Quality data are essential for the reliability of the pre-assessment. Therefore, specific attention was given to select state-of-the-art data originating from trusted sources, built following a well-documented procedure and having undergone a thorough quality check.
Strategy for identifying suitable data
The constraint that the tool should be able to carry out an assessment for any location in the world implied that, either the one dataset on a global scale should be used, or that regional data should be used and combined so that no region in the world would be free of data.
For the sake of consistency, global products were preferred to self-built blends of regional datasets. In most cases, the global data still originate from various sources but they have been blent into one product using well-documented objective rules and have undergone a comprehensive quality check. This procedure also ensures that the reliability of the pre-assessment is the same for any location.
In addition, a compromise had to be made between the level of detail (i.e. spatial en temporal resolutions) and the overall reliability of the data. As such, it was requested that the data would be sufficiently representative of reality, would be made available by a trusted source, and would undergo a thorough quality check before release. Priority was therefore given to measurements or to numerical model output including a quality check and validation against measurements, published by established scientific and/or (supra-)governmental entities.
Within the various types of numerical models and simulations that are available, so-called reanalyses were preferred over so-called analyses. In a reanalysis, a numerical model is used to model a period in the past (for instance 1993-2010) and that includes data assimilation of historical observations carried out during the modelled period. In a reanalysis, the model is constrained to compute the best solution that satisfies the underlying equations and still approaches the observed state as best as possible, with the a-priori knowledge of all available measurements. In an analysis, the model is used to compute a short-term forecast (typically a few hours up to a couple of days) from the time of observations, whereby the model has only knowledge of observations that have been carried in the past (relative to the moment the model is currently computing). A reanalysis should therefore lead to a more consistent (in space and time) representation of the observed system an analysis does.