From logic-based to algebra-based specification of geo-analytical workflows

Abstract

This thesis explores the use of a domain-specific algebra as an input for a SAT-based workflow synthesis system. With this algebra, I try to make the gap smaller from logic-based input to natural-language-based input for the use of workflow synthesis systems in the Geoscience domain. The thesis extends the Automatic Pipeline Explorer (APE) to use the Geospatial Analytical Transformation Algebra(GATA) as an extra input. The goal is to let a user specify their intent using a GATA expression and APEwill find a correct workflow that recreates that input. This is possible because components of workflows also can be represented by GATA. A workflow is correct if It represents the input GATA expression.To accomplish this, GATA-adapted APE transforms the additional input and domain information intoCNF constraints so that APE can use this information. I use the structure of the expressions to define constraints to use in APE. I use the clear order within a GATA expression to define sets of constraints.This approach can only approximate the correct workflows. To ensure only correct workflows appearin the result, a post-process step is added. It reconstructs the GATA expression of the workflow andcompares it to the workflow specification. If the specification expression cannot be recreated from aworkflow, the workflow is discarded. Only the correct workflows remain.I tested GATA-adapted APE on four ArcGIS examples and compared it to the APE approach. Thenew approach resulted in a smaller amount but more domain-specific workflows. This makes it easier fora user to find a workflow if they provide a GATA expression. However, the time needed to generate theconstraints and to post-process the results increases drastically for more extended workflows.