Verification, Validation and Vacancy-Profiling (VVV-P) as an assessment of quality and network growth potential of gridded climate data sets

Context. Over the past decades, we have seen important developments in climatological observational gridded data sets. One development is the increasing number of stations included in the gridding process. For any gridded data set, one might ask oneself: (a) How sensitive is the gridded data set to the number of stations? (b) How accurate is the gridded data set? (c) What is the potential improvement of the gridded data set if we add more stations, and does this potential improvement depend on the locally existing station density, the local terrain, etc.?

Approach. We can learn from similar questions in engineering, specifically in the discipline of computational fluid dynamics (CFD). In this field of research, questions (a) and (b) are addressed in a standardized process of ‘verification’ (a) and ‘validation’ (b). The results of a verification and validation study are usually reported in terms of a grid convergence index (GCI) and (cross-)validation results. This approach was standardized by Roache in his work Verification and Validation in Computational Science and Engineering (Hermosa Publishers, 1998). We aim to apply the same procedure to gridded data sets. In this way, our approach is a way of interdisciplinary exchange of methodology.

In addition, we analyze the effects of adding stations by tracking their type of location, terrain, etc. (part of which might also be used as covariates during gridding) when we quantify their effect on the gridded data set. In this way, we can train a simple machine learning model of how sensitive the gridded data set is to inclusion of stations with specific characteristics. The aim is to identify which type of stations, based on their characteristics, would be the most valuable addition to the data set. The last step (c), we name ‘vacancy-profiling’.

Application. As a first study, we apply this approach to the E-OBS gridded data set for daily mean wind speed. This is an interesting data set, since the network density has increased significantly over the years, because the gridding process includes the use of covariate information which gives more details in the verification and validation processes.