A framework to use crop models for multi-objective constrained optimization of irrigation strategies

• Published in: Environmental modelling & software : with environment data news Vol. 86; pp. 145 - 157
• Main Authors: Cheviron, Bruno, Vervoort, R. Willem, Albasha, Rami, Dairon, Romain, Le Priol, Camille, Mailhol, Jean-Claude
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.12.2016; Elsevier Science Ltd; Elsevier
• Subjects: Agricultural economics; Agricultural management; Case studies; Constraints; Crop model; Crop yield; Crops; Dispersions; Environmental Sciences; Framework; Irrigation; Learning; Learning algorithms; Machine learning; Mathematical models; Modelling; Multiple objective analysis; Optimisation; Optimization; Parameter sensitivity; Parameter uncertainty; Scenarios; Sensitivity analysis; Studies; Uncertainty; Uncertainty analysis; Scenarios; Modelling; Crop model; Optimisation; Uncertainty analysis; Framework
• ISSN: 1364-8152; 1873-6726
• DOI: 10.1016/j.envsoft.2016.09.001

This paper discusses an innovative framework to use crop models which combines sensitivity analysis, uncertainty analysis and constrained optimisation runs for irrigation optimisation purposes, facing competing constraints on several agricultural variables (e.g. crop yield, total irrigation amount, financial expectations). For simplicity, this ex-post optimisation relies on direct calculations only, exploiting the dispersions on the target variables. The screening of the parameter space for sensitivity analysis yields a reference dispersion which is expectedly reduced by reducing the uncertainties in the sensitive parameters and/or climatic forcings. Additional dispersions are calculated to evaluate if the management controls on irrigation strategies (amounts, triggers, periods) are more influential on model predictions than the remaining uncertainties on the soil, plant, irrigation and climatic inputs, eventually allowing optimisation. As a case study, the Optirrig model is used. A discussion proposes future ways to convert diagnostics into real-time near-optimal decision rules, for example through learning algorithms. •This ex-post optimisation of irrigation strategies involves direct calculations only.•The Optirrig model (Irstea Montpellier, France) is chosen for application.•Both the feasibility and the achievement of irrigation optimisation are handled.•Dispersion is reduced through sensitivity, uncertainty then optimisation runs.•Perspectives are discussed to convert diagnostics into real-time decision rules.