The use of machine learning considerably increased in recent years, mainly due to advances in computing power and an easier access to it. At the same time, open data policies at European level enabled an easier access to many data sources (Southworth 2021). Notably, thanks to the Copernicus satellite imagery programme, a wide range of land use mapping applications have been developed. In this context, access to quality data is essential to properly train and evaluate machine learning algorithms (Gounari et al., 2022). This allows robust and generalizable predictors to be applied over large areas. At the European level, the combination of spatial imagery data (Sentinel) with data from the Land Parcel Identification System (LPIS; within the framework of the Common Agricultural Policy, CAP) is increasingly being used to map agricultural land use (Dobrinić et al., 2021; Orynbaikizy et al., 2020; Blickensdorfer et al., 2022). The use of these data in the training and evaluation of algorithms is decisive for the efficiency and generalization of the predictors obtained (Blickensdorfer et al., 2022). The research unit INRAE-SADAPT (French National Institute for Agriculture, Food and Environment) provides annual data on crop filiation and sequence at islet and parcel level for 2007-2014 and 2015-2020, respectively. These data are constructed from the information in the LPIS within the framework of the CAP using the RPG Explorer software (Levavasseur et al., 2016). They provide access to rich information, both temporally and spatially, on the whole of France's agricultural rotation. Our work presents the datasets of crop sequences produced from 2007 to 2020 at an islet/parcel level, and aims to validate or identify new uses of these data with respect to digital soil mapping, and make a possible generalization at the European level.