Uncertainty is more than a number or colour : Involving experts in uncertainty assessments of yield gaps

CONTEXT: Yield gap analysis plays an important role in determining potential food availability. The Global Yield Gap Atlas maps yield gaps of crops from point to regional scale across the globe. The calculated yield gaps are based on comparisons between modelled potential yields with actual farmers' yields derived from statistical sources. The calculations are subject to uncertainty due to various sources, including measurement errors, modelling limitations, and scaling issues. OBJECTIVES: An important goal of the Atlas is to convey an uncertainty evaluation of the yield gap analysis. The aim of this paper is to provide a practical methodology that can make the assessment of the uncertainty by experts explicit and accessible for users of the Atlas. METHODS: We developed an uncertainty protocol and guidelines listing several sources of uncertainty to be considered by country agronomists who were involved in the calculation of the yield gaps. These experts are asked to score the level of uncertainty of each source, as well as the relative impact of each source. Both scores are combined into uncertainty scores for each source. Aggregated uncertainty scores for yield gaps, potential and actual yields are mapped as colours in the Atlas to indicate ranking. Moreover, experts are encouraged to provide a justification for their scores, which are also made available to users of the Atlas. RESULTS AND CONCLUSIONS: The uncertainty protocol was applied to 189 country-crop combinations by fourteen experts. They ranked lack of data for model calibration, model sensitivity to specific conditions, weather data, and the data quality on cropping system as the most important uncertainty sources for potential yields. The quality of yield data was ranked as the highest source of uncertainty for actual yields. The justifications provided by experts suggest which uncertainty sources may be reducible with relatively little effort, while other uncertainty sources may be more difficult or im