But they can give a very accurate count of the minimum number in any on areaI have colleagues doing formal survey work with thermal drones, and attempting to establish the boundaries of the measurement error.
As you might expect, they’re very good for large things in open spaces. They get rapidly worse as things get smaller and vegetation thicker.
The emerging consensus is that they will speed up and reduce the cost of counting, but the margin of error isn’t going to change by much. With things like roe and muntjac in woodland, it looks like they’re about as useless as every other way of counting them!