Author
|
Submitted to: Meeting Abstract
Publication Type: Abstract Only Publication Acceptance Date: 2/15/2018 Publication Date: 4/11/2018 Citation: Goslee, S.C. 2018. This is like that, only bigger and messier.{Abstract}.1. Interpretive Summary: No interpretive summary is required for this Abstract. JLB. Technical Abstract: Cluster analysis is a core tool of vegetation science; we have always wanted to divide a complex world into manageable chunks. In vegetation science, we classify both vegetation and sites. Both have clear management applications. Various types of spatial classifications are used to delineate agroecological land units and standardize management recommendations, such as MLRAs and forage suitability groups. These classifications consider qualitative differences in soils and climate, but often do not have a consistent quantitative basis. Can these agricultural land unit classifications be improved? Problems appear almost immediately if the classification is to be derived from geospatial data at fine resolutions and large extents. Few clustering algorithms can handle such datasets easily on current computing hardware. Ensemble clustering is often used to compare different methods across a single dataset, but can equally well be used to combine the results of a single method carried out repeatedly on subsets of a larger dataset. When clustering complex environmental data, there are rarely clear divisions between groups. A standard crisp clustering hides this variation: a unit that is 52% similar to group A but 48% similar to group B will be classified in A. When considering management recommendations, the best choice may be to look at the options for both A and B. Fuzzy clustering extends the idea of cluster analysis: instead of "is it A or B," the question becomes, "how similar is to to A, and to B?" Recent work on categorizing agricultural phosphorus loss risk in Pennsylvania explored the utility of four types of cluster analysis (hierarchical, partitioning, density-based, and model-based) for grouping complex environmental data into useful agroecological clusters. Both ensemble methods and fuzzy clustering techniques were employed. For complex environmental problems there is no one clear "correct" method. Instead, the desired use must be balanced against the assumptions about cluster shape and size inherent in each technique. Advances in methodology make a greater range of techniques usable, and improve the ability of cluster analysis to provide meaningful and useful results. |
