Global changes have led to a renewed interest in time series of environmental monitoring. In France, for example, the French Research Institute for the Exploitation of the Sea (Ifremer) has been managing for 40 years several networks with hundreds of active sites, with annual to fortnightly sampling frequencies, measuring dozens of variables. These long-term datasets are difficult to analyse due to their characteristics (e.g. missing data, outliers, changes in sampling frequency, shifts).For this large number of time series, this paper proposes a semi-automatic procedure based on Dynamic Linear Models, detailed from data pre-processing (e.g. time unit definition, aggregations, transformations), through model specification, automatic and manual intervention, outlier and shift handling, to model hypothesis testing.When applied to three time series combining the above features, the results showed that missing data and changes in sampling frequency were adequately handled. Outliers and structural breaks were identified automatically, but also added manually. Highlighted shifts were identified as artefactual (e.g. probe drift), anthropogenic (e.g. ministerial decree) and ecological changes (e.g. storm impact).Finally, the presented treatment has been successfully applied routinely to more than 19,000 time series with a common and simple model structure. The broad theoretical framework offered by dynamic linear models opens up fruitful perspectives for improving and extending the results presented here, in particular for dealing with measurement quantification limits and time-varying observation variances.