Modelling uncertainty of a land management map derived from a time series of satellite images

Abstract

The agricultural management practice of leaving land fallow during winter is a key pressure on ground water quality in Canterbury, New Zealand. This is because any nitrate present is likely to be leached down through the soil profile since there is no plant uptake. Remote sensing imagery has been successfully used to identify land with low potential for nitrate uptake due to having bare soil or dead vegetation for significant periods in winter and early spring. This was achieved by use of classification rules on a time series of three 20–30-m resolution satellite images. Our rules are based on percentage live vegetation cover, so a regression relationship between percentage live vegetation cover (from field data) and a satellite-image-derived vegetation index was derived. This paper describes the rules and analyses the sources of error in this low-nitrate-uptake land classification process. The objective was to provide a low-nitrate-uptake map that included uncertainty, and to determine the most significant error source, a task made more difficult due to a lack of reference data. Sources of error include suitability of the logical model, timing of image acquisition, incorrect reference data, geometric errors between images, and radiometric variation both within and between images (month to month and year to year). A pragmatic approach was adopted where each potential error source was systematically examined in turn, quantified where possible, and combined to produce a map indicating where predictions were less reliable or certain. This approach is suitable for situations where there is very limited reference data and more quantitative approaches cannot be applied. It has the advantage of using subsidiary information about errors, and produces spatial information on uncertainty for end users, albeit qualitative and very dependent on expert opinion. The most significant source of uncertainty in the rule-based maps of low nitrate uptake was missing images or too large an interval between images.

Acknowledgements

This work was supported by the New Zealand Foundation for Research, Science and Technology, contract number C09X0306. David Pairman and Daniel Rutledge are thanked for their helpful comments, as are Stephen Burgham, who assisted with the field data, and Stella Belliss, who prepared the images. Two anonymous reviewers are thanked for their efforts which allowed us to improve the manuscript.