Scripts and preprocessed dataset for An identification method for two types of particular behaviors in stream temperature time series. Application to a national dataset in mainland France

The measurement of the stream water temperature signal is subject to various issues and environmental phenomena. Accurate interpretations of the data composing water temperature time series (WTS) often requires a high-level human expertise during data preprocessing steps to sort out meaningful temperature signals. This study proposes a method to highlight two main types of particular behaviors encountered in WTS, apart from outliers: intensified data and buffered data. The method uses a metric based on the WTS itself to identify periods with particular data. It enables the identification and the visualization of regular and irregular particular behaviors in a given WTS. The method was applied to a large national dataset collected in mainland France. The dataset contains 993 WTS with a wide range of data quality and environmental measurement conditions. Data identified as particular behavior accounts for up to 7% of the dataset. Depending on the measurement conditions, up to 25% of a given WTS data can be considered as ”occasional particular behavior” and potentially not exploitable. Buffered data mostly occur during winter months with no apparent spatial pattern. Intensified data occur mainly in summer months and a spatial pattern shows WTS containing the highest percentage of intensified data in the south-east part of the study area. The identification method was also applied to several known situations where a high-level human expertise was available. It provided robust identification performances at regional scale confronted to human expertise as well as at national scale, on a large dataset. Such methods can facilitate the selection of exploitable data in large datasets which are more widely available today. Potentially problematic data becomes straightforward and subsequent data qualification or correction is facilitated.