discussion section from report

classification-systems/emergent/discussion.md

@@ -2,7 +2,7 @@
 
 Classification schemes are incredibly important in describing how we understand our natural world. While always purposeful, they are often not systematic. Land use is inherently not any single thing. It is some combination of qualities at a particular scale (or resolution) which can be assembled to provide meaningful information. At present, our classification schemes are based on a small set of available data, most of which is not directly fit for purpose, along with a massive amount of expert knowledge in the form of interpretation, inference, rules, assumptions, and probabilities. This form of expert knowledge is largely implicit, which negatively affects the interoperability of the classification schemes.
 
-As an example, consider Figure 3, which represents the concept of horticulture as it is described (at a high level) in several previous land classifications discussed in Section 7 of the final report [CITE}]. Each classification scheme in figure 3 is uniquely coloured, showing links to concept definition or description as it relates to defining the class along with broader or narrower concepts where applicable. The first thing to notice is the similarities. One could argue these should not be separate entities at all; merging these categories would allow for a richer description and property association of the overall concept (as a class). However, most, if not all, of the descriptions and/or definitions are not rich enough to determine the differentia, or similarity, between them. As such, important questions arise. Which crops does orchard include or exclude? Do land management practices need to be considered; and if so, when? What are the intended spatial units (or range)? Which data were used in determining, and identifying, the concepts of horticulture as applied in practice? In the absence of transparent methodologies (including the possibility of numerical boundaries to distinguish certain concepts) in existing classifications, there may be no other option but to loosely infer the answers to these questions.
+As an example, consider Figure 3, which represents the concept of horticulture as it is described (at a high level) in several previous land classifications discussed in Section 7 of the final report [CITE}]. Each classification scheme in figure 3 is uniquely coloured, showing links to concept definition or description as it relates to defining the class along with broader or narrower concepts where applicable. The first thing to notice is the similarities. One could argue these should not be separate entities at all; merging these categories would allow for a richer description and property association of the overall concept (as a class). However, most, if not all, of the descriptions and/or definitions are not rich enough to determine the differentia, or similarity, between them. As such, important questions arise. Which crops does the concept ```orchard``` include or exclude? Do land management practices need to be considered; and if so, when? What are the intended spatial units (or range)? Which data were used in determining and identifying the concepts of horticulture as applied in practice? In the absence of transparent methodologies (including the possibility of numerical boundaries to distinguish certain concepts) in existing classifications, there may be no other option but to loosely infer the answers to these questions.
 
 Perhaps the classes are similar because they are based on the same data. Does the available data drive our questions? What we do not have easy access to is a clear understanding of which supporting data was used, how it was used or generated (subset, processed, etc.) and which other data may have been incorporated. For example, LUNZ used LCDB with AgriBase to determine horticulture classes. How many other categories rely on this information, directly or indirectly?
 
@@ -11,8 +11,8 @@ Any classification scheme, or system of categories, adhering to the principles o
 Another way of expressing this idea is to make the point that the preceding NZLUM classification system is a “top down” classification system that begins from a consideration of what classification structure would be useful for most intended use cases. It does not necessarily entail that sufficient data and/or information actually exists to fill, or represent, the categories in classification structure in practice. Therefore, another (complementary) approach is to consider a “bottom up” classification system that first establishes what information exists, including its quality, availability, and access rights, and only then systematically determine what land uses classes might be achievable in practice.
 
 Ultimately, the ideas of top-down and bottom-up classification systems are complementary, since once it is established what information is available for use within a land use information system, one still needs to determine what conceptual ideas to aim at expressing as knowledge. Doing both establishes the existence of data gaps that must be addressed before complete realisation of a useful land use classification system.
-The shortcoming of a top-down classification system is that it is an open-ended problem. Although it is possible to satisfy the needs of end users with respect to a land use classification system when designed top down, from a bottom-up perspective there is always the potential for new data which needs to be integrated into the knowledge graph. As such, we have not produced a complete example of this idea for this report. An example of this approach, clearly specifying the attributes to be captured for the classification of land use for soil quality monitoring, based on the on-site visitation of sampling sites is described in Cavanagh and Whitehead (2022) and further developed in Cavanagh and Whitehead (2023). 
- 
+
+The shortcoming of a top-down classification system is that it is an open-ended problem. Although it is possible to satisfy the needs of end users with respect to a land use classification system when designed top down, from a bottom-up perspective there is always the potential for new data which needs to be integrated into the knowledge graph. As such, we have not produced a complete example of this idea for this report. An example of this approach, clearly specifying the attributes to be captured for the classification of land use for soil quality monitoring, based on the on-site visitation of sampling sites is described in Cavanagh and Whitehead (2022) and further developed in Cavanagh and Whitehead (2023).
 
 Figure 3. The concept of 'horticulture' as represented in previous land use classifications.
 ![CMap horticulture figure](../../figs/CLU_horticulture_v4_trim.svg)