Table 2. Sources of error.

	Procedure	Result	Source of variability	Variability	Notes and suggestions for future work
1	Supervised classification	Area of penguins at each colony	Interpretation error: manual interpretation of which pixels constitute penguins as opposed to snow, guano or shadow. Variability here stems from being able to accurately determine penguins in the image, and repeatability between operators.	Less than 10% with most imagery but progressively worse with poorer imagery. Can be as much as 50% out in worst cases.	Depends upon the quality of the imagery. We suggest that future satellite acquisitions should avoid images with heavy guano staining.
2	Chick/adult area assumption	Area of adults at each colony	Chick adult ratio error: we make the assumption that the ratio of pixels showing as penguin in the satellite imagery remains constant to the number of adult pairs: i.e. That the area of larger chicks and fewer adults seen late in the season (November) is equal to the area of adults seen by the satellite earlier in the season (September, when chick are virtually invisible to the satellite).	Unknown at this stage, but the high correlation in good imagery from robust regression analysis confirms that the assumption is broadly true.	We suggest further work is needed to assess the variability. At present there is not enough ground truthing linked to satellite imagery over the period when the imagery is acquired.
3	Ground truthing estimates	Number of adults at selected colonies	From ground counts a mixture of error sources, mainly the error associated with counting an areas and scaling up to the whole colony. In aerial counts there can be variability in the manual interpretation of how many penguins (especially chicks) are on an image.	Approximate variability of ground truthing is around 10% using aerial photography, but can be higher for ground counts especially at larger colony sites	Low level vertical aerial photography is recommended to minimize ground truthing errors.
4	Statistical analysis	Estimated of adults at each colony at time of image	Statistical error: conversion of the pixels to penguins relies on a regression between area identified as penguin and the number of adults from ground truthing. Enough good ground truthing, concurrent with satellite imagery must be available to make this regression accurate.	1.75% based on Monte-Carlo analysis	More ground truthing over the entire season is recommended to improve the statistical procedure.
5	Seasonal assumption	Autumn population estimate	errors in the ground truthing and fluctuations between the dates of the ground truthing and satellite imagery	How this varies on a daily or weekly basis is at present unknown	Data from colonies where counts on seasonal variability would be useful. Especially if data exists on daily and weekly fluctuations in adult and chick numbers.
6	yearly population estimate	population estimate for 2009	Conversion between spring population and total population. Literature suggests that only 10% of birds are non- breeders	No variability estimate in literature	Further investigation required. Ground data from long term monitoring sites needed.
7	Inter-annual variability	Mean population estimate	Inter-annual changes at each emperor colony	Different estimates between colonies. Possibly size dependent (see text)	Monitor all colonies over multiple years by satellite to assess population change

The various sources of error; see section on Accuracy and uncertainty in the Discussion for further details of each area.