Why do raw carcass counts underestimate true bird and bat fatalities at wind farms?

Q: Why do raw carcass counts underestimate true bird and bat fatalities at wind farms?

Raw carcass counts at a wind farm always underestimate the true fatality rate because three biases strip carcasses out of the count: searchers miss carcasses that are present, scavengers remove carcasses before searches happen, and many carcasses fall outside the searched area entirely. The IFC PCFM Handbook requires correcting for all three to produce a defensible fatality estimate.

Raw carcass counts at any wind farm undercount the true mortality rate. The number of carcasses observers actually find is only ever a fraction of the number generated by the facility, because three independent biases pull carcasses out of the count before they can be recorded. The IFC PCFM Handbook treats this as the foundational problem PCFM exists to solve:

The term “bias” with respect to PCFM means that counts of carcasses are lower than the true number of fatalities. Bias in PCFM studies is generated in three principal ways:

Searchers do not find all carcasses (searcher efficiency bias).

Not all carcasses persist long enough to be found (carcass persistence bias).

Not all carcasses fall within searched areas (unsearched and unsearchable area bias).

— IFC PCFM Handbook, §3.2.3

What each bias removes from the count

Searcher efficiency bias. Even on the searched ground, observers do not find every carcass present. Small size, cryptic coloration, damaged remains, and dense vegetation all reduce detection. Measured by Searcher Efficiency (SE) trials.
Carcass persistence bias. Carcasses are removed by scavengers, weather, decomposition, and farm operations between searches. The longer the search interval and the smaller the carcass, the larger this bias gets. Measured by Carcass Persistence (CP) trials.
Unsearched-area bias. Carcasses fall across the entire fall zone, much of which is outside the search plot or inaccessible to observers (steep terrain, dense vegetation, hazards). Corrected by Density-Weighted Proportion (DWP).

Why this matters for the final number

Detection probability is the product of these three components plus the sampling fraction. The IFC PCFM Handbook is explicit that this is the mechanism that makes a fatality estimate defensible at all:

In the real world, detection probability is always less than 1 due to the combined effects of factors that can prevent searchers from detecting fatalities. However, the primary benefit of the bias correction methods discussed in this Handbook is that, as long as biases can be characterized and the detection probability can be estimated, it is possible to estimate the “true” or unbiased bird and bat fatality rates generated by a WEF.

— IFC PCFM Handbook, §3.2.4

A detection probability of 0.1 means raw finds represent only 10 percent of actual mortality. Without all four bias components captured cleanly throughout the year, the program has no way to recover the true number, and no way to defend whatever it reports.