Remote-sensing image analysis and geostatistics

The random function theory forms the basis of geostatistics and allows modelling of the uncertainty associated with spatial estimation and simulation. Remote sensing involves gathering information about an object by measuring signals composed of radiation, particles and fields emanating from an object with a sensor that is not in direct contact with it. This article reviews the present state of the art of how geostatistics is used in remote-sensing studies by reviewing the 2000–2010 literature in this field. This article first addresses the issue of stationarity in the context of image analysis and reviews whether this holds as it is a basic assumption of most geostatistical techniques. Ways to relax the assumption are discussed. Following is the use of variograms to quantify image structure and texture, and the use of variograms to address issues of optimal scale of observation. Next, various kriging-based estimation techniques (parametric and non-parametric) and how these can be used to enhance image information, fill missing pixel information and downscale information through super-resolution techniques (e.g. downscaling with preservation of spatial structure at the finer resolution) are discussed. Cokriging techniques are discussed to enable the combination of various variables at different support sizes and to link field and image data. To address issues of uncertainty and to characterize landscape heterogeneity, stochastic simulation techniques are discussed. A bibliometric analysis is presented which places the field of ‘remote sensing and geostatistics’ in a broader geosciences context and explores who are the key scientists and research groups that have contributed to the development of this field. It is concluded that the field of remote sensing and geostatistics has further developed the use of various pre-existing numerical techniques, which has led to new application areas. However, few new geostatistical techniques have been developed for use in remote sensing. In particular, aspects of scale and spatial sampling have been extensively addressed, but issues related to monitoring and space–time analysis have been largely neglected.