Landscape Ecology - Lecture Notes

Landscape Pattern Analysis: Errors, Accuracy, and Scale Effects

---

1. Errors in Landscape Ecological Data

---

Error propagation in landscape analysis using multiple data layers:

From G. Hess (1994):
 
• [...,]  I raise the issue of classification error in the thematic GIS representations of remotely-sensed data.  Landscape ecologists have been using such data to calculate measures of broad-scale landscape pattern, but have devoted no effort to quantifying the uncertainty in these measures.
• The problem is that there is always classification error in the land cover data used to calculate pattern measures.  The GIS data represent a final product of a complicated process that introduces error at many points.
• Yet without statistical confidence in the measures used, scientists cannot evaluate correlations between landscape pattern and ecological processes.
• Without statistical confidence one cannot use measures of pattern to detect differences in landscapes over space, or changes in a landscape over time.
• Progress in testing hypothesized relations between landscape pattern and ecological process requires that both be measured and that errors inherent in the measurements be understood and quantified.
• The lack of error information ultimately limits our ability to draw statistically valid conclusions about the degree of correlation between landscape pattern and ecological process.
• Coordinated efforts are needed among members of the landscape ecology, remote sensing, GIS, and statistics communities to resolve the issues on classification errors.

* Inappropriate choice of methods

* Some landscape metrics do not always do what they are supposed or known to do.  For example:


• Contagion does not always measure "contagion".
• The calculation of fractal dimension in many landscape ecological studies using linear regression may be flawed.

* Problems of Spatial Analysis with Area-Based Data (MAUP; see Jelinsk and Wu 1996)

---


What Is MAUP?

• MAUP stands for “modifiable areal unit problem” (Openshaw1984, Jelinski andWu 1996).
• Scale-related studies in landscape ecology include three distinct but intrinsically linked issues: characteristic scales, scale effects, and scaling.

An illustration of two aspects in areal aggregation: change in scale (here referring to the spatial resolution or grain size) and change in zoning systems or zoning alternatives at the same scale (i.e., the shape and orientation of the areal units). The numbers denote grain sizes. Notice that grain sizes change only along the columns, whereas along each row the way of zoning varies. Two numbers in each group are the respective numbers of rows and columns of basic spatial units or BSUs (i.e., the original grid cells) contained in the aggregated areal unit in the repartitioned data set. 

---

Examples

Examples of scale and zoning problems in statistical analysis. The original data set is a 16 x 16 matrix. To examine scale effect, the data is progressively aggregated across 8 spatial scales at each of which summary statistics (mean and variance) and autocorrelation (Moran's I and Geary's c) are calculated. To examine zoning effect, the original data is repartitioned differently at two spatial scales (16 and 64 BSUs), resulting in 8 zoning alternatives, at each of which the same statistics are computed. In all cases, variance and autocorrelation change markedly with both scale and zoning alternatives, while mean seems resistant to these changes. Also see a schematic description of ways of data aggregation.  (From Wu & Jelinski 1995).

    

Simpson's Paradox

 




• Li, H. and J. Wu. 2004. Use and misuse of landscape indices. Landscape Ecology 19: 389-399.