Are landscape ecologists addressing uncertainty in their remote sensing data?

Abstract

In this quantitative review, we investigate the degree to which landscape ecology studies that use spatial data address spatial uncertainty when conducting analyses. We identify three broad categories of spatial uncertainty that are important in determining the characterisation of landscape pattern and affect the outcome of analysis in landscape ecology: i) classification scheme uncertainty, ii) spatial scale and iii) classification error. The second category, spatial scale, was further subdivided into five scale dependent factors i) pixel size ii) minimum mappable unit, iii) smoothing, iv) thematic resolution and v) extent. We reviewed all articles published in the journal Landscape ecology in 2007 and recorded how spatial data was used and whether spatial uncertainty was addressed or reported in ecological analyses. This review found that spatial uncertainty was rarely addressed and/or reported. Most articles used the default pixel size of the sensor. Only 23% of articles addressed one or more scale dependent factors and 47% reported one or more as issues. A single study of the 59 investigated the effect of classification accuracy on ecological analyses. We demonstrate that spatial uncertainty is not being addressed as standard practice in analyses in landscape ecology, and then describe methods to test for spatial uncertainty and potential solutions that can be developed in the future.

On the Use of Maps and Models in Conservation and Resource Management (Warning: Results May Vary)

“…While the concepts associated with error and uncertainty propagation have been studied in the geospatial literature (Fisher and Tate, 2006;Wilson, 2012), the attempts to raise end-users' awareness in fields like ecology and environmental modeling have failed (Brown and Heuvelink, 2007;. With a few exceptions from the terrestrial literature (e.g., van Niel and Austin, 2007;Livne and Svoray, 2011), and despite repeated calls for the appropriate consideration of error and uncertainty propagation in environmental modeling and mapping (Rocchini et al, 2011;Beale and Lennon, 2012;Lechner et al, 2012), these concepts have yet to be better implemented, especially in a marine context .…”

“…Data quality Van Oort and Bregt, 2005Devillers and Goodchild, 2010Li et al, 2012Robertson et al, 2010Lechner et al, 2012Rocchini et al, 2011Gallo and Goodchild, 2012Moudrý and Šímová, 2012Cros et al, 2014Lecours et al, 2017a Scale Woodcock, 1987Quattrochi and Goodchild, 1997Marceau and Hay, 1999Duncan et al, 2002Goodchild, 2011Zhang et al, 2014Wiens, 1989Borcard et al, 2004Rahbek, 2005Seo et al, 2009Austin and Van Niel, 2011Moudrý and Šímová, 2012Bradter et al, 2013Wilson et al, 2007Anderson et al, 2008Brown et al, 2011Harris and Baker, 2012bRengstorf et al, 2012 …”

Recent Progress on Spatial and Thematic Resolution in Landscape Ecology

“…In remote sensing, spatial resolution is primarily a product of the pixel size of the sensor or the object-scale for data produced through geographic object-based image analysis [22][23][24][25][26]. Pre-and post-processing of spatial data, commonly used for improving classification and geometric accuracy, also affects spatial resolution through: removing features below a minimum area (minimum mappable unit) [27,28], the application of smoothing filters [29] and the characteristics of the remote sensing device [30].…”

“…In this paper we focus primarily on observation scale effects from changing spatial and thematic resolution in the context of GIS analysis using data derived from remote sensing. The combination of GIS analysis with remote sensing data represents one of the most common ways for representing and analysing data in landscape ecology [25,36,37].…”

Translating land cover/land use classifications to habitat taxonomies for landscape monitoring: a Mediterranean assessment

“…The selection of an appropriate LC/LU classification system for habitat mapping applications is a crucial issue for long term monitoring. It is particularly important when working with remote sensing imagery at very high spatial resolution due to the complexity of class description and the limited spectral resolution (few spectral bands) that requires multi-temporal imagery and the integration of ancillary data in order to minimize uncertainty in mapping (Lechner et al 2012). The qualitative review of CLC, IGBP and LCCS for habitat mapping oriented applications in Mediterranean sites indicates that LCCS allows the finest discrimination of natural and semi-natural types with respect to CLC and IGBP by using the simple pure land cover classifiers of the Modular-Hierarchical phase.…”