Metamodeling methods that incorporate qualitative variables for improved design of vegetative filter strips

• Published in: Reliability engineering & system safety 2020-12, Vol.204, p.107083, Article 107083
• Main Authors: Lauvernet, Claire, Helbert, Céline
• Format: Article
• Language: English
• Subjects: Agrochemicals; Buffer zone; Categorical variables; Climate models; Covariance; Design; Design improvements; Environmental risk; Gaussian processes; Kriging; Mathematical models; Mathematics; Metamodeling; Metamodels; Mitigation; Model reduction; Numerical models; Overland flow; Pesticide; Pesticides; Pollutants; Pollution; Pollution control; Qualitative inputs; Reliability engineering; Risk analysis; Risk management; Runoff; Soil conditions; Surface runoff; Surface water; Surrogate; Toolkits; Vegetative filter strip; VFSMOD; Water pollution; Water purification
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2020.107083

•Metamodelling the vegetative filter strip toolkit BUVARD, including VFSMOD.•A kriging method mixing qualitative and quantitative variables is implemented.•Many metamodels are evaluated (linear, additive, and kriging per modality).•Kriging with mixed variables is more stable and efficient than the other methods.•The metamodel is a relevant tool for testing global rules but also local scenarios. Significant amounts of pollutant are measured in surface water, their presence due in part to the use of pesticides in agriculture. One solution to limit pesticide transfer by surface runoff is to implement vegetative filter strips. The sizing of VFSs is a major issue, with influencing factors that include local conditions (climate, soil, vegetation). The BUVARD modeling toolkit was developed to design VFSs throughout France according to these properties. This toolkit includes the numerical model VFSMOD, which quantifies dynamic effects of VFS on site-specific pesticide mitigation efficiency. In this paper, a metamodeling, or model dimension reduction, approach is proposed to ease the use of BUVARD and to help users design VFSs that are adapted to specific contexts. Three different reduced models, or surrogates, are compared: a linear model, GAM, and kriging. It is shown that kriging, implemented with a covariance kernel for a mixture of qualitative and quantitative inputs, outperforms the other metamodels. The metamodel is a way of providing a relevant first approximation to help design the pollution reduction device. In addition, it is a relevant tool to visualize the impact that lack of knowledge of some field parameters can have when performing pollution risk analysis and management.