Sensitivity analysis of logistic regression parameterization for land use and land cover probability estimation

Abstract

The recent availability of regional-scale land use and land cover data made it possible to apply land use and land cover models at regional-, national- and even continental-scale levels. A commonly used modelling approach is based on the assessment of land cover probabilities by means of logistic regression equations. In most cases, however, the error involved in the parameterization of logistic regression equations is not known.

In this article, the sensitivity of logistic regression parameterization for land use and land cover probability assessments is analysed by comparing the results using input maps from different sources. Land cover maps with a range of accuracy levels were collected for a sub-catchment of the Lake Balaton watershed in Hungary. The results show that the parameterization of the logistic regression coefficients is highly dependent on the quality of the input maps. Both the spatial pattern and the area covered by a specific land cover type have a direct influence on the error propagation in the land cover probability maps. These findings should be taken into account when interpreting the results of land use and land cover change models at regional- and national-scale levels.

Keywords:

• data quality
• land cover change
• logistic regression

Acknowledgements

The work done by three anonymous referees is greatly acknowledged. Their comments and suggestions significantly improved this manuscript. This study was financially supported by the Hungarian Scientific Research Fund (OTKA K60203).