How does a GenEst export work for a wind farm PCFM program?
A GenEst export for a wind farm PCFM program is a bundle of five CSV files that match the column layout the GenEst R package and web tool expect as input. GenEst will not run without all five, and it will not run if any of them have the wrong column names, date formats, or missing turbines. The IFC PCFM Handbook frames the analysis step as a non-negotiable extension of the field program:
Generating robust estimates of total bird and bat fatalities at a WEF goes beyond fieldwork; it is necessary to analyze those data using fatality estimation tools and present the technical results of the PCFM analysis so that others can understand how results were derived.
— IFC PCFM Handbook, §5
What is in the bundle
| File | What it represents |
|---|---|
| SE | One row per planted carcass, with detection or non-detection recorded against each scheduled visit. Used to fit the searcher efficiency model. |
| CP | One row per monitored carcass, with timestamps from each persistence check. Used to fit the persistence curve. |
| SS | One row per plot-visit, recording when each turbine or plot was actually searched. Used to compute exposure time. |
| CO | One row per real carcass found during routine searches, with species, date, and location. |
| DWP | One row per turbine, recording the density-weighted proportion of the fall zone that the search plot covers. |
Where exports break
Each file has a specific column layout GenEst expects, and the IFC PCFM Handbook is unambiguous that the entire fatality estimate stands or falls on the integrity of these inputs:
“These sources of bias are why it is important to use field and analytical methods associated with bias correction to estimate detection probability accurately so that estimates of total, project-generated fatality rates are not biased.”
— IFC PCFM Handbook, §5.1.2
In practice this is where most manual GenEst submissions fail: date format mismatches, trailing whitespace in species codes, renamed columns, missing turbines that had zero observations, search dates reconstructed from memory, and DWP delivered weeks after the rest of the bundle by an outside contractor. The model never sees the fieldwork directly; it sees the five CSVs, and the fatality number is only ever as defensible as those files.