Infometric and statistical diagnostics to provide artificially-intelligent support for spatial analysis: the example of interpolation

Abstract

The wider uptake of GIS tools by many application areas outside GIScience means that many newer users of GIS will have high-level knowledge of the wider task, and low-level knowledge of specific system commands as given in reference manuals. However, these newer users may not have the intermediate knowledge that experts in GI science have gained from working with GI systems over several years. Such intermediate knowledge includes an understanding of the assumptions implied by the use of certain functions, and an appreciation of how to combine functions appropriately to create a workflow that suits both the data and overall goals of the geographical analysis task. Focusing on the common but non-trivial task of interpolating spatial data, this paper considers how to help users gain the necessary knowledge to complete their task and minimise the possibility of methodological error. We observe that both infometric (or cognitive) knowledge and statistical knowledge are usually required to find a solution that jointly and efficiently meets the requirements of a particular user and data set. Using the class of interpolation methods as an example, we outline an approach that combines knowledge from multiple sources and argue the case for designing a prototype ‘intelligent’ module that can sit between a user and a given GIS. The knowledge needed to assist with the task of interpolation is constructed as a network of rules, structured as a binary decision tree, that assist the user in selecting an appropriate method according to task-related knowledge (or ‘purpose’) and the characteristics of the data sets. The decision tree triggers exploratory diagnostics that are run on the data sets when a rule requires to be evaluated. Following evaluation of the rules, the user is advised which interpolation method might be and should not be considered for the data set. Any parameters required to interpolate the particular data set (e.g. a distance decay parameter for Inverse Distance Weighting) are also supplied through subsequent optimisation and model selection routines. The rationale of the decision process may be examined, so the ‘intelligent interpolator’ also acts as a learning tool.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 16 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.