Bridging the gap between modelling advice and irrigator solutions through empirical reasoning techniques

Abstract

Science has led to increased on-farm irrigation productivity in the last decade in Australia through the widespread adoption of better crop varieties, management practices and mechanical tools. However, Decision Support Systems (DSS) for farmers, despite proved productivity benefits Inman-Bamber et al. (2005), have seen poor adoption. Inman-Bamber and Attard (2005) noted the vast majority of purpose-built irrigation DSS saw poor or no commercial adoption after development. Australian authors, particularly McCown (2001, 2002b,a), have noted a science-industry 'gap' between science researchers and DSS designers and the agriculture industry resulting in systems not entirely fit for practical purpose. Current feeling in agricultural DSS development internationally is summed up by Matthews who states "there is a need to think beyond technocentric solutions" Matthews et al. (2008). They argue that DSS designers must engage with target users through collaborative approaches during development to prevent this 'gap' forming. The issue central to the gap problem is DSS not catering for the myriad of hard to model, yet influential, factors that affect real-world decisions. This paper details an approach to decision support, implemented using 'technocentric' techniques that may do so. It allows open-ended problem parameterisation and user choice in determining advice relevant to industry that is derived from peers' empirical data. The approach is implemented using a heavily modified Case-Based Reasoning (CBR) Artificial Intelligence (AI) technique with a novel interface. Traditional CBR systems look at the decision scenario at hand - the current case - and attempt to draw parallels to previous scenarios - stored cases - based on similarity measures. They then presents likely solutions to the current case from relevant stored cases. Our CBR variant presents an enhanced version of emulated non-DSS farmer decision making. Added to human memory and reasoning is the power of DSS data storage and computer information manipulation. It allows users to select case parameters of interest to them, via an interface, to be used in case comparison in place of the standard, automated and thus predetermined, similarity measures used by most CBR systems. This approach does not restrict decision input parameters or pathways to a solution, unlike current model-based DSS. Advice delivered is still useful to users when the stored cases contain decisions which have been made using knowledge they do not possess. Analysis of irrigation decision factors is given in Section 1 and the new theoretical approach is given in Section 2. Section 3 details CBR practices that relate to the theory, Section 4 describes a test implementation domain within Australian irrigation, Section 5 details a DSS prototype construction and Section 6 gives initial test results conducted with real irrigation decision data. The aim of his paper is not to detail a complete, new, DSS using yet another AI technique. It is to highlight that at least one technical approach to DSS design exists which may sidestep current ongoing DSS problems. Ever more sophisticated, purely biophysical, model-based, DSS may never do this but Australian agricultural DSS designers need not give up all technical DSS design. Many more as-yet undiscovered approaches to decision support, including hybrid rule-based and CBR systems that require technical implementation may exist and may prove useful to farmers.