6.2. ESTUDIO FINANCIERO
6.2.4. Relación Beneficio Costo
The first goal of this chapter was to identify which available GIS data layers may best describe this highly variable landscape in the context of studying the spatial distribution of (native or alien) plant species. The results of automated workflows as a linear sequence of GIS operations and tools (i.e. Geoprocessing and Spatial Analysis) generated relevant GIS data layers that have been stored in a geodatabase of Banks Peninsula. Geoprocessing is useful because it is repeatable and self-documenting. If it is integrated with spatial analysis (so-called geospatial analysis) as iterative processes or methods of modelling it may generate results and evaluate them at the same time. To enhance the results in order to obtain the best representation not only of heterogeneous environment, there is a clear need for the combined use of GIS, remote sensing, spatial analysis and expert knowledge. Further development of the interoperability between GIS and RS data and tools associated with spatial analysis would be a valuable technique in the study of plant species distribution not only in a highly variable landscape such as Banks Peninsula.
In my study, although the available data layers that best described the Peninsula environments have been identified, there is room for further improvement. For instance, the layers used as surrogates for human-related impact, for example, human population density, distance to buildings, distance to paved and unpaved roads, may be inadequate. The 1983-1988 systematic survey had for example sampled the Peninsula environments with some sampling limitations (e.g. detecting fewer population density areas and in areas near rivers). Land-use history was also inferred from the data rather than assessed independently (see Methods). The arguments about the importance of land-use and -management would be much stronger if an independent assessment of land-use and land-cover will be made. Another limitation can be seen in the fact that Hugh Wilson’s plots were established before the days of GPS units, resulting in an error of±100 m of their spatial position (Susan Wiser personal communication). This means that the intersections with spatial layers need to accommodate this error. However, moving in all the directions for 100 m would have not changed the estimates of the data layers (i.e. explanatory variables) as they had a coarse spatial resolution (i.e. 500×500 m) but excluding DEM layer of 10×10 m spatial resolution. The derived measures from elevation, such as aspect and slope, have been assessed directly on each plot by Hugh Wilson and this may have been given the best estimate for these explanatory variables.
In my case, once these available data have been identified and quantified the second goal was to test if the Banks Peninsula systematic sampling survey that contains the species data is good enough to detect the range of environments found on the Peninsula. Although the systematic and random sampling methods are purely spatial techniques and extra information about the environments is neededa priori
to conduct a stratified sampling survey, the results indicated that the systematic sampling survey is the most accurate method, followed by the stratified and random methods. Thus, consistent with the findings of Hirzel and Guisan (2002) and
Role˘ceket al.(2007), the first two methods (i.e. systematic and stratified sampling) performed well confirming the validity of those approaches when dealing with a heterogeneous landscape such as Banks Peninsula characterized by highly variable climatic, human-related and environmental factors.
Obviously, a range of different sampling methods rather than these unbiased sampling designs (e.g. grid, random points and stratified targeted) can be used[see
Hirzel and Guisan(2002);Rewet al.(2006);Maxwellet al.(2012)]. For instance,
to detect species presence/absence and abundance in early stage of plant invasions, random transects and roadside sampling methods can be used. However, "because of the fragmented distribution of the plant species and to best perform, these methods need to increase the number of samples, leading into more time to detect and into less efficiency" (Morrisonet al.,2008). When plant species populations are patchier and dispersed, like alien plant species at early and later stages of invasion, the adaptive cluster sampling methods are generally proved to be the most time-efficient and effective in detecting plant species (Thompson, 2006). However, the use of adaptive cluster sampling methods has only been recently discovered for surveying alien plant species and it requires good computational effort. In conclusion, we need to keep in mind which sampling methods best achieve the objectives/goals of our study (Maxwellet al.,2012).
Although the systematic survey performed better than the other methods, the latter detected fewer population density areas and was conducted near rivers. This may indicate that the 1983-1988 systematic survey was carried out in less urbanized areas and in areas more freely accessible by chance or because areas with high population density are rare on the landscape. As sampling is a money and time consuming task, prioritisation of requirements is of great importance (Hirzel and Guisan,2002). In this sense, I confirm the findings ofRewet al.(2006) andHirzel and Guisan(2002) that although there may be limitations to the systematic sampling method [e.g. the length of time required, up to 5 years; a restricted choice of
plot selectionsensu Hirzel and Guisan(2002)], or it may be adequate for sampling grassland vegetation in New Zealand (Wiser and Rose,1997;Hurst and Allen,2007) but undersampling in forest and probably in sand dunes and cliffs (Phil Hulme personal communication), this is the "best" design option when the aims are to: (1) conduct a comprehensive floristic survey over a heterogeneous landscape such as Banks Peninsula, and (2) model within it the distribution of native or alien plant species in relation to the spatial distribution of environmental factors (c.f.Huebner,
2007;Role˘ceket al.,2007).
Once the "best" sampling method has been identified, the third goal was to deter- mine if the data used as explanatory variables to describe plant species distribution violated any statistical assumptions. The results show that a series of problems (i.e. outliers, colinearity, non-normality and spatial autocorrelation) were encountered and these may seriously affect the results of an analysis. In all cases, the problems can lead to statistical models that are wrong (Kühn and Dormann,2012). Zuuret al.
(2010) suggested that such problems can be avoided only by applying systematic data exploration before embarking on analysis. In the Banks Peninsula geodatabase for ex- ample, the results have shown that the data layers are either spatially autocorrelated or suffer different degrees of colinearity. Nevertheless, the spatial autocorrelation of those potential variables that varied over distance could potentially reveal the effects of underlying ecological and environmental gradients (Guoet al.,2012). Thus, it was crucial to build a spatial component into the statistical models for a better under- standing of the pattern of species on the Peninsula and to increase confidence in data interpretation (see Chapter 4). If spatial autocorrelation is ignored or removed from among the explanatory variables this would also remove most of the power of the explanatory variables (Guoet al.,2012), and we simply would not know if the results are to be trusted. As already stated byDinizet al.(2003), the presence of residual spatial autocorrelation should always be tested for in spatial ecology and appropriate methods should be used if there is shown to be significant spatial autocorrelation (Kühn and Klotz,2007). Colinearity is another statistical challenge, as it could affect the response variable interactively. On the one hand, if strong colinearity is detected, an easy way to deal with it could be not to include highly correlated variables in the analysis. However, information and understanding may be lost when ecological patterns and processes are influenced by additional factors not selected. On the other hand, adding more variables potentially offers more hypotheses and tests, and interpretations of greater explanatory power (Guoet al.,2012). Thus, a trade off in selecting variables is always necessary. With these issues adequately addressed in
a way that makes ecological sense, questions asked of the dataset, regarding plant species patterns and processes will not be biased by the quality of the data and the chances of making wrong ecological conclusions and poor recommendations will be reduced.