.